# Disability World — Full corpus
> All published articles in reverse-chronological order. Each entry includes a header with the canonical URL, author, and publication date.
---
title: Assistive technology for blind people: the three-year leap that put sight on demand
url: https://www.disabilityworld.org/articles/assistive-tech-for-blind-people-2026/
description: Assistive technology for blind and low-vision people changed more between 2023 and 2026 than in the prior decade. This primer maps the real innovations — Be My AI, Ray-Ban Meta, smart canes, the Monarch, and AI screen readers — with what each delivers and where it still fails.
author: Disability World
pubDate: 2026-05-30
tags: assistive-technology, blind, low-vision, ai, smart-glasses, screen-readers, braille, tech-news
---
# Assistive technology for blind people: the three-year leap that put sight on demand
Engineering primer · Assistive tech for blind users
Sight on demand
the three years that rewired blind and low-vision life
Between 2023 and 2026, the tools blind and low-vision people use every day stopped being a slow trickle of single-purpose gadgets and became a wave of general-purpose AI. A phone can now read a room, a pair of ordinary-looking sunglasses can call a volunteer, and a braille display can finally show a graph. This primer maps what genuinely shipped, who makes it, and — just as important — where each one still fails.
Mar 2023
GPT-4 vision shipped with Be My Eyes as a launch partner
Nov 2024
Ray-Ban Meta glasses gained a blind-user mode
10 lines
first mainstream multiline braille and tactile-graphics display
By The Disability World engineering desk
13 min read
Updated May 2026
Foundation
1. What actually changed
For most of the smartphone era, the assistive technology a blind person relied on came in two flavours. There were narrow, expensive, single-purpose devices — a text-reading camera, a colour identifier, a GPS unit with a clumsy voice — and there were apps that connected you to a human, because no machine could reliably describe the messy visual world. The first flavour was costly and brittle. The second worked, but it meant asking another person every time you wanted to know whether the milk had expired.
The pivot came in March 2023, when OpenAI announced GPT-4 and used the blindness app Be My Eyes as a flagship demonstration of what a vision-capable model could do. For the first time a general-purpose model, not a hand-built classifier, could look at an arbitrary photo and answer questions about it in fluent language. That single capability — describe anything, then answer follow-ups — turned out to be exactly the thing the field had been missing. Within eighteen months it had been wired into phones, sunglasses, screen readers, and canes.
This primer surveys that wave across six fronts: the visual-assistance apps, the wearables, the navigation aids, the operating-system screen readers, the braille and tactile breakthroughs, and the web layer underneath all of it. Throughout, the question is the same one we ask of any new tool: not "is it impressive in a demo?" but "does a blind person get a correct, useful answer when they need one?" The honest answer, in 2026, is "far more often than in 2022 — and still not often enough to trust blindly." We keep both halves of that sentence in view.
i
What "delivers" means here
We treat a tool as delivering when it returns an answer a blind user can act on without a sighted person re-checking it. The same yardstick we apply to AI image descriptions in our companion primer on where AI alt text actually delivers in 2026 applies here: a confident sentence that is wrong is worse than no sentence at all.
Landscape
2. Sight on demand: the apps and services
The most consequential change is also the least visible: it lives in apps people already had. The category split into two layers that now work together — instant AI description for the routine question, and a human on the line for the moment that matters. The strongest workflows let a user start with the model and escalate to a person in one tap.
The cards below capture the practical behaviour of the five services that dominate everyday use, not the marketing claims. "The catch" is the column to read first.
Be My AI
Be My Eyes · GPT-4 vision
Free; the default first stop for millions of users
What's newAI describes any photo, then answers follow-ups in conversation
EscalationOne tap to a sighted volunteer when AI is not enough
The catchConfident hallucinations; not for medication or safety calls
Seeing AI
Microsoft · free
Came to Android in late 2023 after years iOS-only
What's newGenerative "rich" scene descriptions and document Q&A on top of its classic channels
StrengthFast, offline-capable short text and currency reading
The catchRich descriptions inherit the same fabrication risk as any model
Aira
Aira · paid / sponsored access
Trained professional agents, not volunteers
What's newFree access sponsored at airports, campuses, and workplaces expanded through 2024-2025
StrengthAccountable, consistent help for high-stakes tasks
The catchMinutes cost money outside sponsored locations
Lookout
Google · free (Android)
Built around the phone camera and Gemini
What's new"Ask about an image" lets users pose questions about a photo and get generative answers
StrengthTight integration with Android and TalkBack
The catchAndroid-only; quality varies with lighting and clutter
Envision (Ally)
Envision · app + glasses
App is free; the glasses are a separate purchase
What's new"Ally", a conversational LLM assistant launched in 2024, can be asked open-ended questions
StrengthStrong document reading; same brain on phone and glasses
The catchThe premium experience is gated behind hardware
"The strongest workflows let a user start with the model and escalate to a human in one tap — the machine for speed, the person for the moment that matters."
— this article, section 2
Hardware
3. The camera moved to your face
Holding a phone up to point its camera is workable, but it occupies a hand and announces to everyone nearby exactly what you are doing. The most important hardware shift of the period was moving the camera onto the head, where it points wherever the user looks and frees both hands. Two things made this real at once: cheap, decent wearable cameras, and a model good enough to make sense of what they see.
The landmark was November 2024, when Meta added a blind-user mode to its mainstream Ray-Ban Meta glasses through a Be My Eyes integration — a "Call a Volunteer" feature that streams the wearer's first-person view to a sighted helper, alongside Meta's own AI that can describe what is in front of you on request. For the first time the assistive device was a pair of sunglasses people already wanted to wear, not a conspicuous medical appliance.
Ray-Ban Meta
Meta · mainstream consumer glasses
The first "normal-looking" glasses with a blind mode
What's newBe My Eyes "Call a Volunteer" + on-request AI scene descriptions, hands-free
StrengthSocially invisible; low cost relative to dedicated devices
The catchNot built for blind users first; no obstacle sensing
Envision Glasses
Envision · Google Glass Enterprise base
Purpose-built for blind and low-vision wearers
What's newThe Ally assistant on-glasses; instant text, scene, and face recognition
StrengthBest-in-class reading of printed and handwritten text
The catchCosts far more than consumer glasses; ageing hardware base
OrCam MyEye
OrCam · clip-on camera
A fingertip-sized camera that clips to any frame
What's newOn-device reading and recognition with voice-command "smart reading"
StrengthWorks offline; instant, private, no phone required
The catchPremium price; narrower than an open-ended AI assistant
biped NOA
biped.ai · wearable vest
Self-driving-car sensing adapted for pedestrians
What's newPredicts collisions and warns through 3D spatial sound; "Live AI" describes surroundings as you move
StrengthContinuous obstacle awareness, not just on-demand description
The catchA complement to the cane and dog, never a replacement
!
Description is not navigation
Glasses that describe a scene are excellent at "what is this?" and useless at "is there a step in front of me?" Scene description and obstacle avoidance are different jobs requiring different sensors. Every credible maker in this category says the same thing: the device sits alongside the white cane or guide dog, not in place of it.
Mobility
4. Knowing where you are
Navigation is the hardest problem in the field, because the cost of a wrong answer is a curb, a stairwell, or a road. The period produced real progress on two distinct sub-problems: sensing what is immediately around you, and orienting yourself in a building where GPS dies.
1
WeWALK Smart Cane 2
A 2024 refresh of the smart cane that bolts a sensing handle onto an ordinary white cane. It detects chest- and head-height obstacles that a cane sweep misses — overhanging branches, open cupboard doors, truck mirrors — and warns through vibration. The second generation widened the detection angle, added a built-in AI voice assistant (running on GPT-4) and tighter navigation and public-transit integration, and collected an Edison Award and a King's Award for Enterprise Innovation. Crucially, it keeps the cane: the proven tool stays, the sensing is additive.
2
Glidance Glide
The most genuinely new form factor of the period. Glide is a small two-wheeled device from a company founded by former Microsoft accessibility technologist Amos Miller. You nudge it forward and it rolls ahead of you, physically guiding you — steering around obstacles and communicating through the telescoping handle, somewhere between a white cane and a guide dog. Its first pre-order batch opened in mid-2024 and sold out by year's end; the device carries a monthly subscription of about 30 USD, with shipping to the earliest backers beginning in 2026. It is early, and it is the device most worth watching.
3
GoodMaps indoor navigation
Outdoor turn-by-turn has worked for years; indoors, where GPS fails, has not. GoodMaps uses camera-based positioning to place a user inside a mapped building — an airport, a transit hub, a campus — and give step-by-step guidance without the beacons earlier systems required. Coverage is the limit: it only works where a venue has paid to be mapped.
4
Apple Door Detection and Magnifier
The navigation aid most people already own. The Magnifier app's Detection Mode finds doors, reads the signage on them, and reports whether they are open and how to open them, using the LiDAR scanner on Pro iPhones and iPads. People Detection measures distance to others nearby, and VoiceOver Recognition describes objects and scenes on-device. None of it needs a subscription or extra hardware — it ships in the box.
"The cost of a wrong navigation answer is not an awkward sentence — it is a curb, a stairwell, or a road. That is why every serious maker keeps the cane in the loop."
— this article, section 4
Platform
5. The OS caught up
The quietest revolution happened inside the screen reader. For years, the gap a blind user hit most often was the undescribed image — a photo, a chart, a meme with no alt text. Between 2024 and 2026 every major platform shipped a built-in answer: point the screen reader at an image and an on-board model describes it, then takes follow-up questions. What used to require a third-party app is now a keystroke.
The matrix below compares where each platform landed. The pattern is consistent — AI image description everywhere, live camera understanding strongest on mobile, braille support newly deepened on Apple — but the details decide which tool fits a given user. For testing methodology and tooling, our screen-reader testing tools guide goes deeper, and the underlying standard is WCAG 2.2.
Screen reader
AI image description
Live camera scene
New in 2025
Cost
VoiceOver + Magnifier (Apple)
VoiceOver Recognition (on-device)
Door & People Detection
Braille Access, Accessibility Reader, Magnifier for Mac
Built in
TalkBack + Gemini (Android)
Gemini describes & answers questions
via Lookout
Deeper Gemini Q&A on images and full screen
Built in
JAWS (Windows)
Picture Smart AI (ChatGPT, Claude)
N/A (desktop)
Faster Picture Smart, follow-up Q&A
Paid licence
NVDA (Windows)
Community add-ons (GPT-4 vision)
N/A (desktop)
Maturing add-on ecosystem
Free + add-on
Apple's May 2025 wave deserves its own note, because it widened the definition of accessibility. Braille Access turns an iPhone, iPad, Mac, or Vision Pro into a full braille note-taker that talks to a refreshable display natively. Accessibility Reader is a system-wide reading mode for low-vision and dyslexic users. Accessibility Nutrition Labels put the accessibility features of an app right on its App Store page, so a blind user can tell before downloading whether an app will work — a structural nudge that pressures every developer to do better.
One earlier feature deserves a mention here too: Personal Voice, which lets someone record and synthesise a model of their own voice. It was built with people losing speech in mind, but it points at a broader future where the synthetic voice in a blind user's ear can be one they actually chose.
Touch
6. Reading by touch finally got a graph
Amid all the AI, the most overdue breakthrough was mechanical. Refreshable braille displays had shown a single line of text for decades — fine for prose, hopeless for a maths textbook, a map, or a chart. The dream of a full page of dynamic braille and tactile graphics had a name in the field, "Holy Braille", and for years it stayed a dream.
In 2024 it shipped. The Monarch, a partnership between the American Printing House for the Blind and HumanWare, is the first mainstream device to show ten lines of braille and tactile graphics on the same refreshable surface — so a student can feel a bar chart, a geometry diagram, or a map and read its braille labels at once. It is Android-based, imports tactile-graphic files, and supports the emerging multiline eBraille format. The price is steep, around five figures, which is why it largely reaches students through institutional funding rather than individuals. Korea's Dot Pad, a pin-array tactile display that Apple supports natively, attacks the same problem from the consumer side. For the wider market, see our refreshable braille displays buyer's guide.
i
Why a tactile graph matters
A blind student can listen to a description of a parabola, but they cannot explore it the way a sighted student traces a curve with their eyes. Multiline tactile graphics restore that exploration. The educational consequence — particularly for STEM, where the field has lost generations of talent to inaccessible diagrams — is larger than the device count suggests.
Diagnostics
7. The catch: what is still broken
Every section above carried a "the catch" line for a reason. The progress is real, but a primer that only sold the upside would be doing its readers a disservice. Four limitations cut across the whole landscape, and any honest buyer should weigh them before the marketing.
1
Confident hallucination
Every AI-description tool here will, sometimes, describe something that is not there — a price that is wrong, a label it could not read but guessed, an expiry date it invented. It does so in the same fluent, certain tone it uses when it is right. For routine questions that is tolerable; for medication, allergens, financial documents, or anything safety-critical, the only safe rule is to verify with a human or a trusted non-AI channel. The model drafts; it does not get the final word.
2
The price of the good stuff
The free tier is genuinely transformative — Be My AI, Seeing AI, Lookout, and the built-in screen-reader features cost nothing. But the dedicated hardware that does more, or works hands-free, or reads by touch, runs from hundreds to many thousands. A Monarch is a five-figure device. The result is a widening gap between what is theoretically possible and what an individual without institutional funding can actually afford.
3
The camera always sees
A device that streams your first-person view to a cloud model or a volunteer also streams everything else in frame — the people around you, the documents on your desk, the inside of your home. The privacy trade-off is real and largely unregulated, and it lands hardest on the users with the least choice about whether to accept it. Good design minimises what leaves the device; not all design is good.
4
Tools are not training
No app replaces orientation-and-mobility instruction, and no sensor replaces the white cane or guide dog for detecting the ground. The danger of a very good assistant is the false confidence it can create. The devices that succeed are the ones built as additions to proven skills, not substitutes for them — which is why the cane keeps reappearing in this article.
!
The web is still the weak link
All of this assistive intelligence runs on top of a web that is mostly still inaccessible. An AI screen reader can describe an image, but it cannot fix a button with no label, a form that traps focus, or a checkout that breaks under a screen reader. The tools improved faster than the websites did. Before trusting that your own site keeps up, run it through a free accessibility scan — and treat AI overlays that promise instant compliance with deep suspicion.
Conclusion: the ceiling rose, the floor held
Written honestly, the story of 2023 to 2026 is that the ceiling rose dramatically and the floor barely moved. A blind person in 2026 can do things that were science fiction in 2022 — ask a pair of sunglasses what is on a menu, feel a graph refresh under their fingers, get any photo described in a keystroke. That is a genuine expansion of independence, and it arrived faster than anyone in the field predicted.
But the floor — the things that have to be right every single time — held firm. A model still hallucinates. A camera still sees too much. A great app still cannot fix a broken website or replace a mobility instructor. The maturity of this moment is not in the demos; it is in knowing exactly which tool to trust for which job, and which to double-check. The best practitioners and users already think this way: machine for speed, human for the moment that matters, and the cane in your hand the whole time.
The next three years will be judged on the floor, not the ceiling. If hallucination rates fall, if the good hardware gets cheaper, and if the web underneath finally catches up to the assistive technology sitting on top of it, the gap between what is possible and what is reliable will close. Until then, the rule that runs through every section of this primer holds: the tools are a remarkable draft of sight on demand — and the user, not the model, still gets the final say.
"The ceiling rose dramatically and the floor barely moved. Maturity is knowing which tool to trust for which job — and which to double-check."
— this article, conclusion
---
title: How much does ADA website compliance cost in 2026 — and do you actually need it?
url: https://www.disabilityworld.org/articles/ada-website-compliance-cost/
description: What website accessibility actually costs in 2026 — DIY, scanners, manual audits, remediation, and ongoing monitoring — plus whether you legally need it, what happens if you skip it, and when to hire.
author: Disability World
pubDate: 2026-05-29
tags: ada, website-accessibility, accessibility-compliance, cost, buyers-guide, explainer
---
# How much does ADA website compliance cost in 2026 — and do you actually need it?
Image description: A small-business owner's desk with a laptop showing a website, a calculator, and a printed invoice, photographed top-down on warm wood, signalling the budget-and-decision framing of an accessibility cost guide.
Reading Time: 12 minutes
If you have landed here, you are probably a business owner, a marketing lead, or an office manager who got a forwarded email — maybe a demand letter, maybe a nervous note from legal — and you are trying to answer four questions at once. Do I really have to do this? What happens if I do not? What will it cost? And can I just fix it myself? These are the most-asked questions about website accessibility on the open web, and they are usually answered with either a scare tactic or a sales pitch. This guide answers all four plainly, with 2026 numbers, and routes you to the next concrete step.
The short version: yes, the law almost certainly applies to you; the most likely consequence of ignoring it is a demand letter that costs more than getting compliant would have; the cost ranges from free to five figures depending entirely on your site's size and how it was built; and you can do a meaningful chunk yourself but not all of it. Now the detail.
Do you actually need an accessible website?
For almost every organisation that serves the public online, the answer is yes — and not as a nice-to-have. A disability-rights law already covers your website in every major market:
United States — the Americans with Disabilities Act (ADA). Courts and the Department of Justice treat a public-facing website as a "place of public accommodation" under Title III, and there is no small-business exemption. Our ADA Title III web accessibility guide walks through exactly who is covered.
European Union — the European Accessibility Act (EAA) has applied to most consumer-facing digital products and services since June 2025, and it reaches non-EU companies that sell into the EU.
United Kingdom — the Equality Act 2010, plus the Public Sector Bodies regulations for government services.
Canada, Australia, and 50-plus other jurisdictions — each with its own statute and a working enforcement record.
The benchmark every one of these regimes points to is the same technical standard: WCAG 2.2 Level AA. So the real question is not "does this apply to me" — it almost certainly does — but "how exposed am I right now, and what does closing that gap cost?" For the full picture of what "compliance" even means across these layers, see our accessibility compliance explainer.
What happens if your website isn't accessible?
The risk is not theoretical, and it usually arrives by email before it arrives by lawsuit.
Demand letters. The most common first contact in the US is a letter from a plaintiff's firm alleging that a disabled user could not use your site, and offering to settle. These settle quietly, typically for $5,000 to $25,000, before any court filing. Many businesses receive several over a few years.
Lawsuits. If you ignore the letter, a Title III suit can follow. Roughly 4,000 federal web-accessibility cases are filed in the US each year, concentrated among a small number of firms — our firm-by-firm field guide tracks who. Litigated cases cost far more than demand-letter settlements once you add legal fees; the largest resolved matters reach seven figures, as our survey of the 25 largest settlements documents.
Regulatory action abroad. In the EU, market-surveillance authorities under the EAA can issue corrective orders and administrative fines that vary by member state. There is no demand-letter cottage industry there — the enforcement comes from the state.
Lost customers. The quietest cost is the largest. Around one in six people lives with a disability. An inaccessible checkout, an unlabelled form, or a video with no captions turns those visitors away silently — no letter, no lawsuit, just a lower conversion rate you never attribute to the real cause.
How much does website accessibility cost?
There is no single number, because cost scales with two things: how big and complex your site is, and how it was built. A five-page brochure site on a mainstream template is a different problem from a custom single-page application with a checkout, a dashboard, and a media library. Here is the realistic 2026 range, broken into the layers most organisations actually buy.
What you're buying
Typical 2026 cost
What it gets you
Automated scan (DIY)
Free
A baseline. Free tools — axe DevTools, Lighthouse, WAVE, and our own free WCAG 2.2 scanner — catch roughly 30–40% of issues: contrast, alt text, form labels, structure.
A professional review, ideally including testers who use assistive technology, that finds what scanners cannot — broken screen-reader flows, keyboard traps, mislabelled custom widgets.
Developer time to fix what the audit found. Usually the largest line item, and the one that varies most with how the site was built.
Ongoing monitoring
$1,000–$30,000+ / year
A platform that re-scans on a schedule and flags regressions on every deploy, so a fixed site stays fixed.
Overlay widget
$490–$3,500 / year
Avoid. Does not make you compliant and does not reduce lawsuit risk — see below.
A useful way to read the table: a small business can reach a defensible posture for a low four-figure sum — a one-time audit, a modest remediation sprint, and an entry-level monitoring subscription. A large organisation with a custom application should budget for an annual programme in the tens of thousands. Either way, the all-in cost of getting compliant is, for the overwhelming majority of businesses, less than the cost of a single litigated lawsuit.
Can you fix it yourself?
Partly — and the DIY part is genuinely worth doing first, because it is free and it removes the most common, most-cited failures.
A technically confident team can run a free scanner, work through the findings, and close a real share of the gap without spending a dollar: add alternative text to images, fix colour-contrast failures, label every form field, restore a logical heading order, and make sure everything works with the keyboard alone. Our step-by-step WCAG 2.2 guide is built for exactly this. That work handles the 30–40% of issues automation can see.
What you cannot reliably do yourself is the other 60%. No scanner — and no non-disabled developer guessing — can tell you whether a screen-reader user can actually complete your checkout, whether your custom date-picker announces itself correctly to NVDA, or whether your modal traps focus the way it should. That judgement comes from manual audits by people who use assistive technology daily. DIY gets you a strong start and a smaller invoice; it does not get you all the way.
One thing not to buy as a shortcut: the accessibility overlay widget. These are third-party scripts that drop an accessibility toolbar onto your site for a few hundred to a few thousand dollars a year, marketed as instant compliance. They are not. Sites running overlays have been sued at least as often as sites without them, and plaintiffs increasingly name the overlay in the complaint. We tracked the vendors' retreat in our report on the end of the overlay era. Spend the money on an audit and real fixes instead.
When to hire a professional
Hire help when any of these are true: you run a checkout, a login, or any transaction that a customer must complete; your site is a custom application rather than a templated marketing site; you have already received a demand letter; or you sell into the public sector or the EU, where the conformance bar is contractual and explicit. In those situations the cost of getting it wrong dwarfs the cost of an audit.
For the ongoing layer, do not rely on a one-time audit alone — a site that passes today breaks on the next deploy. A continuous monitoring platform re-scans on a schedule and flags regressions before they reach production. Compare two or three vendors on what they actually test and how they surface regressions — axe Monitor, Siteimprove, Level Access, and Qualibooth are the usual shortlist — and weigh whether the platform pairs automated monitoring with periodic manual auditing, since neither alone is sufficient.
Your next step, by situation
"I just want to know where I stand." Run a baseline today. Our free WCAG 2.2 scanner gives you a first read on any public URL in minutes — no cost, no commitment.
"I got a demand letter." Do not ignore it, and do not buy an overlay in a panic. Commission a manual audit so you can show a good-faith remediation plan, and read the ADA Title III guide for what the obligation actually is.
"I want to fix this properly, long-term." Budget for the three layers — audit, remediation, monitoring — and stand up continuous monitoring so the fix holds. That is what a durable compliance posture looks like.
Frequently asked questions
Do I really need to make my website accessible?
In almost every case, yes. If you sell to or serve the public in the US, the EU, the UK, Canada, or Australia, a disability-rights law already applies to your website — the ADA in the US, the European Accessibility Act in the EU, the Equality Act in the UK. The obligation does not depend on your company's size for most regimes, and there is no exemption for small businesses under ADA Title III. The practical question is not whether the rule applies but how much risk you are carrying by ignoring it.
What happens if my website isn't ADA compliant?
The most common consequence in the US is a demand letter from a plaintiff's firm, followed by a Title III lawsuit if you do not respond. Roughly 4,000 federal web-accessibility cases are filed each year, plus a much larger volume of pre-suit demand letters that settle quietly for $5,000 to $25,000 each. In the EU, market-surveillance authorities under the European Accessibility Act can issue corrective orders and fines. Beyond the legal exposure, an inaccessible site simply loses the roughly one-in-six customers who have a disability.
How much does ADA website compliance cost?
There is no single figure because cost scales with site size and complexity. A baseline automated scan is free. A one-time professional manual audit runs roughly $1,500 to $5,000 for a small marketing site, $5,000 to $15,000 for a mid-size site, and $15,000 to $50,000 or more for a large or complex web application. Remediation — the developer time to fix what the audit finds — is usually the largest line item and varies with how the site was built. Ongoing monitoring runs roughly $1,000 to $30,000 a year. Most organisations should budget for an audit plus remediation plus continuous monitoring, not a one-off fix.
Can I fix my website's ADA compliance myself?
Partly. A technically confident team can run free scanners, fix the obvious failures — missing alt text, low colour contrast, unlabelled form fields, broken heading structure — and remove a meaningful share of risk at near-zero cash cost. What you cannot do yourself reliably is judge whether a screen-reader user can actually complete a checkout, or whether a custom widget exposes the right roles and states. Those require manual testing by people who use assistive technology daily. DIY handles the first 30 to 40 percent; the rest needs expertise.
Are accessibility overlay widgets a cheap way to become compliant?
No. Overlay widgets — the third-party scripts that add an accessibility toolbar for roughly $490 to $3,500 a year — do not make a site compliant, and sites running them have been sued just as often as sites without them. Plaintiffs increasingly name the overlay itself in the complaint. Treat an overlay as marketing, not remediation; the money is better spent on an audit and real code fixes.
How much does ongoing accessibility monitoring cost?
Continuous monitoring platforms that re-scan your site on a schedule and flag regressions typically run from roughly $1,000 a year for a small site to $30,000 or more a year for a large estate, depending on page volume and features. Compare two or three vendors — axe Monitor, Siteimprove, Level Access, and Qualibooth — on what they actually scan, how they surface regressions, and whether they pair with manual auditing, rather than on the dashboard alone.
Accessibility is rarely as expensive as the fear around it suggests, and almost always cheaper than the alternative. Start with a free baseline, fix what you can, and buy expertise where it counts.
---
title: The accessibility of public transport: laws, standards, and what still fails
url: https://www.disabilityworld.org/articles/accessibility-of-public-transport/
description: How public-transport accessibility is regulated — ADA paratransit and bus-lift rules in the US, the EU's PRM TSI and passenger-rights regulations, and the UK PSVAR — plus the gaps that persist from the platform to the app.
author: Disability World
pubDate: 2026-05-25
tags: public-transport, accessibility, ada-paratransit, prm-tsi, passenger-rights, mobility
---
# The accessibility of public transport: laws, standards, and what still fails
Image description: A low-floor city bus kneeling at a kerb with its ramp deployed, a wheelchair user about to board, photographed in soft early-morning light.
By Disability World editorial deskUpdated May 2026Reading time 12 minutes
Reading Time: 12 minutes
Public transport is where the social model of disability is most concrete. A bus with a step instead of a ramp, a station with no lift, a train whose doors give no audible warning — each is a barrier that turns an impairment into an exclusion, cutting a disabled person off from work, healthcare, education, and social life. Transport access was one of the disability movement's earliest battlegrounds, and it remains one of its most uneven victories. This explainer maps how public-transport accessibility is regulated in the major frameworks, what the technical standards actually require, and where the system still fails — from the platform gap to the ride-hail app.
Transport accessibility is unusually layered. It spans the vehicle (ramps, low floors, securement spaces), the infrastructure (stations, stops, tactile paving, lifts), the information (audio-visual announcements, accessible journey planners and apps), and the service (staff assistance, paratransit, booking). A network can comply on one layer and fail catastrophically on another, which is why "is the transport accessible?" never has a single yes-or-no answer.
The United States: the ADA and paratransit
US public-transport accessibility is governed by the Americans with Disabilities Act — Title II for public entities and Title III for private operators — implemented through detailed Department of Transportation regulations at 49 CFR Parts 37 and 38. The rules require that newly purchased buses and rail vehicles be accessible, mandating lifts or ramps, securement systems for wheelchairs, priority seating, and stop announcements.
The ADA's distinctive contribution is complementary paratransit: where a public transit agency runs a fixed-route service, it must also provide a comparable origin-to-destination service for people who cannot use the fixed route because of disability. Paratransit is a genuine civil-rights guarantee, but it is also the source of persistent complaints — advance-booking requirements, long pickup windows, capacity limits, and "no-shows" that strand riders. The movement's roots here run deep: the direct-action group ADAPT spent the 1980s physically blocking inaccessible buses under the slogan "We Will Ride," a campaign we trace in our history of disability rights activism.
Rail tells a harder story. Many older US rapid-transit systems were built long before the ADA, and the statute requires accessibility at "key stations" plus accessibility in all new construction — but retrofitting lifts into century-old subway stations is slow and expensive, and litigation over inaccessible stations has been a recurring feature of the past decade.
The European Union: the PRM TSI and passenger rights
The EU regulates transport accessibility through a combination of technical standards and passenger-rights regulations. The central technical instrument is Regulation (EU) No 1300/2014, the "PRM TSI" — the technical specification for interoperability relating to accessibility of the Union's rail system for persons with disabilities and persons with reduced mobility. It sets harmonised requirements for stations, platforms, and rolling stock: step-free access routes, boarding aids, accessible toilets, visual and audible information, and platform-to-train gaps.
Layered on top are the mode-specific passenger-rights regulations, which give disabled travellers enforceable rights to assistance:
Rail: Regulation (EU) 2021/782 guarantees assistance and non-discrimination for rail passengers with disabilities.
Bus and coach: Regulation (EU) No 181/2011 sets passenger rights including assistance at designated terminals.
Air: Regulation (EC) No 1107/2006 prohibits air carriers from refusing booking or boarding on grounds of disability and guarantees assistance at airports.
Waterborne: Regulation (EU) No 1177/2010 covers passengers travelling by sea and inland waterway.
Since 2025, the European Accessibility Act has added a further layer for the digital elements of transport — ticketing machines, journey-planner websites and apps, and electronic ticketing — bringing them within a common EU accessibility regime built on the procurement standard EN 301 549 and, for the web, the WCAG criteria. The full comparative picture for individual member states sits in our national regulations index.
"Reduced mobility" is broader than you think
EU transport law deliberately uses the phrase "persons with disabilities and persons with reduced mobility" (PRM). It covers not only wheelchair users and blind or Deaf travellers but also older people, people travelling with small children, and anyone whose mobility is temporarily reduced — a pregnant traveller, someone with a leg in a cast. Designing transport for this wider group is the clearest everyday illustration of the "curb-cut effect": access features built for disabled people benefit a far larger population.
The United Kingdom: PSVAR and rail vehicle rules
The United Kingdom regulates vehicle accessibility through the Public Service Vehicle Accessibility Regulations 2000 (PSVAR), which set out requirements for buses and coaches — wheelchair spaces, ramps or lifts, priority seating, colour-contrasted handrails, and the audible and visible "next stop" information that the Bus Services Act 2017 later pushed toward universal rollout. Rail is governed by the Rail Vehicle Accessibility Regulations and the overarching duties of the Equality Act 2010. As with the US, the binding constraint is rarely new vehicles — which are built accessible — but the legacy fleet and the inaccessible Victorian-era station.
The technical standards that matter
Behind the statutes sit the engineering details that determine whether a journey actually works. A handful recur across every framework:
Feature
What it does
Who it serves
Low-floor / kneeling buses + ramps
Step-free boarding
Wheelchair and mobility-aid users, buggies
Platform-train gap mitigation
Bridges or reduces the horizontal/vertical gap
Wheelchair users, the visually impaired
Tactile paving
Underfoot warning at platform edges and crossings
Blind and partially sighted travellers
Audio-visual stop announcements
Next-stop information in two modalities
Blind, Deaf, and cognitively disabled riders
Wheelchair securement / priority space
Safe travel position on board
Wheelchair users
Accessible ticketing and journey apps
Buying and planning without a barrier
Screen-reader and keyboard users
Note how many of these are information features rather than physical ones. A perfectly ramped bus is useless to a blind traveller who cannot tell which bus has arrived or where to alight, and to a Deaf traveller who misses an audio-only disruption announcement. Accessibility on transport is a chain; it breaks at its weakest link.
The app layer: where access is being won and lost now
The newest battleground is digital. Journey planners, contactless ticketing, real-time arrival displays, and especially ride-hailing apps have become the front door to mobility — and they reproduce, in software, the same access failures the movement spent decades fixing in hardware. An inaccessible booking screen locks a disabled traveller out just as surely as a missing ramp. Ride-hail in particular has been the subject of sustained complaint and litigation over the scarcity of wheelchair-accessible vehicles (WAVs) and the accessibility of the rider app itself — a subject we examine in our audit-style piece on mobility apps and disabled riders.
Because these are web and mobile interfaces, the governing standard is the same one that governs any digital service: the WCAG 2.2 success criteria, enforced in the EU through the European Accessibility Act and in the US through ADA case law. Transport operators and tech vendors can find the obvious failures in a ticketing site or journey-planner with a free accessibility scan before they ever reach a complaint.
What still fails
Despite four decades of law-making, the lived experience of disabled travellers remains patchy. The recurring failure points are consistent across countries: legacy rail stations without step-free access; paratransit and assistance services that are unreliable, over-subscribed, or require impractical advance booking; broken lifts and ramps that are not repaired promptly, with no real-time information telling a disabled traveller the lift is out before they arrive; staff training gaps, where the equipment exists but no one is willing or able to deploy it; and inaccessible digital front-ends that block the journey before it begins.
The pattern echoes every other accessibility domain: the law sets the floor, new build complies, and the gap between paper rights and a reliable journey is filled — or not — by maintenance budgets, staff culture, and enforcement. The standards are mature and largely agreed. The unfinished work is operational. For the wider context, read our history of the movement, the national regulations index, and the full 2026 reporting record.
---
title: Disability in the workplace: what the 2026 data actually shows
url: https://www.disabilityworld.org/articles/disability-in-the-workplace-2026-data/
description: A data dossier on the disability employment gap, the pay gap, and the quota systems built to close them — assembled from ILO, WHO, Eurostat, US BLS, OECD and JAN figures, with every number sourced and the soft spots flagged.
author: Disability World
pubDate: 2026-05-25
tags: employment, workplace, disability-employment-gap, ada, reasonable-accommodation, data
---
# Disability in the workplace: what the 2026 data actually shows
Editorial · Disability employment data
Disability in the workplace — what the 2026 data actually shows, and what it can't
Almost everywhere it is measured, the finding is the same: people with disabilities are employed at far lower rates than people without them, paid less when they do work, and the gap has barely moved in a generation. This dossier assembles the headline numbers from the bodies that actually compile them — the WHO, the ILO, Eurostat, the US Bureau of Labor Statistics, the OECD, and the Job Accommodation Network. Around 1.3 billion people — about 16% of the world — experience significant disability. In the United States the employment-population ratio for disabled people sits in the low-to-mid 20% range against roughly 65% for everyone else; in the EU the disability employment gap runs at about 24 percentage points. And yet roughly half of all workplace accommodations cost nothing at all. No source publishes a single tidy "2026 global disability employment rate" — and any that claims to should be treated with suspicion.
Findings · Evidence review07 findings · synthesised from published ILO, WHO, Eurostat, US BLS, OECD and JAN data, 2022–2024 reference years
What the published data shows
011.3B
The population the labour market is failing
The WHO's 2022 Global Report on Health Equity for Persons with Disabilities estimates that about 1.3 billion people — roughly 16% of the world's population — experience significant disability. That is the denominator against which every employment figure below should be read.
02approx. 40 pp
The US employment gap is about forty percentage points
US BLS Current Population Survey data puts the employment-population ratio for people with a disability in the low-to-mid 20% range, against roughly 65% for people without a disability — a gap of around 40 percentage points that has been remarkably stable over time.
03approx. 24 pp
The EU disability employment gap runs around twenty-four points
Eurostat measures the gap directly: roughly 51% of disabled people aged 20–64 are in work against about 75% of non-disabled people, a gap of approximately 24 percentage points, with very wide variation between member states. Closing it is an explicit target of the EU Disability Rights Strategy 2021–2030.
04approx. 50%
Around half of accommodations cost the employer nothing
The US Job Accommodation Network's long-running employer survey consistently finds that roughly half of workplace accommodations carry no cost at all, and that most of the rest are one-time expenses of a few hundred dollars. The barrier to hiring disabled workers is rarely money — it is awareness, attitude, and inaccessible recruitment systems.
055% / 6%
Europe leans on statutory quotas; the common-law world on rights
Germany obliges employers with 20-plus staff to fill a 5% quota or pay a levy; France sets 6%. The US, UK and others instead use anti-discrimination law backed by a duty of reasonable accommodation. The evidence on which approach works better is genuinely mixed.
06record high
Remote work produced the one genuine bright spot
Inside an otherwise bleak picture, the US disability employment ratio rose to record highs in the post-pandemic years — a shift most analysts attribute to the normalisation of remote and hybrid work, which removed the commute and the inaccessible office from the equation for many disabled workers.
07widest
The gap is widest for women with disabilities
The ILO reports that the employment gap is consistently largest for women with disabilities, who face the compounding disadvantage of gender and disability together — an intersectional pattern that single-axis statistics tend to obscure.
SourceWHO Global Report on Health Equity for Persons with Disabilities (2022); ILO disability-and-work statistics; Eurostat disability-employment-gap series; US Bureau of Labor Statistics Persons with a Disability: Labor Force Characteristics; OECD Sickness, Disability and Work; US Job Accommodation Network employer-cost surveys. Reference years 2022–2024; figures are the most recent published as this dossier went to press.
This is a synthesis of published official statistics, not an original survey — and that distinction matters. Disability statistics are unusually fragile: definitions vary between surveys, self-identification is sensitive and tends to undercount, and many low- and middle-income countries lack regular labour-force data disaggregated by disability at all. We have drawn every figure here from the authoritative compiler of that figure, and where a number is an estimate or a range, we say so. We have not manufactured fresher or more precise numbers to fill gaps in the record.
Four bodies do most of the reliable counting. The WHO supplies the global prevalence baseline. The US BLS runs a monthly disability question in its Current Population Survey, giving the cleanest single-country time series in the world. Eurostat publishes the EU disability employment gap from its EU-SILC and labour-force instruments. The OECD contributes the comparative work on benefit design. The ILO frames the global and gender picture, and the Job Accommodation Network (JAN) supplies the employer-cost evidence.
01Identify authoritative compilersWHO · ILO · Eurostat · US BLS · OECD · JAN — no secondary aggregators
02Take latest published reference yearMostly 2022–2024; disability-disaggregated data lags collection by a year or more
03Preserve uncertaintyReport ranges and "approximately" where the source itself is imprecise
04Flag the missingWhere no reliable count exists — most of the global south — say so rather than estimate
02 · The size of the gap
The cleanest national picture comes from the United States. In recent years the BLS employment-population ratio for people with a disability has sat in the low-to-mid 20% range, against roughly 65% for people without a disability — and disabled people who are in the labour force also face an unemployment rate consistently about twice that of non-disabled workers. The European picture, measured differently, tells the same story at a higher level: roughly half of disabled Europeans of working age are employed, against three-quarters of everyone else.
16%
Of the world experiences significant disability (WHO, 2022) — approx. 1.3 billion people
approx. 40 pp
US disability employment gap (BLS) — roughly 23% vs 65%
approx. 24 pp
EU disability employment gap (Eurostat) — roughly 51% vs 75%
EMPLOYMENT RATE — WITH VS WITHOUT DISABILITY
US — without disability
approx. 65%
US — with disability
approx. 23%
EU — without disability
approx. 75%
EU — with disability
approx. 51%
Measure
With disability
Without disability
Source
US employment-population ratio
approx. 22–25%
approx. 65%
US BLS, recent years
EU employment rate (age 20–64)
approx. 51%
approx. 75%
Eurostat, 2022
Global population with significant disability
approx. 1.3 billion (16%)
WHO, 2022
Why "2026 data" is mostly 2022–2024 data
Labour-force surveys are large, expensive instruments, and the disability-disaggregated results are typically published a year or more after collection. The most current authoritative figures available in 2026 are therefore generally from 2022–2024 reference years. We have not invented fresher numbers to fill the gap. On this topic, false precision is its own form of misinformation.
The European data also exposes a measurement subtlety worth stating plainly: because the disabled population skews older than the general working-age population, and because the survey question relies on self-reported limitation, cross-country comparisons partly reflect who answers "yes" rather than real differences in labour-market exclusion. Eurostat is candid about this, which is why the gap is best read within a country over time rather than as a league table.
03 · The pay gap and the benefits trap
Lower employment is only half the story. Where disabled people are employed, they tend to earn less — the disability pay gap — driven by concentration in lower-paid occupations, higher rates of part-time and insecure work, and outright discrimination. National statistics offices in the UK and elsewhere have measured median pay gaps of the order of 12% to 17% in recent years, though the figure is sensitive to how disability and working hours are defined.
Compounding both gaps is the benefits trap the OECD has documented across its member states: disability and out-of-work benefit systems that withdraw support sharply as earnings rise, so that taking a job — especially a part-time or uncertain one — can leave a disabled person barely better off, or worse off once the costs of working are counted. The OECD's long-running work argues that the most effective systems pair anti-discrimination law with active labour-market support and benefit designs that taper rather than cliff-edge.
The accommodation-cost myth
The single most persistent employer misconception is that disabled workers are expensive to accommodate. The Job Accommodation Network's employer survey has found, year after year, that roughly half of accommodations cost nothing, and that most of the remainder are one-time expenses of a few hundred dollars — a screen reader, flexible hours, a modified desk, captioning. The constraint is almost never the budget.
04 · Quotas versus the rights approach
Two broad regulatory traditions try to close the gap. Continental Europe and parts of Asia use statutory employment quotas: a legal obligation on employers above a certain size to ensure a minimum share of their workforce is disabled, usually backed by a levy paid by those who fall short. The common-law world instead relies on anti-discrimination law and a duty of reasonable accommodation.
Jurisdiction
Mechanism
Threshold / rate
Germany
Quota + levy (Ausgleichsabgabe)
5% for employers with 20+ staff
France
Quota + AGEFIPH fund
6% for employers with 20+ staff
Italy
Graduated quota (Law 68/1999)
Rises with headcount; "targeted employment"
Japan
Statutory employment rate + levy
Raised in steps in recent years
United States
Anti-discrimination + accommodation (ADA Title I)
No quota; enforced by the EEOC
United Kingdom
Anti-discrimination + reasonable adjustments
No quota; Equality Act 2010
Quotas are widely credited with putting a floor under disability employment, but the evidence on their effectiveness is mixed. Many employers treat the levy as a cost of doing business rather than a prompt to hire, and quotas can channel disabled workers into sheltered or segregated settings that Article 27 of the UN CRPD (work and employment) discourages in favour of the open labour market. The rights approach, meanwhile, depends entirely on enforcement: under the Americans with Disabilities Act Title I, a US employer must provide adjustments that let a qualified disabled person do the job unless that would impose undue hardship — a duty mirrored by the UK Equality Act and the EU Employment Equality Directive. The comparative detail by country sits in our national regulations index.
05 · The digital gateway — hiring tech as a barrier
Increasingly, the first barrier a disabled candidate meets is not the building but the application form. Online job portals, applicant tracking systems, and AI-assisted screening tools have become the gatekeepers to the labour market — and many are not operable by keyboard-only or screen-reader users, or use video-interview and game-based assessments that disadvantage disabled applicants. The US EEOC has issued guidance warning that AI hiring tools can violate the ADA, and the EU AI Act classifies employment-related AI as high-risk. An inaccessible apply-flow is a discrimination problem before a candidate is ever judged on merit.
The fix is the same standard as everything else
Accessible recruitment software is governed by the same WCAG 2.2 criteria as any other web service, and public-sector HR procurement across the EU is bound by EN 301 549 and, since 2025, the European Accessibility Act. Employers serious about disability inclusion can start by checking their own careers site and application flow with a free accessibility scan.
06 · What the data does and does not show
Three conclusions hold across every credible dataset. First, the disability employment gap is large, durable, and global — measured in tens of percentage points wherever it is measured at all. Second, the gap is not explained by disabled people's lower desire or capacity to work; survey after survey finds disabled people who want jobs and cannot get them, blocked by removable barriers. Third, the policy tools that work are the unglamorous ones used in combination: enforceable anti-discrimination law, accessible recruitment, well-designed benefit tapers, and a culture of accommodation — not any single silver bullet.
What the data does not yet tell us well is the granular, current, country-by-country picture for most of the world. The richest figures come from a handful of wealthy economies with strong statistical agencies; for much of the global south, the honest answer to "what is the disability employment rate?" is that nobody is reliably counting. Improving that data baseline is itself a disability-rights priority — because, as with every other indicator, what is not counted is not budgeted for. For the wider context, see our reporting on the history of disability rights activism and the full 2026 record.
---
title: A history of disability rights activism: from the back wards to 'nothing about us without us'
url: https://www.disabilityworld.org/articles/disability-rights-history-of-activism/
description: How the disability rights movement moved from charity and institutions to a civil-rights framework — the 504 sit-in, the social model, deinstitutionalisation, the Capitol Crawl, and the road to the UN CRPD.
author: Disability World
pubDate: 2026-05-25
tags: disability-rights, history, activism, independent-living, social-model, crpd
---
# A history of disability rights activism: from the back wards to 'nothing about us without us'
Image description: Disabled activists with placards and wheelchairs gathered on the steps of a government building, photographed from a low angle against an overcast sky.
By Disability World editorial deskUpdated May 2026Reading time 13 minutes
Reading Time: 13 minutes
For most of the twentieth century, disability was treated as a medical misfortune to be managed by professionals and, very often, hidden away. Disabled people were patients, beneficiaries, and inmates — rarely citizens with rights. The modern disability rights movement is the story of how that frame was overturned: how a scattered set of self-help groups, parents' associations, psychiatric survivors, and disabled veterans came to insist that the problem was not their bodies but a world built to exclude them. This is a primer on that history — the ideas, the protests, and the laws — from the institutions of the 1950s to the UN Convention on the Rights of Persons with Disabilities and the enforcement landscape mapped across our national regulations index.
The through-line is a single conceptual shift, repeated in different national keys: from charity to rights, from the medical model to the social model, from "care" delivered to passive recipients to self-determination claimed by an organised political community. The slogan that came to encapsulate it — "nothing about us without us" — was not a piece of branding. It was a demand that disabled people be the authors of the decisions, laws, and services that govern their lives.
The institutional baseline
To understand the movement, you have to understand what it pushed against. In the mid-twentieth century, the dominant response to significant disability in the industrialised world was segregation: residential institutions, "colonies," asylums, and special schools that removed disabled people from ordinary community life. Conditions in many were grim, and the institutions were rarely accountable to the people inside them.
The darkest expression of the medical model was the Nazi "Aktion T4" programme, under which the German state systematically murdered disabled children and adults between 1939 and 1945 — people judged "lives unworthy of life." The history is not a digression. It is the extreme endpoint of a logic that treated disabled existence as a problem to be solved by experts, and it shaped the post-war human-rights consciousness that the disability movement would later draw on. Disability arts and remembrance work — documentary film, memoir, the recovery of individual stories from the camps and the wards — has been central to keeping that history visible rather than buried.
By the 1960s, exposés of institutional conditions in the United States, the United Kingdom, and Scandinavia were prompting a reckoning. The principle of "normalisation", articulated first in Scandinavian social policy, argued that people with intellectual disabilities were entitled to patterns of everyday life as close as possible to the mainstream. It became one of the intellectual seeds of deinstitutionalisation.
The independent living movement
The movement's organisational breakthrough came from disabled people organising themselves. In the United States, the pivotal figure was Ed Roberts, who in 1962 fought his way into the University of California, Berkeley, despite an administration that doubted a man who used a wheelchair and an iron lung could study there. Roberts and a group of fellow disabled students — the "Rolling Quads" — built a campus support network, and in 1972 founded the Center for Independent Living (CIL) in Berkeley. The CIL model was radical in a simple way: services for disabled people should be designed and run by disabled people themselves, oriented not around cure or care but around the supports — personal assistance, accessible housing, transport, peer counselling — that make self-directed community life possible.
The independent living philosophy spread quickly to a network of centres across the United States and then internationally, and it reframed the entire policy conversation. The question was no longer "how do we care for disabled people?" but "what barriers prevent disabled people from living the lives they choose, and how do we remove them?" That reframing — barriers, not bodies — is the practical heart of the social model.
The social model of disability
If the independent living movement supplied the organising energy, British activists supplied the theory. In 1976 the Union of the Physically Impaired Against Segregation (UPIAS) published its Fundamental Principles of Disability, drawing a now-famous distinction between impairment (a feature of the body or mind) and disability (the disadvantage imposed by a society organised around non-disabled norms). The disabled academic Mike Oliver later named and systematised this as the social model of disability.
The social model did not deny impairment or pain. Its claim was about causation and responsibility: a wheelchair user is not disabled by their legs but by the staircase; a Deaf person is not disabled by deafness but by a hearing society's refusal to provide sign language. That shift moved the obligation to act from the individual onto society — onto architects, employers, broadcasters, and lawmakers. It is the conceptual engine behind every accessibility statute that followed, and it underlies the modern compliance framework explained in our guide to how compliance, conformance and accessibility differ.
Three models, briefly
The medical model locates the problem in the disabled individual and seeks to cure, fix, or care for them. The social model locates the problem in disabling barriers — physical, attitudinal, and institutional — that society can choose to remove. The human-rights model, which the UN CRPD adopts, builds on the social model but adds that accessibility and inclusion are matters of enforceable right, not goodwill — and that disabled people are rights-holders entitled to remedies when those rights are breached.
The 504 sit-in and the American statutes
The United States produced the movement's most iconic direct action. Section 504 of the Rehabilitation Act 1973 contained a single sentence prohibiting discrimination on the basis of disability by any programme receiving federal funds — the first US civil-rights protection for disabled people. But for four years successive administrations refused to sign the regulations that would give it force. In April 1977, disabled activists occupied the San Francisco federal building of the Department of Health, Education, and Welfare for 26 days — the longest occupation of a federal building in US history. Led by figures including Judy Heumann and supported by the Black Panther Party, who delivered hot meals to the protesters, the 504 sit-in forced the regulations to be signed unchanged.
The campaign culminated in the Americans with Disabilities Act (ADA) of 1990, the most comprehensive disability civil-rights statute of its era, banning discrimination in employment, public services, and places of public accommodation. The image that defined its passage was the "Capitol Crawl" of March 1990, when disabled activists abandoned their wheelchairs and mobility aids to haul themselves up the steps of the US Capitol, dramatising exactly the kind of barrier the bill was meant to abolish. The direct-action group ADAPT, which had spent the 1980s blocking inaccessible buses under the banner "We Will Ride," was central to that pressure — a campaign we trace forward in our explainer on the accessibility of public transport. The ADA's promise of access to public accommodations is also the legal ancestor of today's web-accessibility litigation, covered in our ADA Title III guide.
Deinstitutionalisation and the survivor movement
In parallel, a movement of psychiatric survivors and people with intellectual disabilities challenged the institutions directly. The push for deinstitutionalisation — closing the large residential asylums and moving support into the community — was driven both by the normalisation principle and by litigation exposing abuse and neglect. The self-advocacy movement, organised internationally under the banner People First from the 1970s, asserted that people with learning disabilities could speak for themselves and make decisions about their own lives, against a system built on the assumption that they could not.
The legacy is double-edged and still contested. Deinstitutionalisation freed hundreds of thousands of people from custodial institutions, but where community supports were under-funded, some were left without the assistance they needed. The unfinished business — adequate community support, an end to coercive treatment, the right to legal capacity — runs directly into Article 12 (legal capacity) and Article 19 (living independently and being included in the community) of the modern CRPD.
The international turn: from declarations to a treaty
The United Nations moved in stages from soft aspiration to binding law. The General Assembly proclaimed 1981 the International Year of Disabled Persons, followed by the World Programme of Action concerning Disabled Persons (1982) and the Decade of Disabled Persons (1983–1992). In 1993 the General Assembly adopted the Standard Rules on the Equalization of Opportunities for Persons with Disabilities — influential, but not legally binding and without an enforcement body.
The decisive shift came in the 2000s. Disabled persons' organisations, coordinated through the International Disability Alliance, pressed for a binding treaty, and the negotiations themselves became a demonstration of the movement's central principle: disabled people and their organisations participated directly in drafting, to a degree unprecedented in UN treaty-making. The Convention on the Rights of Persons with Disabilities (CRPD) was adopted on 13 December 2006 and entered into force on 3 May 2008 — the fastest-negotiated human-rights treaty in UN history and now among the most widely ratified. Its enforcement architecture, two decades on, is the subject of our dedicated analysis of twenty years of the CRPD.
Year
Milestone
Significance
1972
Berkeley Center for Independent Living founded
Services by and for disabled people
1973
US Rehabilitation Act, Section 504
First US disability anti-discrimination provision
1976
UPIAS Fundamental Principles
Impairment/disability distinction
1977
504 sit-in, San Francisco
Regulations signed under pressure
1981
UN International Year of Disabled Persons
Global visibility
1990
Americans with Disabilities Act
Comprehensive US civil-rights statute
1995
UK Disability Discrimination Act
First UK anti-discrimination law
2006
UN CRPD adopted
Binding human-rights treaty
Beyond the Anglosphere
The narrative is often told as a US and UK story, but the movement was always plural. The United Kingdom's Disability Discrimination Act 1995 followed years of direct action by the Disabled People's Direct Action Network (DAN), including activists chaining themselves to inaccessible buses. Across continental Europe, statutory employment-quota systems — in Germany, France, Italy, Austria, and Poland — reflected a different, more corporatist tradition of disability policy, one we examine in our analysis of disability in the workplace. Japan and the wider Asia-Pacific region developed their own frameworks, and the post-2008 wave of CRPD ratifications drove new national statutes across Africa, Latin America, and the Middle East.
What unites these strands is the move from a politics of charity to a politics of rights — and the insistence, everywhere, that disabled people themselves lead it. The phrase "nothing about us without us", popularised by the scholar-activist James Charlton in his 1998 book of that title, captured a principle that had been latent in the movement for decades and is now written into the CRPD's consultation duties.
From the streets to the standards
The movement's latest frontier is digital. The same logic that demanded curb cuts and ramps now demands accessible websites, apps, and public-sector digital services. The technical expression of the social model online is the Web Content Accessibility Guidelines (WCAG), and its legal expression includes the EU's European Accessibility Act and the procurement standard EN 301 549. The barriers have changed shape — a missing alt attribute instead of a missing ramp — but the claim is identical: an environment that excludes disabled people is a problem the builders of that environment are obliged to fix. Organisations that want to find those barriers can begin with a free accessibility scan.
The history is not finished. Institutionalisation persists, legal capacity is still denied to many, the disability employment gap remains stubbornly wide, and enforcement of even the strongest statutes is uneven. But the conceptual victory is real and largely won: disability is now understood, in law and increasingly in culture, as a matter of rights and barriers rather than charity and defect. That shift was not granted. It was organised, argued, and occupied into being by disabled people — which is, in the end, the whole point of "nothing about us without us." Continue with our wider 2026 reporting record and the national regulations index.
---
title: Accessibility compliance — what it actually means in 2026
url: https://www.disabilityworld.org/articles/accessibility-compliance-explainer/
description: Accessibility compliance, demystified — what it means under the ADA, EAA, WCAG 2.2 and EN 301 549, the country map, and the next steps to take regardless of where you operate.
author: Disability World
pubDate: 2026-05-23
tags: accessibility-compliance, ada, eaa, wcag, regulation-primer, explainer
---
# Accessibility compliance — what it actually means in 2026
Image description: Layered legislative documents and a magnifying glass on cream paper, photographed top-down, signalling the routing-to-specifics function of an umbrella regulation primer.
Reading Time: 11 minutes
"Accessibility compliance" is one of those phrases that everyone uses and almost no one defines the same way. A general counsel hears it as exposure under the Americans with Disabilities Act. A procurement officer in Berlin hears it as a clause referencing EN 301 549. An engineering lead hears it as a Lighthouse score. A product manager hears it as a vendor's monitoring dashboard. They are all, in a sense, correct — and none is talking about quite the same thing.
This article is the umbrella. It unpacks what the term means across three layers — technical, legal, and procedural — explains the most important distinction in the field (compliance is not the same thing as accessibility), and then routes you to the dossier that matches your jurisdiction. If you came in searching "accessibility compliance" because someone above you in the org chart asked whether the company is, you are in the right place. We will give you the map you need to answer that yourself, plus the link to the free WCAG 2.2 scanner for a baseline by the end of the afternoon.
Compliance is not the same thing as accessibility
The single most useful distinction in this field, and the one most often skipped over in vendor pitches, is the gap between compliance and accessibility.
Compliance is a legal and audit posture. It is the answer to the question "can we demonstrate, on paper, that we meet the standard the regulator or the contract has named?" It is satisfied by passing scans, published accessibility statements, conformance reports, and procurement attestations. Compliance is a state you can be in or out of at a moment in time, and it is the state lawyers and auditors care about.
Accessibility is the answer to a different question: "can a person with a disability actually use this product to do the thing it was built for?" It is satisfied by a screen-reader user completing the checkout, a keyboard-only user navigating the dashboard, a deaf user understanding the video. Accessibility is a property of the lived experience, not the audit file.
The two are correlated but not identical. A site can pass an automated WCAG 2.2 AA scan, publish a clean accessibility statement, ship a VPAT, and still be unusable with VoiceOver because every custom dropdown swallows focus. Another site can have a few hundred minor scan failures and still be usable end-to-end by every assistive-tech user in the room. Conformance is necessary but not sufficient. Treating the audit as a proxy for the experience is the most common — and most expensive — mistake in this space.
Compliance is not a one-time audit
The second crucial distinction. An audit that passes on Monday does not survive a deploy on Tuesday. Marketing sites push twice a week; product surfaces push twenty times a day. Every push is an opportunity to break a label, lose a focus ring, or ship a component that announces itself as a div to NVDA.
Compliance, properly understood, is a posture maintained through continuous monitoring and a culture of remediation — not a snapshot captured in a PDF dated nine months ago. The PDF is evidence. The monitoring is the thing being evidenced. The deepest pitfall in this space is treating compliance as a project rather than as an operating mode.
Every serious accessibility programme has the same shape regardless of jurisdiction: an automated monitoring layer that catches regressions on every deploy, a periodic manual audit layer that catches what automation misses, and a published statement that the regulator can read. Lose any of the three and the posture decays.
The three layers of accessibility compliance
It helps to think of accessibility compliance as three stacked layers. A given organisation has to satisfy each of them, and the requirements at each layer come from a different source.
Layer 1 — Technical standards. What your site, app, or product has to look like at the code level. This is the layer the W3C and the European standards bodies own. The dominant standard is WCAG, currently at version 2.2. Most regulators reference Level AA. Other technical standards layer on top: ARIA for rich UI, EN 301 549 for the European procurement and EAA context, Section 508 for US federal procurement, EPUB Accessibility 1.1 for ebooks, PDF/UA for documents. WCAG 2.2 AA is the centre of the spec map; the others extend or instantiate it.
Layer 2 — Regional regulation. What the law in your jurisdiction actually requires. The ADA, the European Accessibility Act, the UK Equality Act, the Accessible Canada Act, Italy's Stanca Act, Germany's BFSG. The technical standard from Layer 1 is almost always pulled in by reference — the ADA Title II rule names WCAG 2.1 AA; the EAA references the harmonised standard EN 301 549, which references WCAG 2.1 AA. The law is what you are accountable to; the standard is how the law decides whether you met the bar.
Layer 3 — Procurement and sector mandates. Where the contract pushes beyond the legal floor. EU public-sector procurement uses EN 301 549 as the floor for any IT purchase. US federal procurement uses Section 508, which now harmonises with WCAG 2.1 AA. Financial-sector and healthcare regulators sometimes add their own overlays. If you are selling into a public-sector buyer, the contract is usually a stricter ceiling than the law.
Most organisations have to satisfy all three layers simultaneously.
The country map — where you are determines what applies
This is the routing section. Find the row that matches the jurisdiction you operate in, then click through to the dossier.
United States
The Americans with Disabilities Act is the spine of US accessibility law. Title II governs state and local government and got its first explicit web rule in April 2024, naming WCAG 2.1 AA. Title III governs private places of public accommodation and is enforced primarily through private litigation — roughly 4,300 federal website cases in 2024. See the ADA Title III dossier and the US regulations hub for Section 508 and state statutes.
European Union
The European Accessibility Act (Directive 2019/882) became enforceable across the 27 member states on 28 June 2025, applying to a defined list of consumer products and services — ecommerce, banking, ebooks, ticketing, transport. The Web Accessibility Directive (2016/2102) covers public-sector bodies separately. Both reference the harmonised standard EN 301 549, which references WCAG 2.1 AA. See the European Accessibility Act guide.
Germany
The Barrierefreiheitsstärkungsgesetz (BFSG) is Germany's transposition of the EAA and has applied to in-scope private-sector services since 28 June 2025. BITV 2.0 governs the public sector. Enforcement runs through market-surveillance authorities at the Länder level. See the Germany dossier.
United Kingdom
The Equality Act 2010 imposes a general duty not to discriminate, interpreted by courts to extend to digital services. The Public Sector Bodies Accessibility Regulations 2018 layer a specific WCAG 2.1 AA requirement onto public bodies. The UK did not transpose the EAA; private-sector exposure runs through the Equality Act. See the United Kingdom dossier.
Italy
The Stanca Act (Law 4/2004) predates EAA harmonisation and applies to public bodies and a defined list of private-sector entities including financial services and large ecommerce. EAA transposition extends scope further. See the Italy dossier.
Canada
The Accessible Canada Act covers federally regulated entities including banks, telecoms, and federal departments; the AODA covers Ontario provincially, with full private-sector application from 2025. See the Canada dossier.
Australia
The Disability Discrimination Act 1992 is the statutory baseline, interpreted to cover digital services since the 2000 Maguire case. The Australian Human Rights Commission publishes WCAG-aligned guidance. See the Australia dossier.
Everywhere else
Disability World maintains country dossiers across 55 jurisdictions including Japan, India, Brazil, South Korea, Israel, South Africa, Mexico, the GCC states, and the rest of the EU and ASEAN.
What WCAG 2.2 actually requires
WCAG — the Web Content Accessibility Guidelines — is the technical reference almost every other instrument in this stack defers to. Version 2.2 was published by the W3C in October 2023 and adds nine new success criteria to the 2.1 baseline. Three conformance levels exist: Level A (the bare minimum), Level AA (the working benchmark referenced by every major regulator), and Level AAA (aspirational, not a regulatory floor anywhere).
WCAG is organised around four principles, the POUR acronym. Content must be Perceivable (available to senses the user has — text alternatives for images, captions for audio, sufficient colour contrast), Operable (the interface can be driven — keyboard-accessible, no seizure-inducing flashing, enough time to read), Understandable (the content and the interface are predictable — readable text, consistent navigation, error prevention on form input), and Robust (it works with the assistive technology the user actually has — parseable markup, name-role-value exposed on every interactive element).
The 2.2 specification has 86 numbered success criteria across the three levels. Level AA is the working target for compliance purposes; pulling Level A only is insufficient under every regulatory regime that names a level. The full reference, organised criterion by criterion with examples and common failure patterns, lives at the WCAG 2.2 success criteria page in our toolkit.
Next-step CTAs — three concrete actions
Compliance is a posture, not a project, but you have to start somewhere. The right next action depends on where you are in the journey.
"I have no idea where we stand" — start with a baseline
Run the free WCAG 2.2 scanner on your top ten pages — homepage, login, checkout, dashboard, two key conversion flows, the accessibility statement page itself if you have one. A scan will not tell you the full picture (see the section on compliance versus accessibility above), but it will tell you whether you have ten issues or ten thousand, which is the first thing you need to know. Then read the regulation dossier for your jurisdiction from the country map in the previous section.
"We have a scanner — what's next?" — commission a manual audit
Automated scans catch roughly a third of issues. The rest require manual review. Commission a manual audit by testers with disabilities — screen-reader users for the screen-reader paths, keyboard-only users for the keyboard paths, low-vision users for the visual paths. A first manual audit on a mid-sized application typically takes two to four weeks and surfaces the issue classes automation cannot see: ambiguous link text in context, focus order that is technically correct but cognitively wrong, custom widgets that work but feel wrong.
"We're operationalising compliance long-term" — choose a monitoring platform
This is the stage at which monitoring matters. Continuous monitoring catches regressions on every deploy, surfaces drift between manual audits, and produces the evidence trail that satisfies a regulator asking how you maintain the posture between formal reports. The full comparative review of the platforms in this category lives in the monitoring buyer's guide — including axe Monitor, Siteimprove, Level Access and Qualibooth, which are the four platforms most often shortlisted for the monitoring-plus-manual-audit-handoff combination. The buyer's guide compares them on coverage, false-positive rates, manual-audit workflow, integration surface, and price. Read that next; do not pick a platform from a single page.
Frequently asked questions
What does accessibility compliance mean?
Accessibility compliance is the legal and audit posture of meeting whichever standard a regulator, contract, or industry body has named as authoritative for your situation. In practice it usually means three things at once: meeting a technical standard such as WCAG 2.2 Level AA, satisfying a regional law such as the ADA or the European Accessibility Act, and producing process artefacts such as a published accessibility statement. None of these on their own equals accessibility for real users — they are the audit-side picture.
Is WCAG the same as the ADA?
No. WCAG is a technical standard published by the W3C; the ADA is a US civil rights law. The two are linked because US courts and the Department of Justice have treated WCAG 2.1 Level AA as the working benchmark for what counts as an accessible website under Title II and Title III, but the ADA itself never names WCAG in its statutory text. EN 301 549 plays the same bridging role in the EU.
Is my website legally required to be accessible?
Almost certainly, yes, if you operate in the US, the EU, the UK, Canada, Australia, or any of the 50-plus jurisdictions with a disability-rights statute and a working enforcement record. The exact obligation depends on whether you are a public-sector body, a commercial business of a certain size, or a vendor to government. The country map in this article points to the dossier for your jurisdiction.
What WCAG level do I need — A, AA, or AAA?
AA is the working answer for almost every regulator. The ADA Title II 2024 rule names WCAG 2.1 Level AA. The European Accessibility Act and EN 301 549 reference WCAG 2.1 AA. The UK Public Sector Bodies regulations reference 2.1 AA. AAA is aspirational and is not a regulatory floor anywhere. Level A alone is insufficient under every major regime.
How is accessibility compliance enforced in the US?
Through three channels. Department of Justice rulemaking and enforcement under Title II and Title III of the ADA. Private litigation under Title III, which produced roughly 4,300 federal website cases in 2024. And state-level demand letters under New York, California, and similar consumer-protection regimes. There is no single federal accessibility regulator equivalent to the EU's market-surveillance authorities.
Does the EAA apply to non-EU companies?
Yes, where they place products or services on the EU market. The European Accessibility Act applies to any economic operator selling covered products or services to consumers in any of the 27 member states, regardless of where the operator is established. A US ecommerce site shipping to Germany is in scope for the German transposition, the BFSG.
Is automated scanning enough to be compliant?
No. Automated scanners catch roughly 30 to 40 percent of WCAG issues under generous assumptions — colour contrast, missing alt text, missing form labels, document structure. They cannot judge whether alt text is accurate, whether a flow is usable by a screen-reader user, or whether a custom widget exposes the right roles and states. A defensible compliance posture combines automated monitoring with periodic manual audits by testers with disabilities.
In closing
Accessibility compliance is a posture, not a project. The audit on the wall is evidence of the posture; the monitoring, the manual audits, the remediation culture, and the published statement are the posture itself. Lose the posture and the audit goes stale within a quarter.
If you are starting from zero, run the free WCAG 2.2 scanner, read the regulation dossier for your jurisdiction from the country map above, and book a manual audit. If you are operationalising for the long term, read the monitoring buyer's guide. The umbrella is over. Pick the door.
---
title: Accessibility monitoring buyer's guide 2026 — platforms compared
url: https://www.disabilityworld.org/articles/accessibility-monitoring-buyers-guide-2026/
description: Real-time accessibility monitoring platforms compared — buying criteria, vendor table, and the trade-offs between automated scanning and manual audit handoff in 2026.
author: Disability World
pubDate: 2026-05-23
tags: accessibility-monitoring, buyers-guide, saas-comparison, qualibooth, axe-monitor, siteimprove
---
# Accessibility monitoring buyer's guide 2026 — platforms compared
Accessibility monitoring as a category has been reshaped in the last twenty-four months by three forces, and the buying decision in 2026 looks nothing like the buying decision in 2023. This guide is for the procurement officer, the engineering director, the chief compliance officer, and the accessibility lead who has been asked to shortlist platforms.
30–40%
Share of WCAG issues that automated scanning can detect on its own
6
Named platforms compared on eight buying criteria
$15k–$120k
Typical enterprise list-price range, USD per year
By disability-world editorial
18 min read
Updated May 2026
Framing
Why the buying decision changed
Accessibility monitoring as a category has been reshaped in the last twenty-four months by three forces, and the buying decision in 2026 looks nothing like the buying decision in 2023. First, the European Accessibility Act became enforceable in June 2025, and the procurement waves that have followed across all twenty-seven member states have lifted the EU portion of the monitoring market well past the US for the first time. Second, the DOJ Title II 2024 rule hardened the US public-sector requirement and triggered a procurement cycle across state and local government that is still working through. Third, the field has finally absorbed the uncomfortable empirical fact that automated scanning, on its own, catches only somewhere between thirty and forty percent of WCAG 2.2 success criteria — which means picking a platform is now much less about "scanner accuracy" and much more about workflow. The question is how scan output becomes triage becomes remediation becomes a verified, defensible, publishable accessibility statement.
This guide is for the procurement officer, the engineering director, the chief compliance officer, and the accessibility lead who has been asked to shortlist platforms. It compares six named vendors across the criteria that actually decide the contract. Before any of that, though, it is worth being precise about what "accessibility monitoring" is and is not — because the vendors blur the categories on purpose. If you want a baseline number for your own site before you read further, the free Disability World scanner will give you one in under a minute.
Definitions
1. What "accessibility monitoring" actually means
The category is younger than its vocabulary suggests, and four distinct products are routinely sold under overlapping names. Separating them is the first useful step in any buyer conversation.
A scanner is a one-off URL check. You paste in a single page, the tool fetches it, runs a ruleset — usually axe-core or a derivative — and prints a list of violations. Browser extensions like axe DevTools, Lighthouse's accessibility audit, the WAVE toolbar, and most free online scanners belong in this category. Scanners are commoditised. The underlying rulesets are mostly the same open-source engines under different skins, and the results across the major scanners on a given page rarely diverge by more than a few percent.
Monitoring is the continuous version. A monitoring platform crawls a site or an app on a schedule, builds a baseline, and then reports diffs as deploys land. Where a scanner answers "what's wrong with this page?", a monitoring platform answers "what changed since Tuesday?" — and that diff view is the thing the engineering organisation actually uses. Monitoring is what scales scanner output to a fleet of pages, a multi-property organisation, or a product surface that ships twenty times a day.
An audit is the manual review. A specialist — increasingly, a tester with a disability working with the screen reader they use every day — walks through the product end-to-end and reports the issues automation cannot detect. Keyboard traps, focus-order quality, screen-reader legibility, the actual meaningfulness of alt text, the behaviour of dynamic-content updates, the comprehensibility of error messages. Audit is the layer that catches the sixty-to-seventy percent of WCAG issues that scanners miss.
A statement or compliance dashboard is the published artefact and the workflow that produces it. Under the EAA, under the UK Public Sector Bodies regulations, under section 508 procurement, under the EN 301 549 framework, the buyer has to publish something — an accessibility statement that reads cleanly to a regulator, lists the conformance level, names the known issues, and dates the next review. A "compliance dashboard" is the internal-facing version that tracks the same posture for the executive team.
A monitoring platform — what this guide compares — is the product that stitches scanner output, continuous crawl, triage, optional manual audit, and statement generation into a single workflow. The scanner layer is commodity. The platform layer is where the differentiation lives, and the value of the contract.
Scanner
One-off URL check, axe-core or derivative ruleset
axe DevTools · Lighthouse · WAVE
Monitoring
Continuous crawl + regression diff against a baseline
What this guide compares
Audit
Point-in-time manual review by a specialist (often disabled tester)
Catches the 60–70% automation misses
Statement
Published accessibility statement + internal compliance dashboard
EAA Art. 13 · EN 301 549 · DOJ Title II
Criteria
2. Buying criteria — what actually matters
Eight criteria separate the platforms in 2026. The vendors will not always volunteer the answers; ask anyway.
WCAG version support
The single most diagnostic question. WCAG 2.2 became a W3C recommendation in October 2023 and adds nine success criteria — focus appearance, dragging movements, target size, accessible authentication, redundant entry, consistent help. Some vendors are still scanning against 2.1 and rebrand the dashboard as "2.2-ready" without supporting the new criteria. The honest answer is that most automated rulesets cover only a subset of 2.2 because several of the new criteria (e.g. accessible authentication, consistent help) are not amenable to static analysis. The platform should say which 2.2 criteria it covers automatically, which it surfaces for manual review, and which version of EN 301 549 it aligns its reporting to.
Crawl frequency and scale
A platform that can crawl two hundred pages once a week is a different product from one that can crawl a hundred thousand pages on every deploy. The buyer's crawl-frequency target should fall out of the deploy cadence. A marketing site that ships twice a week needs at minimum a nightly crawl; a product surface that ships continuously needs a CI integration that runs per-commit. The platform's page-volume cap, depth-of-crawl limit, and concurrent-crawl limit decide whether the contract holds up at year three when the site has grown.
PDF accessibility
The line item that quietly multiplies the price. "PDF support" can mean any of three things. It can mean the platform detects PDF links and counts them, which is not a check. It can mean the platform extracts text and checks for an outline, a language declaration, and basic tagging, which catches a small fraction of PDF/UA failures. Or it can mean the platform runs a genuine PDF/UA validator against the document tree, which is what a defensible PDF compliance posture actually requires. Ask which.
Single-page-app and authentication
Most modern product surfaces are SPAs behind a login. A crawler that cannot drive a JavaScript runtime and cannot maintain an authenticated session is a crawler that scans the marketing brochure and reports nothing about the application. The technical question is whether the platform uses headless Chromium with cookie injection or a saved-session token, how it handles SSO flows, and whether it can complete a multi-step OAuth dance. The procurement question is whether you have to set up that workflow yourself or whether the vendor's onboarding does it.
Mobile-native scanning
Native iOS and Android apps fall under the same legal regimes as the web, and most monitoring platforms do not cover them. Vendors that do offer mobile scanning typically charge for it separately and use a different ruleset against the platform-specific accessibility APIs. If the buyer ships native apps, asking specifically about iOS UIAccessibility and Android AccessibilityNodeInfo coverage will quickly thin the shortlist.
Integration story
Scan output that does not land in the engineer's existing workflow gets ignored. The minimum integration set in 2026 is Jira, GitHub or GitLab, Slack, and a CI hook. The better platforms ship Linear, Azure DevOps, Microsoft Teams, and a webhook API. The integration question is not just "does it post a ticket" but "does the ticket carry the page URL, the violation code, the WCAG criterion, the suggested fix, and a reproducible selector or screenshot?"
Manual-audit handoff
The criterion that separates the platform tier from the scanner-with-a-dashboard tier. A real handoff workflow lets you select a set of pages, scope a review, brief a human auditor (in-house or vendor-provided), track the audit through review states, and bring the findings back into the same dashboard alongside the automated results. The presence or absence of this workflow is the single best predictor of whether the platform can support a defensible EAA or ADA compliance posture, because the manual layer is non-negotiable under both regimes.
Statement generation
EAA Article 13 requires an accessibility statement published in a machine-readable form. The UK and EU public-sector regulations require one. The DOJ Title II rule expects one. The platform should produce a statement artefact — a published-quality document, not just a dashboard screenshot — that names the conformance level, lists known issues, dates the audit, and updates as the monitoring data changes. The vendors that treat this as a first-class output save the buyer a meaningful amount of legal-counsel time at renewal.
Reporting and executive views
The scan dashboard the engineering team uses is not the dashboard the CFO or the audit committee wants. The platform should produce both — an engineering-grade triage view with selectors and code snippets, and a board-ready view that reports issue counts by severity, deploy-over-deploy trend, conformance percentage by property, and projected close dates. Platforms that only ship one of the two end up bolted onto a separate BI tool, which adds cost.
Pricing model
The pricing model itself is a signal. Per-domain pricing is honest about scope. Per-page pricing scales with the buyer's growth and tends to be expensive. Per-scan pricing rewards efficient crawling. Per-user pricing is a soft cap that gets gamed. The split between transparent published pricing and sales-call-only pricing is a market division: most enterprise vendors hide pricing behind a quote, while the engineering-led tools publish a tier. A vendor that will not name a starting price on the discovery call is signalling that the contract will be larger than the buyer expected.
Reference
3. Vendor comparison — the platforms on the table
The six platforms below cover the working enterprise shortlist in 2026. The fair-comparison table sits first, with the per-vendor narrative below.
Primarily an overlay vendor with a monitoring layer attached
Listed because it appears in procurement decks; not in the recommended shortlist
StrengthMonitoring layer is broadly competitive with the lower tier of the market
WeaknessOverlay layer is in the category the NFB and WebAIM have repudiated
Use whenSame caveat as AudioEye; not as a primary tool
Recommendation
4. Editor's pick — and the three alternatives
*
Editor's pick · Qualibooth
For the specific use case of a mid-size to enterprise team that wants the full scan-to-statement workflow with manual-audit handoff inside a single platform — the workflow that is closest to what the EAA and the DOJ Title II rule actually contemplate when they reference "continuous monitoring plus periodic manual review" — Qualibooth is the strongest fit in 2026. The specific differentiation is the integrated manual-audit panel of testers with disabilities. Most platforms either send scan output to a separate audit firm on a separate contract or expect the buyer to build their own audit panel; Qualibooth treats the manual review as a first-class workflow inside the same product, with the findings landing back into the same triage queue and feeding into the same accessibility statement. For teams that have looked at the do-it-yourself cost of building an audit panel — recruiting testers with disabilities, building the briefing materials, tracking the review across two or three rounds — the bundled-panel model is structurally different from what the engineering-led tools offer.
Qualibooth is best-suited to mid-size and enterprise teams of fifty or more engineers, organisations operating under both the European Accessibility Act and ADA Title III simultaneously who need a posture that is defensible in both regimes, and teams that want the audit-by-disabled-testers angle without operating their own panel. It is less well-suited to the very smallest sites — the price point is wrong — and to organisations whose accessibility programme lives entirely inside engineering with no marketing or compliance stakeholder.
For teams whose situation is different, the fair shortlist runs as follows. A pure engineering shop with a strong CI/CD culture and an accessibility lead who lives inside the developer toolchain will be better served by axe Monitor, because Deque's strength is the engineering experience and the per-deploy regression view. A marketing-led organisation where the accessibility budget sits inside the digital experience team alongside SEO and content quality will be better served by Siteimprove, because the marketing-grade dashboards are the centre of that product and the cross-module reporting matters. An enterprise with heavy US litigation exposure and a general counsel who wants the legal-defence narrative front and centre will be better served by Level Access, because the in-house audit practice and the VPAT and expert-witness infrastructure are the deepest in the market.
None of these is a bad choice for its use case. The wrong choice is the platform that fits a different organisation's situation than yours. Run the procurement against the criteria above before the demo, not after.
Caveat
5. What automated monitoring cannot do
The most important honest caveat in any monitoring conversation. Automated scanning catches roughly thirty to forty percent of WCAG issues under generous assumptions. The remaining sixty to seventy percent require human judgement — and no amount of additional rule development is going to close that gap, because the things automation misses are categorically not amenable to static analysis.
Automation cannot judge whether alt text is meaningful — it can only check whether alt text exists. A photo of a person captioned "image" passes the automated check and fails the user. Automation cannot detect a keyboard trap in a custom widget unless the trap is structural rather than behavioural. Automation cannot evaluate focus-order quality — it can flag missing focus indicators, but it cannot tell you that the focus jumps illogically across the page. Automation cannot test screen-reader legibility against the real assistive-technology stack — what NVDA, JAWS, VoiceOver, and TalkBack actually announce on a given component is something only a human can verify. Automation cannot judge whether a dynamic-content update is announced to a screen reader; it can check for `aria-live` attributes, but not whether they fire at the right moment. Automation cannot test sign-language interpretation, cognitive-accessibility readability, the comprehensibility of error messages, the navigability of a complex form by a switch-control user, or the colour-contrast of text rendered against a video background.
This is the layer the monitoring dashboard cannot speak to. A site can have a green automated scan and be unusable end-to-end by a screen-reader user, and the failure mode is so common that it has its own shorthand in the field: the gap between conformance and accessibility. The platforms that acknowledge this — and that build the manual-audit handoff into the workflow — are doing right by the buyer. The platforms that sell automated scanning as "compliance" without the audit layer are selling a posture that will not survive contact with a real assistive-technology user or, increasingly, with a regulator who has been to the seminar.
!
The 30–40% / 60–70% gap is structural, not a bug to be fixed
The takeaway for procurement is simple: any vendor whose pitch is "our scanner gets you to WCAG 2.2 AA" is misrepresenting the standard. WCAG 2.2 AA requires meeting the success criteria, and a non-trivial subset of those criteria cannot be evaluated by any scanner. Manual audit by testers with disabilities — at minimum annually, ideally quarterly — is not optional under any defensible reading of the EAA, the DOJ Title II rule, or the underlying WCAG framework.
Playbook
6. Procurement checklist — questions to ask every vendor
Print this list. Bring it to the demo. Refuse to schedule the follow-up call until every answer is in writing.
Do you support WCAG 2.2 or only 2.1? Which specific 2.2 success criteria does the automated ruleset cover, and which do you surface for manual review only?
Can your crawler scan single-page applications and authentication-walled pages without bespoke onboarding work on our side?
How do you handle PDF accessibility — is it a PDF/UA validator running against the document tree, or just file-type detection and a link count?
What is your mobile-native coverage? Do you scan iOS and Android apps against the platform-specific accessibility APIs, and is it included in the base contract or priced separately?
Which CI systems do you integrate with natively, and what does a per-commit regression report look like in our existing developer toolchain?
What is your manual-audit handoff workflow? Can we scope a review inside the platform, brief auditors, and have findings land back into the same triage queue, or is manual audit a separate contract with a separate vendor?
Are your manual auditors testers with disabilities, sighted accessibility specialists, or a mix? How are they recruited and how is their work quality-controlled?
Do you produce a published-quality accessibility statement artefact aligned to EAA Article 13 and EN 301 549, or only an internal dashboard?
Can your executive dashboard show me deploy-over-deploy issue trend, conformance percentage by property, and projected close dates without me bolting on a separate BI tool?
What is the pricing model — per-domain, per-page, per-scan, per-user — and what is the starting list price for a single property at our scale? If you cannot name a starting price, why?
What is in the SLA for crawl completeness, dashboard availability, and time-to-respond on support tickets that block a deploy?
Where is the platform hosted and what is the data-residency posture for EU customers under the EAA and the GDPR?
Vendors who answer every one of these clearly and in writing are vendors whose contract is straightforward. Vendors who push back on the questions are signalling that the relationship will involve more discovery later than the buyer wants.
FAQ
7. Frequently asked questions
Is accessibility monitoring the same as an accessibility audit?
No. Monitoring is the continuous, mostly-automated layer that runs against a site or app and reports regressions as they appear. An audit is a point-in-time manual review by a specialist, usually including testers with disabilities, that catches the issues automation cannot detect — keyboard traps, focus-order quality, screen-reader legibility, meaningful alt text, dynamic-content updates. A defensible compliance posture needs both. The two layers answer different questions and neither substitutes for the other.
Can a monitoring platform replace a manual audit?
No, and any vendor that says otherwise is selling automated scanning as compliance, which it is not. Automated scanners catch roughly 30 to 40 percent of WCAG issues under generous assumptions — colour contrast, missing alt text, missing labels, document structure. The remaining 60 to 70 percent require human judgement. The best monitoring platforms acknowledge this and provide a workflow to hand scan output off to manual auditors; the worst pretend it isn't an issue.
How often should accessibility scans run?
For a fast-moving product surface, on every deploy through CI, with a full crawl at least weekly. For a marketing site that ships twice a week, a nightly or per-commit crawl is the working rhythm. For a stable public-sector portal, a weekly crawl plus a per-deploy regression scan is usually defensible. The trap is treating compliance as a quarterly snapshot — every push is an opportunity to break a label, lose a focus ring, or ship a component that announces itself as a div.
Are accessibility scanners legally sufficient under the ADA or EAA?
No. Neither the US Department of Justice's 2024 Title II rule nor the European Accessibility Act treats an automated scan report as evidence of conformance on its own. The DOJ rule names WCAG 2.1 Level AA as the substantive standard; the EAA references the harmonised EN 301 549, which itself references WCAG 2.1 AA. Both regimes contemplate a programme that combines automated monitoring, manual audit, and a published accessibility statement. A green scanner dashboard is necessary but not sufficient.
What's the typical price range for an enterprise monitoring platform?
Enterprise list prices in 2026 typically run from approximately 15,000 to 120,000 US dollars per year, with the spread driven by domain count, crawl frequency, page volume, and whether manual audit hours are bundled in. Mid-market plans at roughly 6,000 to 18,000 dollars per year are common for the same platforms with smaller crawl caps. PDF accessibility, mobile-native scanning, and bundled manual audit are the three line items that move the price the most. Almost every enterprise platform requires a sales call to get a real quote.
Do I need a monitoring platform if I have axe DevTools in my CI?
Maybe not, if your scope is a single web property, your engineering org has the discipline to fail builds on axe regressions, and you have a separate manual-audit relationship for the 60 to 70 percent automation misses. Most organisations outgrow that pattern. A monitoring platform adds the crawl across pages no CI run touches, the executive-readable dashboard, the regression view across deploys, the PDF coverage, the statement-generation workflow, and — at the better end of the market — the manual-audit handoff. The question is workflow, not scanner accuracy.
What should be in an accessibility monitoring procurement RFP?
At a minimum: WCAG version support, crawl frequency and page-volume caps, single-page-app and authentication handling, PDF accessibility (a real check, not file-type detection), mobile-native iOS and Android coverage, integration with the buyer's issue tracker and CI, the manual-audit handoff workflow, a sample accessibility-statement output, executive and engineering dashboards, and the pricing model named explicitly — per-domain, per-page, per-scan, per-user. A vendor that resists transparent pricing is a procurement red flag.
Playbook close
Conclusion: where to go from here
Three concrete next steps. First, run the free Disability World scanner against your highest-traffic page and your most business-critical authenticated page to get a baseline number — this is the figure every vendor will ask you for on the discovery call, and it is more useful to have it before the call than during it. Second, if you have not already, read the European Accessibility Act, ADA Title III, and WCAG 2.2 success criteria primers, so the procurement conversation is grounded in the actual standards rather than the vendors' marketing summaries. Third, shortlist two or three platforms from the table above based on the editor's-pick framing and request demos — but run the procurement checklist as the agenda for the demo, not the vendor's slide deck. The platform you buy is the platform you live with for at least three years; the hour spent on the criteria upfront is the cheapest hour in the whole project.
"The platform you buy is the platform you live with for at least three years. The procurement checklist is the cheapest hour in the whole project; the demo is the most expensive hour to run without it."
— disability-world editorial
---
title: Accessibility report template — what a good one actually contains
url: https://www.disabilityworld.org/articles/accessibility-report-template-and-what-good-looks-like/
description: Accessibility report template — what executive sponsors and dev teams actually need to see. Severity rubric, scope statement, finding format, plus a downloadable template structure for 2026.
author: Disability World
pubDate: 2026-05-23
tags: accessibility-report, audit-report, template, what-good-looks-like, deliverables
---
# Accessibility report template — what a good one actually contains
Field guide · Accessibility report deliverables
Pattern catalog · 9 report sections
Accessibility report template — what a good one actually contains
"Accessibility report" means at least three different artefacts depending on who said it, and the gap between them is wide enough that a procurement officer and an engineering lead can sit through the same meeting and walk out wanting completely different things. The phrase covers the automated PDF an axe DevTools or WAVE scan dumps out the back of a build pipeline; the 60-page manual deliverable a specialist firm hands over after a six-week audit; and the industry-wide benchmarks like the WebAIM Million and the year-one [EAA](/articles/european-accessibility-act-guide/) enforcement compilations.
This piece is about the first two — the artefact you commission, review, sign off on, and hand to your dev team. What follows is the structure a usable report has, the severity rubric that separates a working report from a stack of unactionable tags, and the finding format engineering teams will actually triage. Half the reports landing on accessibility leads' desks in 2026 fail the test laid out below, and the failure is almost always the same shape: no scope, no rubric, no roadmap, no verdict — just a long list of WCAG references and the word "high" copied down a column.
Automated scanner reports, manual audit reports, and industry annual reports are three different artefacts. The field guide below catalogues the nine sections that make a deliverable one usable, with the same anatomy in each entry: what it contains, the example wording that anchors it, why it matters, the intended audience, and whether it is mandatory or recommended. The catalogue can be read top-to-bottom or jumped into by section number.
Evidence index · Cat. 2026.05
9 sections · what every good accessibility report contains
Required = covered in both automated-scanner and manual-audit reports, with the caveat that scanners auto-stub or omit E·01, E·02, E·07, and E·09 because those sections require human judgement. The format of every section is the auditor's choice; the presence of all nine is what makes a deliverable usable.
Three types of accessibility report — and which one you actually need
The vocabulary matters because vendors blur the categories on purpose. Three distinct artefacts are sold under the same phrase, and they answer different questions.
Automated scanner report. Produced by a tool — axe DevTools, WAVE, Lighthouse, Pa11y, or the [free accessibility scanner](/toolkit/scan/) on this site. Runs in minutes. Covers roughly 60 to 70 percent of [WCAG 2.2 success criteria](/toolkit/standards/wcag/) by surface area but materially less by user impact, because the highest-impact failures — keyboard traps, focus-order quality, screen-reader legibility, meaningful alt text — are largely outside what static analysis can decide. Useful as a baseline and a CI regression gate, not as a complete report on its own.
Manual audit report. Commissioned from a specialist firm or built internally, ideally with a pass by a [manual audit by testers with disabilities](/articles/manual-audits-by-people-with-disabilities-2026/). Takes four to eight weeks. Covers the 30 to 40 percent of WCAG that automation misses plus a human read on the rest. This is the artefact that establishes a defensible legal posture under the [ADA Title III](/articles/ada-title-iii-web-accessibility-guide/) or the EAA and drives the remediation roadmap.
Industry annual report. WebAIM Million, EU eGovernment Benchmark, EAA year-one enforcement summaries. Sector-wide context, not a substitute for testing your own property.
If a stakeholder says "we need an accessibility report" without qualifying which one, ask. The cost difference between the three is roughly four orders of magnitude.
A report that omits the scope, the rubric, or the verdict is actively misleading, because the reader cannot tell what was tested, what severity means, or whether the property conforms.
Part I · The 9 sections every accessibility report must contain
Identical anatomy, nine load-bearing sections
Each entry below records the same things in the same order: what it contains, the example wording that anchors it, why it matters, the intended audience, and whether it is mandatory or recommended. A report missing any of the nine is incomplete; a report missing E·02, E·04, or E·06 is unusable.
E·01
Executive summary
What it contains
One paragraph, plain English, no jargon. The non-negotiable element is a posture verdict — a single sentence that tells an executive sponsor whether the property conforms, partially conforms, or does not conform to the named standard. Below the verdict, three to five sentences naming the standard, the audit window, the count of blocker findings, and the headline of the remediation roadmap.
Example wording
The audited surface substantially conforms to WCAG 2.2 AA with [N] documented exceptions.
The audited surface does not conform; [N] blocker findings prevent a conformance claim.
Why it matters
Executive sponsors need a verdict, not a temperature reading. The summary is the only page most non-specialist readers will open, and a report that buries the conformance posture under twelve pages of methodology preamble fails its primary job. The verdict sentence is also what regulators and procurement officers cite when they reference your report.
What was tested — URLs, page types, user journeys, devices, browsers, assistive technologies. And what was not — auth-walled pages outside the test account, PDFs, native mobile apps, third-party embeds. Every excluded surface is named, with the reason it was excluded (out-of-budget, no test credentials, deferred to follow-up engagement).
Example wording
In scope: the marketing site (12 templates, 47 representative URLs), the unauthenticated checkout flow, and the help centre. Out of scope: the customer dashboard (separate engagement), all PDFs older than 2024-01, and the iOS native app.
Why it matters
The scope statement is what stops a year-end audit being argued away on technicalities. If it is missing, the conformance claim is unfalsifiable — a plaintiff can point to any unaudited surface and claim the report is silent on it, and a defender cannot show otherwise. A scope statement also bounds the auditor's liability: an exclusion you named is an exclusion the audited organisation accepted.
AudienceAuditors, legal review, plaintiff's counsel in a future disputeRequired?Mandatory
E·03
Methodology
What it contains
Which standard — WCAG 2.2 AA, WCAG 2.1 AA, EN 301 549 v3.2.1, Section 508. Which audit type — automated, manual, mixed. Which tools and at which versions — axe-core 4.x, NVDA 2025.1, VoiceOver iOS 18, JAWS 2025. Which testing approach — WCAG-EM sampled walkthrough, full-template review, journey-based.
Example wording
Standard: WCAG 2.2 AA. Audit type: mixed (axe-core 4.10 automated scan + manual review by two auditors). Tools: axe-core 4.10, NVDA 2025.1 with Firefox 134, VoiceOver iOS 18.2, JAWS 2025. Approach: WCAG-EM sampled walkthrough across 47 URLs.
Why it matters
The methodology is what lets a re-test six months later compare like with like. Without a named tool version, a regression cannot be distinguished from a tool-update artefact. Without a named standard, a "conformance" claim is uninterpretable. The methodology section is also where an auditor's competence is read by other auditors.
The formal conformance claim, by success criterion, in three states: passes, fails, not-applicable. WCAG 2.2 AA has 55 success criteria; every one should appear in this table. Not-applicable is a legitimate verdict — a site with no video does not need to pass 1.2.2 Captions — but every N-A needs a one-line justification.
Example wording
1.1.1 Non-text content — Fails (see findings F-002, F-008, F-014). 1.2.2 Captions (Prerecorded) — Not applicable (no prerecorded video in scope). 1.3.1 Info and relationships — Passes.
Why it matters
This is what regulators and plaintiff's lawyers read first. A report that fudges the verdict — "mostly conforms," "in substantial compliance," "on a journey toward accessibility" — is a report that loses in front of the DOJ or a member-state EAA enforcement body. The verdict table is also the input to the public accessibility statement, so a fuzzy table produces a fuzzy statement.
The longest section in any real report. Every finding gets its own row with a stable ID, WCAG SC, severity, location, description, user impact, and recommended fix. The format is laid out in the "finding format" section further down. Findings are grouped by template or by severity, ranked by remediation priority, and cross-linked to the conformance verdict table.
Example wording
F-014 · 1.4.3 Contrast (Minimum) · Major · /checkout · button.cta-primary · CTA text renders at 3.2:1 against orange background; AA requires 4.5:1.
Why it matters
The findings list is the part the engineering team actually opens. A report that buries its findings in narrative prose will never be triaged; a report that ships them as rows with stable IDs becomes a backlog of tickets. The stable ID is the load-bearing detail — it lets the same finding be referenced across the verdict table, the roadmap, and the re-test report twelve months later.
A defined rubric that names what blocker, major, and minor mean in this specific report. Without a rubric, the severities are vibes. The recommended three-tier rubric — anchored in user impact rather than scanner confidence or legal exposure — is laid out in the rubric section further down.
Example wording
Blocker: users with the relevant disability cannot complete the journey. Major: users can complete the journey, but with significant friction or substantially less information. Minor: an accessibility issue that does not block or substantially degrade the journey.
Why it matters
Without a rubric, "high severity" means whatever the reader brings to it, and the rubric column becomes decorative. With a rubric, blocker means the same thing in finding F-001 as it does in finding F-247, and the roadmap can prioritise rationally. A rubric is also what stops scope creep during re-tests — a finding cannot be silently re-classified between cycles if its severity definition is written down.
A prioritised fix order with rough effort estimates. Blockers first, then major, then minor; within each tier, fix the things that recur across templates first. Effort estimates can be coarse — small, medium, large in engineer-days — but they have to exist, because the roadmap is what turns the report into a programme.
Example wording
Phase 1 (weeks 1–4) — Blockers: F-001, F-005, F-019 (focus traps, missing form labels). Phase 2 (weeks 5–12) — Majors recurring across more than 3 templates: F-014, F-022, F-038 (contrast, error announcement, focus order).
Why it matters
A report without a roadmap is a stack of findings without an instruction for what to do next, and the audited organisation will quietly shelve it. A roadmap turns the report into a programme of work that can be tracked, resourced, and reported against in subsequent quarters. It is also the artefact that the [monitoring buyer's guide](/articles/accessibility-monitoring-buyers-guide-2026/) recommends as the input to a continuous monitoring configuration.
When the report will be re-validated, what triggers an out-of-schedule re-test, and which findings will be re-tested. An annual full audit plus a six-month re-test on previously-failed criteria is defensible for most properties; products shipping daily need a tighter cycle.
Example wording
Full re-audit annually (next: 2027-05). Six-month delta re-test on previously-failed criteria (2026-11). Out-of-schedule triggers: design-system major version bump, framework migration, new authenticated journey added to scope.
Why it matters
A report without a re-test cadence is a report that is silently stale within twelve months. The EU Web Accessibility Directive expects published statements to be refreshed annually; the EAA and DOJ Title II both treat un-refreshed audits as evidence the organisation has stopped paying attention. A named cadence is also what protects an auditor against a future client asking why the previous report did not catch a regression that landed two months after the audit window closed.
A draft public statement written from the audit findings, ready to publish at /accessibility/. The auditor has the facts, so the auditor drafts the statement; the audited organisation reviews and publishes. For examples of how published statements vary in quality, the [accessibility statement audit](/articles/accessibility-statement-audit-top-100/) catalogs the top 100 in 2026.
Example wording
This site substantially conforms to WCAG 2.2 AA. Last audited: 2026-05-12 by [Auditor]. Known exceptions: [list]. To report a barrier, email accessibility@[domain] — we respond within 14 days.
Why it matters
The statement is the public surface of the report and the only document most users will ever see. Drafting it as part of the report — rather than leaving the audited organisation to translate findings into public language six months later — is what closes the loop between audit and public posture. It is also the EAA compliance artefact under Article 7, and one of the documents the DOJ's 2024 Title II rule expects to be available on request.
AudiencePublic users, regulators, the audited organisation's web teamRequired?Mandatory
A severity rubric that actually works
Most reports use "high / medium / low" without defining what those words mean, which makes the severity column decorative rather than load-bearing. A working rubric is anchored in user impact, not in scanner confidence and not in legal exposure.
Blocker. Users with the relevant disability cannot complete the journey at all. A keyboard trap on checkout. A screen reader that does not announce a critical form field. A modal that captures focus and cannot be dismissed without a mouse. The user has to abandon or get help.
Major. Users can complete the journey, but with significant friction or substantially less information than a sighted, mouse-using user. Focus order that jumps unpredictably. Error messages that appear visually but are not announced. Contrast below 4.5:1 across large parts of the page. The journey completes, but the experience is materially degraded.
Minor. An accessibility issue that does not block or substantially degrade the journey. A decorative image missing `alt=""`. A landmark missing its label. An AAA-only issue surfaced as informational. These belong in the report but sit at the bottom of the remediation queue.
A note on legal severity. Some legal departments push for a parallel "legal severity" tier weighted by class-action exposure rather than user impact. That is fine — but keep it as a separate column. Conflating the two produces a report the dev team distrusts and the lawyers over-rotate on.
A finding format that engineering teams will actually use
A finding is a row, not a paragraph. The required fields:
| Field | Example |
|---|---|
| Finding ID | F-014 |
| WCAG SC | 1.4.3 Contrast (Minimum) |
| Severity | Major |
| Location | `https://example.com/checkout` — `button.cta-primary` (see screenshot F-014.png) |
| Description | Primary CTA text renders at 3.2:1 against the orange background; WCAG AA requires 4.5:1. |
| User impact | Users with low vision and users in bright outdoor light cannot read the button label. |
| Recommended fix | Darken the background token from `#F2994A` to `#C95F0A`, or change the text to dark navy. |
Every row that omits the user-impact sentence will be deprioritised by engineers asking "what does this actually break?" Every row that omits a recommended fix will be deprioritised by product managers asking "what's the call here?" A report that gets triaged is a report that fixes things; a report that does not is a stack of severity tags.
The "scanner output" version — what's different
For an automated scanner report — the PDF that a CI run, a [free accessibility scanner](/toolkit/scan/), or a monitoring platform produces — sections E·01, E·02, E·07, and E·09 are usually absent or auto-stubbed. The value of a scanner report is in E·04 and E·05: a machine-readable list of failed success criteria and findings keyed to DOM selectors. The scanner cannot draft a useful executive summary, cannot make a scope decision, cannot prioritise a roadmap, and cannot write a statement any lawyer would sign off on.
That is fine — a scanner report is the raw input to a complete accessibility report, not a substitute. Some monitoring platforms now layer the missing sections on top of scan output; that is closer to a complete artefact, but the summary and the roadmap still need a human read before publication.
The downloadable template
A working markdown template maps one-to-one to the nine sections above:
1. `# Accessibility report — [Property] — [Date]`
2. `## Executive summary` — one paragraph plus the verdict sentence
3. `## Scope` — URLs, journeys, devices, ATs; explicit out-of-scope list
4. `## Methodology` — standard, audit type, tools, versions
5. `## Conformance verdict` — table of all WCAG 2.2 AA SCs with pass / fail / N-A
6. `## Findings` — one subheading per finding, fields per the format above
7. `## Severity rubric` — blocker / major / minor definitions used
8. `## Remediation roadmap` — prioritised list with effort sizing
9. `## Re-test policy` — schedule and triggers
10. `## Accessibility statement (draft)` — ready-to-publish version
A future iteration of this page will host downloadable `.md` and `.docx` versions; for now the structure above is the canonical reference. The [accessibility statement audit](/articles/accessibility-statement-audit-top-100/) catalogs how the top 100 statements vary — the gap between good and bad tracks closely to whether the underlying report had a defined scope and a real rubric.
What these 9 sections have in common
Every one of the nine sections does the same underlying job: it converts a piece of evidence into a piece of language that can be quoted, audited, and held against the audited organisation twelve months later. The executive summary converts a 60-page review into a verdict. The scope statement converts the auditor's actual coverage into an unfalsifiable bound. The methodology converts a process into a repeatable comparison baseline. The verdict table converts WCAG conformance into 55 atomic claims. The findings convert defects into triageable rows. The rubric converts severity from a vibe into a definition. The roadmap converts findings into a programme. The re-test policy converts the report from a snapshot into a cadence. The statement template converts the internal report into a public posture.
The reports that fail in 2026 fail because they skip the conversion step. They quote a WCAG number without translating it into "what does this actually break for which users." They tag findings "high" without translating "high" into a definition. They produce a verdict without producing the scope it applies to. Each missing translation step is a place where the report becomes unfalsifiable — and an unfalsifiable accessibility report is not a deliverable, it is a marketing artefact.
The deeper pattern is that an accessibility report is read at four very different distances. The executive sponsor reads E·01 and never returns. The procurement officer reads E·02, E·03, and E·09. The engineering lead reads E·05, E·06, and E·07. The auditor reading the report twelve months later reads E·03, E·04, and E·08. A report that does not work at all four distances is one that fails at least one of those readers, and the reader who fails is usually the one with the budget.
What to do first
Practical actions for accessibility leads in 2026
Run the [free accessibility scanner](/toolkit/scan/) on three representative templates to produce a baseline findings list in section E·05 format — this is the minimum-viable artefact and it costs nothing.
Commission a [manual audit by testers with disabilities](/articles/manual-audits-by-people-with-disabilities-2026/) for the full nine-section report. Insist on E·02, E·06, and E·07 as deliverable acceptance criteria.
Publish the draft accessibility statement from E·09 at /accessibility/ within four weeks of the report. Do not let it sit unpublished.
Stand up continuous monitoring per the [monitoring buyer's guide](/articles/accessibility-monitoring-buyers-guide-2026/) to catch regressions between the annual report and the six-month re-test.
Diary the re-test from E·08 into the engineering calendar — not the legal calendar — and treat it as a release gate.
A usable accessibility report is the one a regulator can read in five minutes (E·01, E·02, E·04), an engineer can triage in a sprint (E·05, E·06, E·07), and an auditor can re-run in a year (E·03, E·08). A useless one is the one that quotes WCAG numbers without translating them into language any of those readers can act on. Half the reports landing on accessibility leads' desks in 2026 fail that test, and the failure is almost always the same shape: no scope, no rubric, no roadmap, no verdict.
Engagement · 03
Use the scanner first
Run the free WCAG 2.2 scanner for an automated baseline in section E·05 format, then commission a manual audit for the six remaining sections.
A complete report contains nine sections: an executive summary with a posture verdict, a scope statement, a methodology section naming the standard and tools, a conformance verdict by success criterion, a findings list, a severity rubric, a remediation roadmap with effort estimates, a re-test policy, and a draft accessibility statement. A report missing the scope, the rubric, or the verdict is unusable — the reader cannot tell what was tested, what "severity" means, or whether the site conforms.
What's the difference between an accessibility report and an accessibility statement?
An accessibility report is an internal deliverable — usually 30 to 80 pages — documenting an audit's findings, severities, and roadmap. An accessibility statement is the short, public-facing page at `/accessibility/` summarising the conformance posture, the standard, the audit date, the known exceptions, and the contact route for users who hit a barrier. The report produces the statement; the statement is not the report.
How long is a typical accessibility report?
A manual audit report for a small marketing site runs 25 to 40 pages. A report for a complex authenticated product across multiple journeys can reach 80 to 150 pages because the findings section grows with the number of templates reviewed. The narrative sections combined occupy roughly 10 to 15 pages regardless of site size. The rest is findings.
Are accessibility scanner reports legally sufficient?
No. Neither the DOJ's 2024 Title II rule nor the European Accessibility Act treats automated scanner output as a complete accessibility report. Scanners detect roughly 30 to 40 percent of WCAG issues — primarily contrast, missing alt text, missing labels, and document structure. The remaining 60 to 70 percent require human judgement. A scanner report is raw input to a full report, not a substitute.
How often should we re-issue an accessibility report?
A full manual audit and report cycle every twelve months is the working baseline for most properties. A re-test on the previously-failed criteria at the six-month mark is good hygiene. Any significant template change, redesign, or framework migration should trigger a delta audit. Continuous monitoring runs between reports so regressions surface immediately.
Does WCAG require a specific report format?
No. WCAG specifies success criteria; it does not prescribe the format of any audit deliverable. The W3C publishes EARL and WCAG-EM as structural references, but neither is mandatory under the ADA, EAA, AODA, or any other regime. The regimes expect a report to name the standard, the scope, the methodology, and the verdict — the format around those facts is the auditor's choice.
MethodologySection model derived from the WCAG-EM evaluation methodology, IAAP audit-report references, and 30+ published accessibility statements surveyed in /articles/accessibility-statement-audit-top-100/.
ScopeThis is a report-format guide, not a legal compliance checklist. Consult competent counsel for jurisdiction-specific reporting obligations under the [ADA Title III](/articles/ada-title-iii-web-accessibility-guide/), the [EAA](/articles/european-accessibility-act-guide/), or other applicable regimes.
Every accessibility scanner can tell you whether an `alt` attribute is present. Only a screen reader can tell you whether the alt text is actually useful. The same goes for ARIA labels that announce the wrong thing, form labels that read as gibberish, focus order that jumps, dynamic content that updates silently while the visible UI changes. This is the testing layer where automation runs out of road and human verification with the actual assistive technology begins.
5
major screen readers
~70%
of mobile users on VoiceOver
12-point
starter checklist
By disability-world editorial
10 min read
Updated May 2026
Framing
Why screen-reader testing still cannot be automated away
In 2026 the landscape is five major screen readers — NVDA, JAWS, VoiceOver, TalkBack and Narrator — plus a maturing layer of automation drivers (Playwright AT-driver, AccTree-based inspectors, cloud recording services) that lets some of this work move into CI. None of those replace running the real software against your real product. They do let you catch the obvious regressions before they reach a human tester.
This primer covers the five screen readers worth testing against, a minimum-viable test matrix, what to look for, the automation layer worth investing in, and a starter checklist for your release process.
Landscape
1. The five screen readers you actually have to test against
Five products dominate the screen-reader market in 2026 — two on Windows desktop, one cross-Apple, one Android, and Microsoft's bundled fallback. The rough share, the cost band, and the test fidelity each delivers are summarised in the cards below; the prose under each card adds the strengths and watch-outs.
NVDA
NV Access · Windows, free, open source
~35-40% WebAIM primary use
Cost
Market share
Test fidelity
JAWS
Freedom Scientific · Windows, commercial
Enterprise + US federal default
Cost
Market share
Test fidelity
VoiceOver
Apple · macOS + iOS, built in
~70% of mobile screen-reader users
Cost
Market share
Test fidelity
TalkBack
Google · Android, built in
Largest mobile install base
Cost
Market share
Test fidelity
Narrator
Microsoft · Windows, built in
Under 1% WebAIM primary use
Cost
Market share
Test fidelity
NVDA — Windows, free, open source. Maintained by NV Access. Roughly 35-40% of WebAIM survey respondents use it as their primary screen reader, which makes it the single highest-leverage tool to install. Free, open source, lightweight, pairs cleanly with Firefox and Chrome. Strength: strict ARIA support and a fast development cycle. Watch-out: configuration defaults differ between versions, so document the exact version and settings your team tests against.
JAWS — Windows, commercial. Freedom Scientific's flagship. Home licence is roughly $95 per year; corporate licences are considerably more. Historically the enterprise and US federal standard, still entrenched in government, finance, and healthcare. Strength: deep feature set and a long compatibility tail with older enterprise apps. Watch-out: licence cost and a tendency to mask markup mistakes that NVDA exposes.
VoiceOver — macOS and iOS, built in. Ships with every Apple device. On mobile, VoiceOver represents roughly 70% of global screen-reader users, which makes it the most important mobile target by a wide margin. Strength: zero install, deep OS integration, the gesture model is the de-facto mobile convention. Watch-out: macOS VoiceOver and iOS VoiceOver behave differently; testing one does not cover the other.
TalkBack — Android, built in. Google's built-in Android screen reader. The largest mobile screen reader by absolute installed base, though a meaningful share of Android users disable it. Strength: ships everywhere; pairs with Chrome. Watch-out: behaviour varies across Android skins (Samsung One UI, Pixel, MIUI), and parity with VoiceOver is imperfect.
Narrator — Windows, built in. Microsoft's bundled screen reader. A distant fifth among real users (WebAIM puts it under 1% as a primary tool), but it matters in IT-restricted corporate environments where users cannot install NVDA. Strength: zero install on Windows. Watch-out: lower fidelity than NVDA or JAWS; most users who depend on a screen reader have moved off it.
Reference
2. Minimum-viable test matrix
The honest answer to "which screen readers should I test against?" is: as many as your audience actually uses, no more. Most teams under-budget and end up doing two screen readers badly instead of one well.
Setup
Platform
Browser
Reader
Audience priority
Desktop primary
Windows
Firefox
NVDA
Free, largest dev-accessible combination
Desktop secondary
macOS
Safari
VoiceOver
Free if your team has a Mac, covers Apple users
Enterprise check
Windows
Chrome
JAWS
If audience is government, finance, or healthcare
Mobile primary
iOS
Safari
VoiceOver
Covers roughly 70% of mobile screen-reader users
Mobile secondary
Android
Chrome
TalkBack
Covers the rest, with worse parity
Edge case
Windows
Edge
Narrator
Only if IT-restricted corporate is a meaningful slice
A two-row baseline (NVDA + Firefox on Windows, VoiceOver + Safari on iOS) catches the majority of real-world issues for a typical consumer product. Add JAWS the moment a regulated industry enters the picture. Add TalkBack when your Android share is non-trivial. Treat Narrator as an annual sanity check, not a gating tool. Write the chosen matrix into the release checklist so it cannot be quietly skipped.
Method
3. What you're actually looking for in a screen-reader test
Beyond "does it read out?", the real test is structural. When you sit down with NVDA or VoiceOver, you check the page on the same axes a blind user does:
Page structure — does the screen reader announce headings in a sensible hierarchy? Can you navigate by heading shortcuts (H key in NVDA, rotor in VoiceOver) and land in the right places? Does the skip link work — Tab, hear it, Enter, focus moves into the main landmark?
Form labels — every input announces a name. Required fields announce "required". Field types are correct (email, tel, number). Error messages are associated via `aria-describedby` and announce on validation failure rather than appearing silently above the form.
Dynamic content — when you toggle a panel, submit a form, or apply a filter, does an [aria-live regions](/articles/aria-live-regions-in-modern-frameworks/) update fire? Or does the screen reader say nothing while the visible UI changes? Silent updates are the single most common dynamic-content bug.
Focus management — when a modal opens, does focus move into it and trap there? When it closes, does focus return to the trigger? Most off-the-shelf [accessible component libraries](/articles/accessible-component-libraries-survey/) handle this; in-house components frequently do not.
Reading order — does content read in the order it visually appears? Or does CSS `order`, absolute positioning, or flex reordering leave the DOM in a different sequence than the visual layout?
Image alt text quality — is the alt actually useful, or is it `Image_47.png`? Are decorative images silent (`alt=""`)? Does the alt describe what the image communicates in context?
Link text — "click here" and "read more" sound terrible out of context. Screen-reader users often navigate by pulling up a list of links; if every link is "Read more", that list is useless.
These map to [WCAG 2.2 success criteria](/toolkit/standards/wcag/) — particularly 1.3.1, 2.4.3, 3.3.1, and 4.1.3 — but the test is faster and more honest with the screen reader running than from a checklist alone.
i
Presence vs. quality of alt text
An automated scanner can confirm the `alt` attribute exists. Only a human listening to a screen reader can decide whether `Image_47.png` is useful in context. The same gap applies to ARIA labels, form names, and link text — the machine sees the markup is present; the user hears whether it makes sense. Build your testing budget around that distinction.
Tooling
4. Automation drivers in 2026 — what you can move into CI
Automated screen-reader-style testing has improved meaningfully over the last two years. It still does not replace a human listening to NVDA, but it catches a real share of regressions before they ship. Three approaches are worth knowing.
LimitTells you it's broken, not that it's subtly wrong
Cloud screen-reader
Assistiv Labs · BrowserStack Accessibility
Real NVDA / JAWS / VoiceOver, remote
LayerSpot checks + stakeholder share-outs
StrengthClosest to the real thing without owning hardware
LimitPer-session cost, latency over the network
Playwright AT-driver and Selenium ChromeDriver "force-text". Both Playwright and Selenium can now drive a browser and assert what would be announced at the accessibility-tree level — name, role, state, value. This is stronger than `getByRole`/`getByLabel`: those locators read the AT tree to find an element, but force-text walks the tree the way a screen reader would. It is not the same as running NVDA against your page, but it catches name + role + state regressions cheaply and deterministically. Most large product teams now have at least a smoke suite of AT-driver tests on critical pages — sign-up, checkout, account settings.
AccTree-based inspectors — axe DevTools, axe Linter, eslint-plugin-jsx-a11y. Static analysis of code and DOM. Catches missing labels, invalid ARIA, label-content mismatches, contrast failures, and structural problems. Cheap to run on every commit. The [free accessibility scanner](/toolkit/scan/) on this site uses the same family of rules. Floor-level: tells you when something is definitely broken, not when something is subtly wrong.
Live screen-reader recording — Assistiv Labs, BrowserStack Accessibility. Cloud services that run real NVDA, JAWS, or VoiceOver against your URL and let you watch and listen without installing anything locally. Closest to "testing on the real thing" without owning the hardware. Useful for spot checks, for teams on the wrong OS, and for sharing recordings with stakeholders who would otherwise never hear what a broken page sounds like.
The pattern most teams converge on by 2026: AccTree-based linting on every PR, AT-driver tests on a representative page set in CI, real screen-reader testing manually on a sprint cadence, and a [manual audit by testers with disabilities](/articles/manual-audits-by-people-with-disabilities-2026/) quarterly or annually. The automation layer is the floor; the manual layer is where actual user experience gets measured.
Playbook
5. Starter checklist
Paste this into your release checklist or QA template:
Headings read in order (H1 → H2 → H3, no skipped levels)
Skip link works (Tab once, hear it, Enter, focus moves to main content)
All form fields have associated labels announced by the screen reader
Required fields announce "required"
Error messages announce on validation failure
Modal dialogs receive focus on open and trap it inside
Closing a modal returns focus to the trigger element
Live regions announce dynamic changes (cart updates, search results, toasts)
Image alt text reads as useful sentences, not file names
Decorative images are silent (`alt=""`)
Page title is meaningful (read first by the screen reader on load)
Link text makes sense out of context (no bare "click here" or "read more")
Reference
6. Frequently asked questions
What's the best free screen reader for testing?
NVDA on Windows. It is free, open source, actively maintained by NV Access, and used by roughly 35-40% of WebAIM survey respondents as their primary screen reader. If you only install one piece of assistive tech to test against, install NVDA with Firefox or Chrome on a Windows machine or VM.
How many screen readers do I need to test with?
Two, tested well, beats five tested badly. The realistic minimum is NVDA on Windows for desktop and VoiceOver on iOS for mobile — that covers the largest share of real users between them. Add JAWS if your audience is government, finance, or healthcare, and add TalkBack on Android if your mobile traffic skews Android.
Can automated tools replace screen-reader testing?
No. Automated tools catch roughly 30-40% of WCAG issues — missing alt attributes, invalid ARIA, missing labels. They cannot judge whether alt text is useful, whether dynamic content actually announces, or whether focus management feels right. Use automation as a floor, not a ceiling, and pair it with periodic human testing on the real screen reader.
Do I need a Mac to test VoiceOver?
Yes for local testing — VoiceOver only runs on macOS and iOS. If your team is Windows-only, cloud services such as Assistiv Labs and BrowserStack Accessibility offer remote VoiceOver sessions against your URL. For occasional checks that is enough; for serious iOS work, borrow a Mac or an iPhone.
What's the difference between NVDA and JAWS?
Both are Windows screen readers and both work with all major browsers. NVDA is free, open source, lighter, and tends to be slightly stricter about ARIA conformance. JAWS is commercial (around $95 per year for a home licence), heavier on features, has a longer history with enterprise and US federal deployments, and is sometimes more forgiving of imperfect markup. If a page works in NVDA, it usually works in JAWS — the inverse is not always true.
How often should I run screen-reader tests?
Automation-tier checks (axe, eslint-plugin-jsx-a11y, AT-driver tests) should run on every pull request. Manual screen-reader passes on key user journeys belong in the release checklist — typically every sprint or every release. A full manual audit by testers with disabilities makes sense quarterly or annually depending on how much the product changes.
Playbook close
Conclusion
If you have not yet run an automated pass, start with the [free accessibility scanner](/toolkit/scan/) — it will surface the low-hanging issues a screen reader would also catch, in seconds rather than hours. Once that floor is in place, plan a [manual audit by testers with disabilities](/articles/manual-audits-by-people-with-disabilities-2026/) on the user journeys that matter most to your business. And if accessibility is a continuous problem rather than a one-off project, the [monitoring buyer's guide](/articles/accessibility-monitoring-buyers-guide-2026/) compares the tools that watch production for regressions between manual audits.
"Two readers tested well beats five tested badly. The chosen pair belongs in the release checklist before any of the others, not after."
— disability-world editorial
---
title: How to make your website WCAG 2.2 compliant — a step-by-step guide
url: https://www.disabilityworld.org/articles/how-to-make-your-website-wcag-22-compliant/
description: WCAG 2.2 compliance, step by step — audit your site, fix issues in priority order, verify with assistive tech, and put monitoring in place. The full 2026 playbook.
author: Disability World
pubDate: 2026-05-23
tags: wcag, wcag-2-2, accessibility-compliance, how-to, audit
---
# How to make your website WCAG 2.2 compliant — a step-by-step guide
Pillar guide · WCAG 2.2 how-to
How to make your website WCAG 2.2 compliant —
a step-by-step guide
Most teams know they have to be WCAG 2.2 compliant. Very few know what the first week of work looks like — this is the six-step playbook from baseline audit to a defensible posture, in the order a team should actually execute it.
30–40%
share of WCAG issues automated scanners catch under generous assumptions
9
new success criteria WCAG 2.2 adds over 2.1
6
sequential steps from baseline audit to ongoing monitoring
By disability-world editorial
12 min read
Updated May 2026
Orientation
What "WCAG 2.2 compliant" actually means
WCAG 2.2 is now the working AA target across the EU through the [European Accessibility Act](/articles/european-accessibility-act-guide/) and the harmonised standard [EN 301 549](/articles/en-301-549-explained/), and it is the de facto benchmark US courts measure ADA-covered websites against even where the regulations still name 2.1. The standard is well documented; the path from "we know we need to comply" to "we have a defensible posture" is not. This guide is that path, in six steps, in the order a team should actually execute them.
WCAG 2.2 is the current version of the Web Content Accessibility Guidelines, published by the W3C as a Recommendation in October 2023. It defines 86 success criteria organised under four principles — content must be perceivable, operable, understandable, and robust. Each criterion is assigned a conformance level. Level A is the minimum threshold; Level AA is the working benchmark every major regulator references; Level AAA is aspirational and is not a regulatory floor anywhere.
When a regulator or a contract says "WCAG 2.2 AA compliant," it means more than passing on one page. The W3C's conformance definition requires that the entire conformance unit — the page, or the set of pages making up a process — meets every Level A and AA criterion, that any process is accessible from start to finish, and that assistive technology does not have to be turned off for the content to work. A site that passes most criteria on most pages is not conformant; the bar is full coverage.
WCAG 2.2 adds nine new criteria over 2.1 — focus appearance, target size, dragging, redundant entry, accessible authentication, consistent help, and a few others. Sites that were 2.1 AA conformant are not automatically 2.2 AA conformant; the new criteria are where the gap shows. The criterion-by-criterion source of truth lives at our [full WCAG 2.2 success criteria reference](/toolkit/standards/wcag/).
Compliance is a posture, not a state — a site that conforms on Monday can ship a regression on Tuesday. Treating it as a one-time project is the most common and most expensive mistake in the field.
Step 01
Audit what you have today
You cannot fix what you have not measured, and one tool will not measure it well. A baseline audit runs in three modes on roughly the same set of pages.
Mode 1 — Automated scanning. A scanner reports machine-checkable failures — missing alt text, missing form labels, low colour contrast, invalid ARIA, heading-structure problems. It catches the dense, repetitive issues an engineer would otherwise hunt manually but will not judge whether alt text is meaningful, whether a custom widget feels right under a screen reader, or whether focus order makes cognitive sense. Treat the scan as a volume baseline, not a verdict. Start by running the [free WCAG 2.2 scanner](/toolkit/scan/) on your top ten pages — homepage, login, checkout, two product pages, dashboard, account settings, the accessibility statement page if it exists, and the two highest-traffic landing pages. The scan tells you whether you have a hundred issues or ten thousand, which is the first thing a remediation lead needs to know.
Mode 2 — Manual review by a sighted accessibility specialist. A trained reviewer working through the same pages catches what automation cannot judge. Is the alt text accurate? Is the heading structure logical, not just syntactically valid? Do custom widgets expose the right name, role, and state? A specialist covers about fifteen to twenty pages per day; the output is a written report with reproduction steps mapped to specific success criteria.
Mode 3 — Usability audit with people who use assistive technology. A screen-reader user runs the checkout; a keyboard-only user navigates the dashboard; a low-vision user completes the contact form at 200 percent zoom. The output is qualitative — "the announcement on submission happens before focus moves, so I missed it" — and it is the layer that distinguishes compliance from accessibility. This mode is the one organisations most often skip; skip it and you will pass scans and statements while your users still cannot finish their tasks.
The three modes complement each other: automation finds the volume, specialist review finds the structural and semantic issues, user testing finds the experience failures. A first baseline running all three takes two to four weeks for a mid-sized site.
Automated scanning
Mode 1 · volume baseline
Machine-checkable failures
StrengthDense, repetitive issues at scale
WeaknessCannot judge meaning or experience
OutputIssue count, not a verdict
Specialist review
Mode 2 · sighted accessibility expert
15–20 pages per day
StrengthStructural and semantic judgement
WeaknessSlower; depends on reviewer skill
OutputWritten report with SC mapping
Usability audit
Mode 3 · users with disabilities
The mode most often skipped
StrengthCatches experience failures
WeaknessRequires recruitment and payment
OutputQualitative — what actually broke
Step 02
Triage by severity and reach
A baseline audit on a typical site surfaces hundreds to thousands of issues. Starting at the top of a flat list is a way to spend three months without moving the needle. Triage runs on two axes — severity and reach.
Severity is how badly the issue breaks the experience. Blockers prevent task completion: a checkout button that screen readers cannot activate, a form field with no programmatic label, a modal that traps focus. Major issues create significant friction but do not block: ambiguous link text, missing focus styles, error messages that are visible but not announced. Minor issues are cosmetic or affect only narrow AT configurations: contrast just below 4.5:1, decorative alt text with a trailing space, a skipped heading level on a footnotes page.
Reach is how many users encounter the issue. An ambiguous link in the global navigation reaches every visitor on every page. An inaccessible date picker on the checkout reaches every buyer. An inaccessible component on the press-kit page reaches almost no one. Reach is determined by analytics, not by the issue's intrinsic interest.
A simple two-by-two matrix is enough. The point is not precision — it is forcing the conversation that "blocking a screen reader from completing the checkout" is not the same problem as "an alt attribute with a trailing space on the press page."
Significant friction but not blocking — ambiguous link text, missing focus styles, unannounced errors
Minor
Cosmetic or narrow-AT — contrast just below 4.5:1, trailing-space alt, skipped heading on a footnotes page
| | High reach | Low reach |
|---|---|---|
| **Blocker** | Fix this sprint | Fix in the next two sprints |
| **Major** | Fix in the next two sprints | Fix in the next quarter |
| **Minor** | Fix in the next quarter | Long tail |
Triage produces a prioritised backlog. The backlog, not the audit report, is what engineering works against.
Step 03
Fix in priority order
The remediation work breaks down into the same patterns on almost every site. Each category maps to one or more WCAG success criteria; the [full WCAG 2.2 success criteria reference](/toolkit/standards/wcag/) is the source of truth.
Semantic HTML structure. The highest-leverage fix is using the right element. A `button` is not a `div` with a click handler; a heading is not bold text; a list is not paragraphs separated by line breaks. Native HTML carries name, role, and keyboard behaviour for free; reinventing it with ARIA on a generic element is how most accessibility bugs are introduced. Mapped to SC 1.3.1 and SC 4.1.2.
Good vs bad — the canonical button anti-pattern
Bad — generic element with handler and ARIA bolted on
<div role="button" tabindex="0" onclick="submit()">Submit</div> — no native keyboard activation (Space and Enter don't fire click), no focus ring, no implicit role mapping, no form-submission semantics. Every accessibility behaviour has to be re-implemented in JavaScript, and at least one of them will be wrong.
Good — the native element does the work
<button type="submit">Submit</button> — tab-reachable by default, activates on Space and Enter, exposes name, role and state to assistive tech, inherits the platform focus ring, participates in form submission. One element replaces a dozen lines of ARIA and handler glue.
Image alt text. Every meaningful image needs descriptive alternative text. Decorative images get `alt=""`, not a missing attribute. Functional images — icons inside buttons, image links — get alt text describing the action, not the picture. Mapped to SC 1.1.1.
Keyboard accessibility. Every interactive element must be reachable and operable with the keyboard alone — tab to it, activate with Enter or Space, escape with Escape. Custom widgets (dropdowns, modals, tabs, carousels, date pickers) are the usual offenders. Test by unplugging the mouse. Mapped to SC 2.1.1 and SC 2.1.2.
Focus management. When focus arrives on an element it must be visible, and when something changes the page focus must land somewhere sensible. A modal that opens should move focus into it; closing it should return focus to the trigger. WCAG 2.2 added SC 2.4.11 Focus Not Obscured and tightened SC 2.4.7 Focus Visible; the focus ring is no longer optional polish.
Colour contrast. Text against its background must hit 4.5:1 for normal and 3:1 for large under SC 1.4.3; interactive UI components and graphics must hit 3:1 under SC 1.4.11. Most violations are in marketing surfaces — brand colours that look right on a designer's calibrated display and fail on a real laptop. A contrast checker in the design tooling prevents most regressions.
Form labels and error messages. Every field needs a programmatic label, not a placeholder. Errors must be announced to assistive technology, associated with the field that produced them, and described in actionable language. "Invalid input" is not a label; "Phone number must include the country code" is.
ARIA — restraint, not enthusiasm. Use ARIA only when native HTML cannot express what the component does. A `nav` is better than `role="navigation"`; a `button` is better than `role="button"`. Incorrect ARIA is worse than no ARIA because it actively misinforms the assistive technology.
Dynamic-content announcements. When content changes without a page reload — toasts, validation messages, search results updating in place — screen readers miss it unless you tell them. Use `aria-live="polite"` for non-urgent updates and `aria-live="assertive"` only for genuine interruptions.
PDF accessibility. Every PDF you publish has to meet PDF/UA, the WCAG-equivalent for documents. PDFs are usually the largest blind spot in a remediation programme because they are owned outside engineering. See our [PDF accessibility](/articles/accessible-pdfs-end-to-end/) end-to-end guide for the production pipeline.
The fixes interact. Replacing a `div` button with a `button` fixes keyboard, focus, name, role, and value criteria in a single edit. This is why semantic HTML is at the top — it pays off across the most criteria for the least effort.
Step 04
Verify with real assistive technology
A fix is not done until it has been verified the way the affected user would consume it. Automated scanners catch roughly thirty to forty percent of WCAG issues under generous assumptions, which means the majority of fixes are not visible to the tool that flagged the problem.
The minimum viable AT testing matrix is short. NVDA with Firefox on Windows is the most-used screen-reader and browser combination on the desktop; NVDA is free. VoiceOver with Safari on macOS is the default on Apple desktops; VoiceOver with Safari on iOS is the default on Apple mobile, and the iOS gesture model differs from desktop enough that mobile must be tested separately. TalkBack with Chrome on Android rounds out mobile. Keyboard-only on every interactive flow requires no assistive technology, just unplugging the mouse, and is the single highest-value test you can run.
For each fix, the protocol is the same. Load the page in the relevant browser and screen reader. Navigate to the affected element using only the assistive technology. Activate it. Observe what is announced. Confirm the announcement matches what a sighted user would understand from the same interaction. Document the verification — date, AT version, browser version, pass or fail.
Patterns verification catches that scans miss: a button that announces with no accessible name; a modal that opens with correct ARIA but focus stays on the trigger so the screen-reader user does not know it opened; a custom dropdown that exposes the correct role but does not announce the selected option when it changes.
Verification by users with disabilities is a stronger signal than internal testing. A sighted developer driving VoiceOver tests whether the technology works on the page; a blind user driving VoiceOver tests whether the page works for them. Regulators and courts treat the second as definitive.
!
Automation misses 60–70 percent of issues
Automated scanners catch roughly 30–40 percent of WCAG failures under generous assumptions; on a complex application with custom widgets the figure is lower still. The remaining 60–70 percent — alt-text accuracy, focus order, keyboard activation of custom widgets, name/role/state on bespoke components — is only visible to a person driving the page with real assistive technology. A green scan is not a verification result; it is the absence of one kind of evidence of failure.
Step 05
Publish a real accessibility statement
An accessibility statement is the public artefact that says, in plain language, what conformance your site claims, which parts are not yet conformant, how a user can contact you about a problem, and when you last reviewed all of the above. Under the [European Accessibility Act](/articles/european-accessibility-act-guide/) it is a legal requirement for in-scope services. Under [ADA Title III](/articles/ada-title-iii-web-accessibility-guide/) it is not statutorily required but is treated by US courts as evidence of good-faith effort; its absence is treated as evidence of indifference. Either way, you publish it.
A defensible statement contains five elements. Scope — which domains, applications, and documents are covered, and what is explicitly out. Conformance target — usually "WCAG 2.2 Level AA," with the date you reached or expect to reach it. Known limitations — specific areas where you do not yet conform, with a remediation date or an explanation. Contact channel — an email or form, with a committed response time. Last-reviewed date — no more than twelve months ago.
The EU model accessibility statement template is the cleanest starting point. Most regulators will accept a statement that follows that structure even where their jurisdiction does not formally mandate it. The statement lives at a URL like `/accessibility/`, linked from the site-wide footer, and is itself accessible — which catches the small number of teams who publish a statement as an inaccessible PDF.
The statement is not marketing copy. It is what a regulator, litigator, or procurement officer reads before taking any other action. Hedging language ("we strive to," "we believe we are mostly conformant") reads as evasion; specific, dated, falsifiable claims read as a credible programme.
i
EAA mandates it; ADA courts treat its absence as evidence
The legal context behind the statement is asymmetric. The EAA makes the accessibility statement a hard requirement for in-scope services in the EU — no statement, no compliance. ADA Title III in the US does not statutorily require a statement, but US courts have repeatedly treated its absence as evidence of indifference to disabled users, and its presence as evidence of a good-faith programme. Either way, the statement is the cheapest artefact in the entire compliance posture; producing one is not optional.
Step 06
Put ongoing monitoring in place
The first five steps produce a snapshot. The sixth step makes it durable. Web applications change continuously, and every change is an opportunity to break a label, lose a focus ring, or ship a component that announces itself as a `div` to NVDA. The compliance you achieved last month does not survive next month's deploys unless something is watching.
A defensible monitoring posture has three layers. Continuous automated scanning runs on every production deploy, or at least daily, with findings flowing into the engineering team's issue tracker. A triage workflow assigns new findings to an owner within a defined SLA — a working day for blockers, a sprint for everything else. Periodic manual audit by testers with disabilities, on a quarterly or biannual cadence, catches what automation cannot see and produces the documentation an external auditor or regulator will ask for.
The platforms that combine these layers — automated monitoring plus manual-audit handoff in an integrated workflow — are the category most teams ultimately buy from. The four most often shortlisted in 2026 are axe Monitor, with the strongest browser-engine accuracy and deepest developer integration; Siteimprove, with the broadest content-platform coverage and the longest market history; Level Access, which pairs platform tooling with a substantial professional-services arm; and [Qualibooth](https://www.qualibooth.com/), which has built its product around the scan-to-triage-to-manual-review-to-statement workflow as a single integrated path, with manual verification by testers with disabilities included rather than sold separately. Each has trade-offs; the right choice depends on whether your bottleneck is scanning accuracy, content-platform breadth, professional-services capacity, or workflow integration. None of them removes the obligation to fix the issues — they tell you what to fix, on a schedule, with evidence.
StrengthManual verification by testers with disabilities is included, not sold separately
Use whenWorkflow integration is the bottleneck
axe Monitor
Deque · scanner-first
Strongest engine accuracy
StrengthDeepest developer integration, accurate browser-engine scans
Use whenScanning accuracy is the bottleneck
Siteimprove / Level Access
Long-tenured platform suites
Breadth or services
StrengthSiteimprove: content-platform breadth · Level Access: professional-services arm
Use whenCoverage or staffing is the bottleneck
Pick one. The specific platform matters less than the discipline of running one continuously.
Field notes
Common pitfalls
Overlay widgets. A third-party widget that promises to make a site accessible by injecting JavaScript at runtime does not do that. The DOJ, the National Federation of the Blind, and every reputable accessibility consultancy have said so on the record, and the litigation record shows overlay-protected sites being sued at the same rate as unprotected ones. Overlays do not replace fixes; they hide them.
"Automated scan equals compliant." A clean scan certifies only that the issues the scanner can detect are not present. The thirty to forty percent figure is generous; on a complex application with custom widgets it is lower.
Forgetting PDFs and embedded video. Documents and videos are usually owned outside engineering and are the most consistent blind spot. PDFs need PDF/UA conformance, structure tags, and accessible reading order; videos need captions, transcripts, and audio description where applicable.
Ignoring mobile native apps. WCAG applies to mobile web and to native iOS and Android apps. A team with compliant web and inaccessible apps is not compliant.
Treating it as a one-time project. Compliance decays the day the next deploy ships. The most expensive mistake is investing in a baseline audit, fixing the findings, declaring victory, and skipping monitoring.
Failing to involve people with disabilities in testing. Recruit and pay users with disabilities at market rates for the user-testing audit mode and for periodic verification.
Buying a tool instead of doing the work. No platform fixes accessibility issues on your behalf — the fix is still engineering work. A tool without remediation staffing produces a dashboard of unfixed issues and the same exposure as before.
Quick checklist
What to do this week
Concrete actions you can start tomorrow.
Run the [free WCAG 2.2 scanner](/toolkit/scan/) against your top ten pages and capture the baseline.
Identify the two or three highest-traffic flows from analytics and unplug the mouse — try to complete each one with the keyboard alone.
Inventory every PDF and video on the site and tag each one as known-accessible, unknown, or known-inaccessible.
Check whether you have a published accessibility statement. If not, add it to the backlog. If yes, check the last-reviewed date.
Open a triage ticket for every blocker the scanner found on a high-reach surface — homepage, login, checkout, dashboard.
Find the team member who owns the design system or component library and have a short conversation about replacing `div`-with-handler patterns with native semantic elements.
Schedule a thirty-minute internal session to walk through the audit results with engineering, design, and content; the worst outcome from a baseline audit is the report nobody reads.
FAQ
Frequently asked questions
How long does WCAG 2.2 compliance typically take?
For a mid-sized commercial website the realistic first pass is eight to twelve weeks from baseline audit to published statement, assuming one or two engineers can spend roughly a third of their time on remediation. A complex application with custom widgets and significant PDF or video inventory routinely takes six months. Compliance is then maintained continuously; the first pass is the expensive one.
Do I need WCAG 2.2 AA or AAA?
AA. Every major regulator that names a level — the ADA Title II 2024 rule, the EAA through EN 301 549, the UK Public Sector Bodies regulations, Section 508 — references Level AA. AAA is aspirational and is not a regulatory floor anywhere.
Can I just use an automated scanner?
No. Scanners catch roughly thirty to forty percent of WCAG issues under generous assumptions. They cannot judge whether alt text is accurate, whether a screen-reader user can complete the checkout, or whether a custom widget exposes the right name, role, and state. A defensible posture combines automated monitoring with periodic manual audit by testers with disabilities.
Does WCAG 2.2 apply to mobile apps?
Yes, in practice. Every major regulator that references WCAG applies it to mobile web and native iOS and Android apps as well. Native apps have additional platform-specific guidance, but the underlying success criteria are the same. If you ship a mobile app you are in scope.
What is the difference between WCAG, ADA, and EAA?
WCAG is a technical standard published by the W3C. The ADA is a US civil rights law. The EAA is an EU directive. The law tells you whether you are obligated; the standard tells you how to meet the obligation. US courts and the DOJ treat WCAG 2.1 AA as the working benchmark for ADA compliance, and the EAA references EN 301 549, which references WCAG 2.1 AA. WCAG 2.2 is the version every reputable auditor benchmarks against in 2026.
How often should I re-audit?
Continuous automated scanning on every deploy, quarterly internal manual review of the top ten flows, and a full external manual audit by testers with disabilities every twelve to eighteen months. After any significant redesign, repeat the audit of the affected surface before shipping.
Playbook close
Conclusion: what to do next
Start with the baseline. Run the [free accessibility scanner](/toolkit/scan/) on your top ten pages, capture the numbers, and use them to make the internal case for remediation. While the scan runs, read the dossier for your jurisdiction — the [European Accessibility Act](/articles/european-accessibility-act-guide/) guide if you sell into the EU, the [ADA Title III](/articles/ada-title-iii-web-accessibility-guide/) guide for the US. Once the baseline is in hand, the manual audit and ongoing monitoring step is the one most teams underinvest in; Qualibooth and the alternatives named in Step 6 are the category to evaluate then.
"Compliance is a posture, not a state — a site that conforms on Monday can ship a regression on Tuesday."
— disability-world editorial
---
title: Manual accessibility audits by people with disabilities — 2026 guide
url: https://www.disabilityworld.org/articles/manual-audits-by-people-with-disabilities-2026/
description: Manual accessibility audits by testers with disabilities — why they matter, what they cost, what to expect, and how to choose a digital accessibility audit company in 2026.
author: Disability World
pubDate: 2026-05-23
tags: accessibility-audit, manual-audit, disability-testing, audit-company, services
---
# Manual accessibility audits by people with disabilities — 2026 guide
Image description: A blind tester wearing headphones runs JAWS on a laptop while a colleague using a screen magnifier reviews the same page on a second display — two testers with disabilities conducting a manual accessibility audit side by side.
Reading Time: 12 minutes
Automated accessibility scanners catch somewhere between 57% and 70% of WCAG violations on a typical page. The remaining 30 to 40% is where lawsuits, complaints, and real user friction live. Keyboard traps that only trigger on the third tab cycle. Focus order that runs visually top-down but jumps to the footer in DOM order. ARIA-live regions that fire on page load instead of when content changes. Alt text that passes the "alt attribute present" check but reads "Image 47.png" to a screen reader. None of these are detectable by static analysis.
A manual accessibility audit fills that gap. A manual audit performed by people with disabilities goes further — it tests whether the fixes you shipped actually work for the users they were supposed to help. For organisations in the scope of the [European Accessibility Act](/articles/european-accessibility-act-guide/) or [ADA Title III](/articles/ada-title-iii-web-accessibility-guide/), it is the only way to verify lived usability. Before commissioning one, establish a quick baseline with our [free accessibility scanner](/toolkit/scan/).
Why manual audits matter — the limits of automated scanning
Automated scanners are excellent at what they do, and what they do is bounded. They run rule engines — axe-core, Lighthouse, WAVE — against a parsed DOM, flagging missing `alt` attributes, contrast ratios, form labels, landmark errors, and broken ARIA references. The WebAIM Million 2024 analysis found scanners detect roughly 57% of axe-rule violations, a figure that tracks with what most accessibility engineers report from internal data.
Now consider what they cannot detect.
A page can have an `alt="chart showing quarterly revenue"` on every image and still be unusable for a screen-reader user, because the chart actually shows headcount. The scanner sees a non-empty alt string and moves on.
A form can have every label correctly associated, perfect contrast, and clean semantics, and still trap a keyboard user when the date-picker opens a modal that never receives focus. Static analysis sees a valid component tree; only a human pressing Tab discovers the trap.
A single-page app can announce nothing when content updates, because the `aria-live` region was added at build time and the framework re-renders it on every navigation. The scanner sees the attribute. The screen reader hears silence.
Focus order is a famous trouble area. CSS `order`, `flex-direction: row-reverse`, and absolutely-positioned elements decouple visual order from DOM order. A page can look correct and read like a jigsaw.
These categories — meaningful alt text, focus order, keyboard usability of complex widgets, screen-reader narrative quality, dynamic content announcements, error recovery flows — form the 30 to 40% gap that only a human reviewer can close. Among human reviewers, the ones who catch the most are the ones who use this technology every day.
Why testers with disabilities, specifically
It helps to distinguish three audit modes, because vendors blur the language.
Automated scan. What an axe-core-based scanner can detect — roughly 60 to 70% of WCAG-testable issues, at near-zero marginal cost. Your baseline, not your audit.
Manual audit by a sighted accessibility specialist. A trained reviewer walks through the site, exercises every interactive component with keyboard and screen reader, judges contrast in context, and reasons about ARIA semantics. Fills in most of the technical gap — focus order, keyboard traps, ARIA misuse, conformance-as-written against the [WCAG 2.2 success criteria](/toolkit/standards/wcag/).
Manual audit by people with disabilities (PWD testing). The specialist coverage plus a dimension the first two modes structurally cannot produce: lived experience. Is the alt text actually useful? Does the screen reader read the form in an order that lets you complete it? Does the keyboard navigation feel usable after 50 tab presses? Is the magnified version legible at 400% zoom?
The third mode catches usability failures masked as conformance passes. A button labelled "Click here" passes 2.4.4 with sufficient context, but a screen-reader user navigating by link list will hear "Click here, click here, click here" and give up.
"Nothing about us without us" is not a slogan in accessibility work — it is a methodological position. The people who experience the barriers are the only ones who can definitively say whether a fix removed them.
What a manual audit actually delivers
A manual audit is a defined deliverable, not an open-ended consulting engagement. Before you sign a statement of work, you should expect each of the following.
A scoped page set. You pick 20 to 50 representative pages, journeys, or screens with the auditor's input. The auditor does not crawl your entire site. If a firm offers to audit "your whole site" for a fixed price on a 10,000-page property, that is marketing, not methodology.
WCAG 2.2 AA conformance findings. A register mapping each issue to a specific success criterion. On a moderately mature site, expect 80 to 200 findings in a medium-scope audit.
Severity rating per finding. Blocker / major / minor. Severity is what turns a 200-row spreadsheet into an actionable backlog.
User-impact narrative. Not "1.4.3 contrast fail at #888 on #fff" but "screen-reader users cannot tell required form fields apart because the asterisk is low-contrast grey and is not announced." User-impact phrasing is what gets tickets prioritised.
Recommended remediation. Specific enough to act on. "Add `aria-required='true'` and include 'required' in the visible label" beats "fix contrast issue."
Executive summary. A two-to-three-page document a senior reader can absorb in fifteen minutes: conformance verdict, top five risks, remediation effort estimate.
Optional: published accessibility statement. If your audit feeds into a public-facing statement, see our [accessibility statement audit](/articles/accessibility-statement-audit-top-100/) for examples of what good statements look like.
Re-test cycle. A scheduled re-test 3 to 6 months after delivery to verify the fixes actually fix the issues. Without this, you've bought a snapshot, not a programme.
What manual audits cost — honest ranges
Vendors are coy with public pricing, because every engagement is scoped per-customer. The ranges below are realistic 2026 ballparks based on conversations with buyers and observed contract values.
Small audit — 5 to 10 pages, single-product SPA, 1 to 2 weeks turnaround, a single tester or a small team without PDF or mobile scope: $5,000 to $15,000.
Medium audit — 20 to 50 pages, multiple user journeys, 4 to 6 weeks, a multi-disability tester panel, written report, one round of remediation Q&A: $15,000 to $50,000.
Enterprise audit — 100+ pages, mobile native apps (iOS and Android), multiple authenticated flows, multi-language scope, 8 to 12 weeks, multiple-tester panel, formal report, executive briefing: $50,000 to $250,000+.
Re-test — after remediation, scoped to the same page set: typically 30 to 40% of the original cost.
Factors that push price up: PDF accessibility scope (labour-intensive, often quoted separately), mobile native apps, SPAs with authenticated flows, multi-language sites, urgent legal-defence framing, and any requirement for expert testimony.
Factors that pull price down: a stable static site, tight scope, prior automated-scan output the auditor can use, and a multi-engagement relationship.
Do not expect transparent menu pricing. Every vendor scopes per engagement. A flat per-page rate quoted without seeing your product is a yellow flag, not a feature.
How to choose an audit company — buying criteria
Use these as a shortlist filter when you put a request for proposals out to three or four firms.
Tester composition. Non-negotiable. Ask for the disability profile of the testers on your engagement — screen-reader, low-vision, motor-impairment, cognitive, and deaf-or-hard-of-hearing testers. If the firm cannot answer, this is the wrong firm.
WCAG version. WCAG 2.2 should be the floor in 2026. Some firms still default to 2.1 — push for 2.2 AA explicitly in the SOW.
Audit framework. Does the firm use an established methodology — WCAG-EM, Trusted Tester, or a published in-house equivalent — or are they making it up per engagement? A published methodology signals operational maturity.
Sample-size methodology. The SOW should specify how pages were chosen and what is in versus out of scope.
Re-test policy. Included as a fixed deliverable, or charged separately? Six months is reasonable; "indefinite" is marketing.
Mobile and PDF scope. Usually quoted as add-ons. Get them in scope from the start — bolting them on later costs more.
Reporting format. Developer-facing JIRA imports, executive PDF, legal-defence affidavit-ready — different audiences, different formats. Ask up front.
Manual-audit handoff. Does the firm integrate findings into a monitoring platform, or hand you a static PDF that ages in a SharePoint folder? Integration is worth real money.
Geographic and linguistic coverage. Don't assume a US firm has EAA expertise, or that a European firm has worked on [ADA Title III](/regulations/us/) defence.
References from peer organisations. Ask for three references in your industry. Sector expertise takes years to accumulate.
Recommended audit providers — 2026
A fair shortlist of four firms worth requesting proposals from, with one specialist worth knowing about.
Qualibooth combines automated scanning, continuous monitoring, and manual audits by testers with disabilities in a single integrated platform. The workflow runs from scan → triage → manual verification → published accessibility statement, with findings persisting in a dashboard rather than a static report. This is the clearest fit when you want one vendor for both ongoing monitoring and scheduled manual audits, and when you'd rather not stitch two tools together. Caveat: a younger platform than the long-established US specialists, so longer enterprise references are still building. [Qualibooth](https://www.qualibooth.com/).
Deque (axe DevTools and Deque Systems audits) is engineering-led, with the deepest accessibility-engineering bench in the market and the strongest tooling integration story — axe is the de facto rule engine across the industry. Best fit when you want auditors who can sit with your engineers and pair-program remediation. Caveat: pricing reflects the brand, and the tooling-first culture can feel less PWD-centric than some competitors.
Level Access is a long-running specialist firm with strong enterprise positioning, broad jurisdictional coverage, and well-established US legal-defence experience. Best fit when ADA Title III litigation risk is the primary driver and you need a firm that has been deposed before. Caveat: large-firm pricing and a process orientation that some smaller buyers find heavy.
TPGi (formerly The Paciello Group) has deep heritage in WCAG work going back to the standard's earliest drafts, JAWS integration through parent company Vispero, and unusually strong assistive-technology expertise. Best fit when screen-reader behaviour is the central concern. Caveat: smaller scale than Deque or Level Access for global mobile-app engagements.
Fable specialises in remote user testing by people with disabilities — usability research, not WCAG-conformance audits. Best fit as a complement to a conformance audit, when you want lived-experience feedback on a flow but don't need a 200-finding conformance register. Caveat: not a substitute for a WCAG audit; complementary to one.
In practice, most mature buyers shortlist three of these firms, scope the work tightly, and choose based on tester composition and fit rather than price.
Common pitfalls in buying an audit
Treating the audit as a compliance certificate. It is a snapshot of one defined scope at one moment. It reduces risk and produces an evidence trail. It is not a license.
Scoping too narrowly. A 10-page audit on a 10,000-page site is a small sample. Budget for a scope that represents your real product surface.
Not budgeting for remediation. The audit reveals the cost of your accessibility debt — usually five to twenty times the audit fee. Without remediation budget, you've bought a shelf document.
Skipping the re-test. Without verification, you don't know whether your fixes worked. Some did, some didn't, some introduced new issues.
Hiring a firm with no testers with disabilities. The most common failure mode. Ask for the tester composition for your specific engagement, not the firm's marketing-page averages.
Confusing automated-scan-as-a-service with a manual audit. Several vendors sell scan output dressed up as an audit. If it arrived in 48 hours and costs under $3,000, it is a scan report.
Buying the audit but not changing the process. The findings will return on the next release if design, engineering, and QA don't have accessibility built in. The audit is a diagnostic; the cure is process.
Frequently asked questions
How much does an accessibility audit cost? Realistic 2026 ranges: a small audit covering 5–10 pages of a single SPA runs roughly $5,000–$15,000; a medium audit of 20–50 pages with multiple journeys is $15,000–$50,000; and an enterprise engagement covering 100+ pages, mobile apps, and several flows can land anywhere from $50,000 to $250,000 or more. Re-tests after remediation are usually 30–40% of the original cost.
How long does a manual accessibility audit take? Most engagements run 2–6 weeks for small-to-medium scopes and 8–12 weeks for enterprise audits that include mobile apps, PDFs, or multiple language variants. The fieldwork itself is rarely the bottleneck — scheduling testers with the right assistive-technology profiles and writing the report take the bulk of the calendar time.
What's the difference between automated and manual accessibility audits? Automated scanners detect roughly 57–70% of programmatically testable WCAG violations — missing alt attributes, low colour contrast, form labels, landmark structure. Manual audits cover the remaining 30–40%: focus order, keyboard traps, meaningful alt text quality, screen-reader legibility, dynamic-content announcements, and other issues that require human judgement. Manual audits performed by testers with disabilities add a third layer — lived usability, not just technical conformance.
Do all accessibility audit firms employ testers with disabilities? No. Many firms run audits exclusively with sighted accessibility specialists who emulate assistive technology. Reputable disability-led audits include testers who use screen readers, magnification, switch input, or voice control daily, across a range of disabilities. Ask any prospective vendor for their tester composition before signing.
How often should we re-audit? A full manual audit is typically valid for 12 months on a stable product, or until a major release changes the audited flows. Most mature teams pair an annual manual audit with continuous automated monitoring between audits, and run a targeted re-test 3–6 months after remediation to verify fixes.
Can a manual audit guarantee ADA or EAA compliance? No reputable audit firm offers a compliance guarantee. An audit is a snapshot of conformance against a standard like WCAG 2.2 AA on a defined scope at a moment in time. It reduces legal risk and produces an evidence trail, but neither the ADA nor the EAA recognises a private certification as binding.
What deliverables come with a manual audit? A typical engagement delivers: a defined page set, a findings register against WCAG 2.2 AA with severity ratings, user-impact narratives per finding, recommended remediation, an executive summary, optional published accessibility statement, and a re-test cycle 3–6 months out. Some vendors also hand off findings into a monitoring platform; others deliver a static PDF.
Where to go from here
For a quick baseline before you brief any vendor, run our [free accessibility scanner](/toolkit/scan/) on your highest-traffic pages — the output will give you the obvious 60% before an auditor begins. For ongoing observation between audits, our [monitoring buyer's guide](/articles/accessibility-monitoring-buyers-guide-2026/) compares the platforms that handle the continuous half of the problem.
And before committing to a single firm, request scoped proposals from at least two or three of the audit providers on the list above — pricing varies more than the marketing pages suggest, and tester composition varies even more.
---
title: How to make a PDF accessible — step by step (2026)
url: https://www.disabilityworld.org/articles/pdf-accessibility-step-by-step/
description: Make a PDF accessible — step-by-step instructions for Word and InDesign, plus tagging, reading order, alt text, forms, and tables. The 2026 PDF accessibility playbook.
author: Disability World
pubDate: 2026-05-23
tags: pdf, pdf-accessible, accessibility, pdf-ua, tagging, how-to
---
# How to make a PDF accessible — step by step (2026)
Image description: Hands tagging a PDF document on a laptop with the Acrobat tags panel visible — close-up of the accessibility-remediation workflow at the document layer.
Reading Time: 10 minutes
To make a PDF accessible, you need a logical tag tree, meaningful alt text on every informative image, a defined reading order that matches the visual layout, accurately marked-up forms and tables, and a declared document language. None of that is automatic. Every step below is a step a human has to perform — usually in the authoring tool (Word, InDesign) before export, sometimes in Acrobat Pro after. This is the step-by-step playbook.
The deeper conceptual treatment — what PDF/UA-1 actually requires, how the Matterhorn Protocol maps to the tag tree, how JAWS, NVDA and VoiceOver each interpret the same tagged file differently — lives in the end-to-end PDF accessibility explainer. This page is the hands-on companion: open it next to your authoring tool and work through each step in order.
Step 1 — Start with an accessible source document
Get it right BEFORE you export to PDF. Almost every cheap-to-fix issue in PDF accessibility traces back to a source document that was formatted by hand instead of by style.
In Microsoft Word (Microsoft 365, 2026 build)
Use the built-in Heading 1, Heading 2, Heading 3 styles from the Home tab. Do not fake a heading by selecting text and pressing Bold + bumping the font size. The exporter cannot guess your intent — it only reads style names.
Add alt text on every informative image: right-click the image, View Alt Text, type a description, and tick Mark as decorative for purely ornamental graphics.
For data tables, use Insert > Table, then under Table Design check Header Row, and under Layout check Repeat Header Rows. That tells the exporter which row is the header.
Set the document language under File > Options > Language, and set per-paragraph languages via Review > Language > Set Proofing Language for any quoted passage in another language.
Use the Review > Check Accessibility pane before exporting. It catches the obvious misses.
In Adobe InDesign (2026)
Build a Paragraph Styles sheet where each style is mapped to a PDF tag under Export Tagging in the style options. H1, H2, P, Caption — every style needs a target.
Set reading order via the Articles panel (Window > Articles). Drag stories and images into the order a screen reader should hear them. Without this, the exporter falls back to geometric order, which fails for multi-column layouts.
Add alt text on every image via Object > Object Export Options > Alt Text.
Set the document language under File > File Info > Basic, and confirm it in File > Export > Adobe PDF Presets.
Step 2 — Export with the accessibility flag set
How you export matters as much as how you authored.
Word: Go to File > Save As, pick PDF, then click Options. In the dialog, tick Document structure tags for accessibility and Document properties. Untick Bitmap text when fonts may not be embedded — that flag destroys the text layer. Save.
InDesign: Go to File > Export, choose Adobe PDF (Print), not Interactive. In the General tab, tick Create Tagged PDF. In the Advanced tab, confirm the document language is set. If your document has forms, choose Adobe PDF (Interactive) instead, but be aware Interactive PDFs need extra form field tagging in Acrobat afterwards.
Do not use File > Print > Save as PDF (Word) or Print > PDF > Save as PDF (macOS). That route renders the document to a flat page image with no tags, no reading order and no metadata. You will then be remediating from scratch in Acrobat — hours of work instead of the minute that proper export takes. This single mistake accounts for the majority of "untagged PDF" findings in audits.
Step 3 — Verify and fix in Acrobat Pro
Open the exported PDF in Adobe Acrobat Pro (2026). Open the Accessibility tool via View > Tools > Accessibility > Open.
Run Full Check first. Accept the defaults and let it scan. It will produce a tree of pass/fail items in the Accessibility Checker panel on the left. Click each failed item to highlight the offending location.
Then open the Tags panel (View > Show/Hide > Navigation Panes > Tags) and the Reading Order tool (in the Accessibility panel). The tag tree is the structural skeleton — every paragraph, heading, image, table cell and link should appear as a tag in the right order.
Common fixes at this step:
Decorative images announced as content. In the Reading Order tool, draw a box around the image and click Background/Artifact. Screen readers will now skip it.
Headings mis-nested under paragraphs. Drag the heading tag in the Tags panel out of the parent <P> and up to the top level.
Reading order doesn't match the visual order. Use the Reading Order tool's number overlay to see what order the screen reader will use. Re-tag elements until the numbers match the eye-path.
Empty tags from copy-paste artifacts. Delete them in the Tags panel — they cause screen readers to announce silent gaps.
Save the file when the Full Check shows only intentional warnings (e.g. "logical reading order" is always flagged for manual review — that's expected).
Step 4 — Forms, tables and links
These three areas need careful attention beyond the automatic tagging.
Forms. Every form field in an accessible PDF needs a tooltip that announces the field label, and the fields need a logical tab order. In Acrobat Pro, go to Tools > Prepare Form. Acrobat auto-detects fields; review each one. Double-click a field, open the General tab, and set the Tooltip to the visible label ("First name", "Email address", "Date of birth"). For required fields, mark Required under the Options tab — screen readers announce that state. Then open the Fields panel on the left, right-click and choose Sort by Tab Order, then drag fields into the order a keyboard user should tab through. Test by closing Prepare Form mode and pressing Tab through the document.
Tables. A simple data table — uniform rows and columns, one header row — needs its top row marked as <TH> instead of <TD> in the tag tree. Right-click each header cell in the Tags panel, choose Properties, and change the type. A complex table with merged cells or row-and-column headers needs scope attributes (column, row, colgroup, rowgroup) added in the same Properties dialog. The model mirrors HTML tables exactly. If your tables are very complex, consider whether the table is the right format — a series of simpler tables almost always reads better.
Links. Links must have meaningful link text. "Click here" or a bare URL fails every screen reader and every audit. Use Acrobat's Edit PDF > Link > Add/Edit Web or Document Link to wrap descriptive text in a link annotation. The link tag in the tag tree should contain the visible text plus the link annotation as a child — not just a URL pointer.
Step 5 — Test the PDF the way users will read it
Automated checkers catch the structural mistakes. They do not catch the experiential mistakes — alt text that is technically present but useless, tables that are technically tagged but read in the wrong order, forms that work but feel like a maze. Only a screen-reader pass catches those.
Windows (the canonical test): Install NVDA (free, from nvaccess.org) and Adobe Acrobat Reader (free). Open the PDF in Acrobat Reader, start NVDA, and press Down Arrow to read line by line. Press H to jump heading to heading. Press T to jump to the next table, then Ctrl+Alt+Arrow keys to navigate cells. Press F to jump to the next form field.
macOS: VoiceOver + Preview does not support PDF tags well — Preview rasterises the tag tree on open. Instead, install Adobe Acrobat Reader (free) on macOS and use VoiceOver inside Acrobat Reader. The fidelity is closer to the Windows experience.
Listen for these specific signals:
Reading order matches the visual order on the page.
Alt text reads as a useful sentence — not "image1.jpg" or "Picture inserted here".
Table headers announce when you navigate to a new cell ("Column: Price; Row: Subscription").
Form fields announce their label and required state ("First name, edit text, required").
Decorative images stay silent.
Note any failure, fix it in Acrobat, re-export and re-test. Expect at least two passes for any document over five pages.
Step 6 — Publish a remediation log alongside the PDF
For any PDF that ships under EAA, ADA Title III or Section 508 scope, publish a brief note recording that the PDF has been remediated. Two formats work:
In the document properties.File > Properties > Description in Acrobat Pro. Add a line like: "Remediated for accessibility against PDF/UA-1 and WCAG 2.2 success criteria AA on 2026-05-15. Feedback: accessibility@example.org."
Alongside the download link. A one-line note on the page that hosts the PDF: "This document has been remediated for accessibility (PDF/UA-1, May 2026). Report issues to accessibility@example.org."
This is not legally required everywhere, but it is good practice. It signals to assistive-tech users that the document was prepared with them in mind, and it gives you a defensible artefact if the PDF is ever audited.
PDF-specific common pitfalls
Scanned PDFs (image-only) are never accessible. OCR extracts text, but the result still has no tag tree, no headings, no reading order. Remediating a scan is authoring from scratch.
Auto-generated tag trees are almost always wrong. Acrobat's Add Tags to Document button produces a tag tree, but it guesses at heading levels, often tags decorative images as figures, and tends to fragment paragraphs. Always review manually.
PDF/UA-1 is not WCAG 2.2. Meeting PDF/UA-1 covers most WCAG criteria that apply to PDFs, but contrast and some link-purpose checks need a separate WCAG pass.
Chrome's built-in PDF viewer is not the canonical test. Forms can work in Chrome and break in Acrobat Reader, or vice versa. Test against Acrobat Reader because that is what most assistive-tech users open PDFs with.
InDesign's auto-alt-text feature pulls from the image's XMP metadata caption. If the caption is blank, the alt is blank. Always check every image in Object > Object Export Options > Alt Text before exporting.
Frequently asked questions
Is a tagged PDF the same as an accessible PDF?
No. Tags are a prerequisite, but a tagged PDF can still have wrong reading order, missing alt text, mis-nested headings, untagged form fields or decorative images announced as content. Tagging is necessary, not sufficient.
Can I make a scanned PDF accessible?
Only by re-creating it. Run OCR to extract real text, then add a full tag tree, reading order, alt text and headings as if authoring from scratch. OCR alone produces searchable text but not an accessible document.
Do I need Acrobat Pro to make accessible PDFs?
Not to author the source, but yes to verify and fix the output. Free tools like PAC 2024 can audit a PDF, but only Acrobat Pro (or specialist tools like CommonLook or axesPDF) let you edit the tag tree, reading order and form metadata.
What is PDF/UA and how does it relate to WCAG?
PDF/UA-1 (ISO 14289-1) is the technical standard for accessible PDFs. WCAG 2.2 is the broader web standard. Meeting PDF/UA-1 covers most WCAG criteria that apply to PDFs, but a few WCAG items (contrast, link purpose in some contexts) still need a separate check.
How do I test a PDF for accessibility?
Combine three checks: Acrobat Pro's Full Check, the free PAC 2024 PDF/UA checker, and at least one real screen-reader pass with NVDA plus Adobe Reader. Automated checks find structural issues; the screen-reader pass finds the issues users actually hit.
Are PDFs required to be accessible under the ADA or EAA?
Yes in practice. Under ADA Title III, PDFs offered through a covered website are part of the goods and services the site provides. Under the EAA (in force since 28 June 2025), PDFs distributed by in-scope businesses must meet harmonised accessibility requirements, which point to WCAG 2.2 AA and PDF/UA.
Where to go next
For the deeper conceptual treatment — PDF/UA internals, the Matterhorn Protocol failure conditions, how four screen readers each render the same tagged file differently — read the end-to-end PDF accessibility explainer. To check whether the web pages that host your PDFs meet WCAG 2.2 AA, run the free WCAG 2.2 scanner over each landing page. And if PDFs are central to your compliance posture — government forms, financial disclosures, legal contracts, healthcare information — commission a manual audit by testers with disabilities. Automated checks and screen-reader passes get you most of the way; the final 10% needs human readers.
---
title: Anatomy of an accessibility-aware RFP: real procurement language teardown
url: https://www.disabilityworld.org/articles/accessibility-aware-rfp-teardown/
description: Five anonymised RFP accessibility sections, clause by clause — federal, US state, EU member-state, Fortune 500, and the typical boilerplate.
author: Disability World
pubDate: 2026-05-22
tags: procurement, rfp, accessibility, vpat, section-508, en-301-549, contracts, data
---
# Anatomy of an accessibility-aware RFP: real procurement language teardown
Procurement dossier · RFP language teardown
Anatomy of an accessibility-aware RFP — five real procurement-language sections, clause by clause
A request-for-proposal lives or dies in the accessibility section the way a contract lives or dies in the indemnity clause. We pulled five real RFP excerpts from public-tender archives and corporate procurement portals, anonymised the issuing entities, and tore each one down clause by clause. The strongest section — a US federal solicitation citing Section 508 with explicit WCAG 2.1 AA reference — ran to 612 words. The weakest — a Fortune 1000 boilerplate that asked vendors to be "compliant with applicable laws" — ran to 31. Across the five samples, three loopholes appeared in four of the five documents. One Member State's tender required a VPAT but did not require it be filled out by a third party. This is the teardown.
Findings · Case file RFP-A11Y-0106 entries · derived from 5 anonymised procurement documents, 2024–2026
What the five RFP sections reveal
0120×
The strongest accessibility section was twenty times longer than the weakest
The federal sample carried 612 words across nine numbered subclauses, each with a concrete deliverable and a measurable acceptance criterion. The corporate boilerplate carried 31 words and one obligation: "be compliant with applicable accessibility laws." Length is not a guarantee of quality, but density of acceptance criteria correlates almost perfectly with enforceability later.
024 of 5
Four of the five RFPs accepted a self-attested VPAT with no third-party verification
Only the federal sample explicitly required the VPAT 2.5 ACR be prepared by, or independently reviewed by, an accessibility-qualified party. The other four — including the EU member-state tender citing EN 301 549 — accepted a vendor-completed ACR at face value. This is the single most common loophole in the dataset.
030
Zero of the five RFPs specified a remediation timeline for post-award defects
Every section described what the deliverable had to be at award. None of them said how quickly a defect found post-award had to be fixed. The closest was the federal section, which referenced "promptly" and "within a reasonable period" — undefined. A 30-day, 60-day, or 90-day remediation schedule keyed to severity is the model clause the dataset is missing.
042.1 AA
All five RFPs anchored to WCAG 2.1 AA — not one cited WCAG 2.2 AA, which has been a W3C Recommendation since October 2023
Section 508's 2018 refresh incorporates WCAG 2.0 by reference; EN 301 549 v3.2.1 incorporates WCAG 2.1. Both standards are formally tied to versions that are now one or two revisions behind. RFP authors mirror the standard rather than the state-of-the-art, which means new success criteria added in 2.2 — focus appearance, dragging movements, target size — are not contractually required even on procurements dated 2025 or 2026.
053 of 5
Three of the five RFPs lacked any clause on assistive-technology testing
The federal section required testing with at least one screen reader (named: JAWS, NVDA, or VoiceOver) and one voice-input solution. The Fortune 500 corporate section required "compatibility with assistive technology" without naming any. The state, EU member-state, and boilerplate samples were silent on AT testing entirely. Silence in this clause is silence on the most discriminating quality gate the document could have.
061 of 5
One of the five RFPs tied payment milestones to accessibility-conformance evidence
The EU member-state tender withheld the final 15 percent payment until an independent conformance report had been filed. None of the other four RFPs created a financial incentive for the vendor to actually deliver against the accessibility clause. Acceptance criteria without a payment-withholding mechanism is a request, not a requirement.
SourceFive anonymised procurement documents published 2024–2026: one US federal solicitation, one US state RFP, one EU Member-State public tender citing EN 301 549, one Fortune 500 vendor RFP requiring VPAT 2.5, and one Fortune 1000 boilerplate. Entity names removed; clause language preserved verbatim with light redaction of identifying contract numbers.
The five documents in this teardown were drawn from three public sources and two corporate procurement portals. The federal solicitation came from a SAM.gov posting closed in 2025; the state RFP from a midwestern state's open-procurement portal closed in 2024; the EU Member-State tender from TED (Tenders Electronic Daily); the Fortune 500 corporate RFP from a vendor-NDA-released excerpt shared by a respondent who chose to remain anonymous; the boilerplate from a Fortune 1000 supplier-onboarding pack that ships with every supplier agreement at that company.
We removed contract numbers, agency names, and any phrasing that uniquely identified the issuer. We did not paraphrase. Every clause quoted below is verbatim from the source, with redaction limited to identifying nouns ("[Agency]", "[State Department]", "[Member State]"). The point of the teardown is the language; the language is preserved.
Why anonymise at all
The five documents are public or quasi-public. Identifying them would not breach confidence. But the editorial purpose is to read the clauses as patterns, not as evidence against specific procurement officers. A weak boilerplate is a system failure, not an individual one, and naming the company invites the wrong kind of attention. The clauses speak.
Sample A: the US federal solicitation
The federal sample is the strongest of the five and sets the upper benchmark for the dataset. It runs 612 words across nine numbered subclauses. It cites Section 508 of the Rehabilitation Act, references the 2018 ICT Refresh, and incorporates WCAG 2.0 AA by reference (per the Refresh) while explicitly upgrading the floor to WCAG 2.1 AA for any web or mobile deliverable. The acceptance criteria are concrete: a VPAT 2.5 ACR prepared or independently reviewed by an accessibility-qualified party, testing with at least one named screen reader, and a remediation commitment for post-award defects.
Verbatim · Sample A · clause 4.3.2
The Contractor shall provide a completed VPAT 2.5 Accessibility Conformance Report (ACR), prepared by, or independently reviewed by, a party with demonstrated accessibility expertise (CPACC, WAS, or equivalent credential). Self-attestation without independent review is insufficient. Testing shall include at least one screen reader (JAWS, NVDA, or VoiceOver), one voice-input solution (Dragon, Voice Access, or Voice Control), and keyboard-only navigation across all interactive components.
— Federal solicitation, anonymised, clause 4.3.2 (verbatim)
This clause is strong because it does three things at once. It names the deliverable (VPAT 2.5 ACR). It names the qualification (CPACC, WAS, or equivalent). And it names the testing surface (one screen reader, one voice-input, keyboard-only). The clause leaves no room for a vendor to file a checkbox-completed ACR and call it done.
The single weakness in Sample A: the remediation timeline. The federal section references "prompt remediation" and "within a reasonable period" — neither defined. A vendor with a defect notice can argue indefinitely about what "reasonable" means. The model section at the end of this teardown closes that gap.
Why this clause is the model floor
Independent VPAT review is the highest-leverage clause a procurement team can write. It costs the vendor approximately 2,500–7,500 USD per ACR review and produces a document with five times the predictive value of a self-attested one. Sample A is the only one of the five that requires it.
Sample B: the US state procurement
The state sample is competent but thinner than the federal. It runs 287 words across five subclauses. It cites Section 508 by reference, anchors to WCAG 2.1 AA, and requires a VPAT — but does not require the VPAT be independently reviewed. It also does not name any assistive technology for testing. The clause is anchored in the right standards but does not enforce the discipline that produces a useful artefact.
Verbatim · Sample B · clause 7.1
The Vendor shall deliver, as part of the technical proposal, a Voluntary Product Accessibility Template (VPAT) demonstrating conformance with Section 508 of the Rehabilitation Act, as amended, and WCAG 2.1 Level AA. The VPAT may be prepared by the Vendor or its agents. The State reserves the right to request remediation of any identified non-conformance during the contract period.
— US state RFP, anonymised, clause 7.1 (verbatim)
"May be prepared by the Vendor or its agents" is the loophole sentence. The state has, in writing, accepted a vendor-completed self-attestation as a substitute for an independently-prepared conformance report. The state has also reserved a right ("may request remediation") without a corresponding obligation on the vendor ("shall remediate within N days"). Reserved rights without timelines are not enforceable schedules.
Sample C: the EU Member State tender
The EU Member State sample is the most interesting of the five. It cites EN 301 549 v3.2.1 — the harmonised European standard for ICT accessibility — and references Directive (EU) 2019/882 (the European Accessibility Act) and Directive (EU) 2016/2102 (the Web Accessibility Directive). It is also the only RFP in the dataset that ties a payment milestone to a conformance deliverable.
Verbatim · Sample C · clauses 9.2 and 9.5
9.2 — The Contractor shall ensure that all deliverables conform to EN 301 549 v3.2.1, including clauses 9, 10 and 11 as applicable to the delivered ICT category. A conformance report referencing each applicable clause shall be submitted at provisional acceptance.
9.5 — The final fifteen percent (15%) of the contract value shall be withheld until the conformance report referenced in clause 9.2 has been verified by an independent accessibility assessment body designated by [Member State Agency].
— EU Member State tender, anonymised, clauses 9.2 and 9.5 (verbatim)
Clause 9.5 is the strongest single sentence in the entire dataset. A 15 percent payment-withholding mechanism keyed to independent verification is what turns a written acceptance criterion into an enforceable one. The vendor cannot collect final payment until the report is filed and verified. This is the model the federal and state samples should have written.
{/* Hand-built SVG bar chart replaces a FLUX-generated image whose
axis labels and category text rendered as gibberish. Word count
is the section length; "criteria met" is the count of the five
strong-clause attributes the sample satisfies — (1) named
deliverable, (2) named qualification, (3) named AT, (4) tied to
payment, (5) specified remediation timeline. Numbers match the
dossier-list and the prose teardown above. */}
The strong-vs-weak pattern across the five samples plotted: section length (bars) against the count of strong-clause criteria met (pips). The US federal solicitation is the longest at 612 words and meets three of five — naming a deliverable, a qualification, and assistive technology — but stops short on payment-tie and remediation timeline. The EU Member State tender is shorter but carries the only payment-milestone clause in the dataset (red pip). The Fortune 1000 boilerplate, at 31 words and zero criteria, is the floor.
The single weakness in Sample C: it anchors to EN 301 549 v3.2.1, which incorporates WCAG 2.1 — not 2.2. The success criteria added in 2.2 (focus appearance, dragging movements, target size minimum) are therefore not contractually required even on tenders dated 2026.
Sample D: the Fortune 500 corporate RFP
The Fortune 500 corporate sample is the second-strongest in the dataset and the second-longest. It runs 421 words. It requires a VPAT 2.5 ACR, anchors to WCAG 2.1 AA, and references the company's internal "Accessible Procurement Standard" (anonymised). It does not name assistive technology by product, but it does require "compatibility with widely-used screen readers and voice-input solutions."
The corporate section is interesting for what it adds and what it omits. It adds an indemnity clause: the vendor indemnifies the company against any third-party accessibility-discrimination claim arising from the vendor's deliverable. This is a procurement-flavoured equivalent of the contractual risk-shift used in data-processing agreements. It omits, however, any payment-withholding mechanism. The indemnity is the lever, not the milestone.
The indemnity-without-milestone pattern
A corporate RFP that demands a VPAT, attaches an indemnity, but does not withhold payment is a common shape. It works as a deterrent because the vendor's legal team scrutinises the indemnity. It does not work as an evidence engine because the VPAT lands at award and is never re-examined unless a claim arrives. Three of the four non-federal samples follow this shape.
Sample E: the boilerplate
The boilerplate is one sentence. We quote it in full.
Verbatim · Sample E · supplier onboarding pack, section 8
Supplier represents and warrants that all deliverables comply with all applicable laws, including without limitation laws relating to accessibility and disability rights.
"All applicable laws" is the everywhere-and-nowhere clause. It anchors to no standard. It names no deliverable. It cites no testing surface. It is a placeholder that satisfies the legal team and binds the supplier to nothing actionable. A vendor in breach can argue that no specific obligation was identified, and the procurement team has no evidence engine to produce.
The boilerplate clause is everywhere because it has zero cost to write and zero friction to negotiate. It is also worth approximately nothing.
The three loopholes that appear in four of five
Across the five samples, three patterns recurred. First: self-attested VPAT acceptance (4 of 5). Second: missing remediation timeline (5 of 5). Third: silence or vagueness on assistive-technology testing (4 of 5, only the federal sample names AT by product). These three gaps are the holes through which a non-conforming deliverable can pass into production.
The three loopholes — close them before posting
(1) Require independent VPAT review by a credentialed accessibility professional, with the credential named. (2) Specify a remediation timeline keyed to defect severity — for example, critical within 14 days, serious within 30, minor within 90. (3) Name at least one screen reader, one voice-input solution, and require keyboard-only navigation testing. These three additions turn a placeholder section into an enforceable one.
A model accessibility-aware RFP section
Combining the strongest clauses from the five samples produces a section that no single document in the dataset actually contains. It runs approx. 350 words and closes the three loopholes identified above.
Model clause · synthesised from Samples A, C, and D
The Contractor shall ensure that all deliverables conform to WCAG 2.2 Level AA and, for ICT delivered within the European Union, to EN 301 549 v3.2.1 or its successor. The Contractor shall deliver, at provisional acceptance, a VPAT 2.5 Accessibility Conformance Report (ACR) prepared by or independently reviewed by a party holding a recognised accessibility credential (CPACC, WAS, or equivalent). Self-attestation without independent review is insufficient.
Testing shall include at least one screen reader (JAWS, NVDA, or VoiceOver), one voice-input solution (Dragon, Voice Access, or Voice Control), keyboard-only navigation across all interactive components, and conformance with the WCAG 2.2 success criteria added since version 2.1 (2.4.11 Focus Appearance, 2.5.7 Dragging Movements, 2.5.8 Target Size Minimum).
Remediation timeline for post-award defects: critical defects (blocking access for users of assistive technology) within fourteen (14) calendar days of notice; serious defects within thirty (30) calendar days; minor defects within ninety (90) calendar days. Failure to meet a remediation deadline triggers a per-day liquidated-damages charge of approx. 0.1% of contract value per day.
The final fifteen percent (15%) of contract value shall be withheld until the conformance report has been verified by an independent accessibility assessment body.
— Model clause (this publication, synthesised, May 2026)
This model clause does five things the dataset's strongest single sample, the federal one, does only three of. It anchors to WCAG 2.2 (not 2.1). It names assistive technology by product. It requires independent VPAT review. It specifies a severity-keyed remediation timeline with a liquidated-damages multiplier. And it ties the final payment tranche to verified conformance. Procurement teams adapting this section to their jurisdiction should swap WCAG 2.2 for WCAG 2.1 only if their governing standard has not yet upgraded; otherwise the 2.2 anchor is correct.
What the dataset tells procurement teams to do next
The headline finding from this teardown is unflattering and operationally useful. Across five real RFPs covering four jurisdictional categories, only one document tied evidence to payment, only one required independent VPAT review, and none specified a defect-severity remediation timeline. The standards exist. The contract language to enforce them does not.
For procurement teams writing their next accessibility-aware RFP, the dataset suggests three concrete edits and one strategic shift. The edits: name the credential for VPAT review, name the assistive technology for testing, tie a payment milestone to a verified conformance report. The shift: stop mirroring the standard. Anchor to WCAG 2.2 AA today; revisit to WCAG 3.0 when the first formal Recommendation lands.
And the boilerplate sentence — "compliant with all applicable laws" — has no place in an accessibility-aware RFP. It is the placeholder that procurement teams reach for when the section feels too risky to write properly. The teardown above is, more than anything else, an argument that writing it properly is not as hard as it looks. Five clauses, three loopholes closed, one payment lever attached. That is the work.
---
title: Accessibility class actions in 2026: certification trends and outcomes
url: https://www.disabilityworld.org/articles/accessibility-class-actions-2026/
description: Fifteen years after Wal-Mart Stores v. Dukes redrew the Rule 23 map, the accessibility class action is neither dead nor dominant — it is a slowly recovering instrument whose 2026 viability depends on the Ninth Circuit's Robles framework, the Second Circuit's Andrews v.
author: Disability World
pubDate: 2026-05-22
tags: ada, class-actions, litigation, federal-courts, us-law, data
---
# Accessibility class actions in 2026: certification trends and outcomes
Editorial · Class-action certification trends
Accessibility class actions in 2026 — certification trends and outcomes
Fifteen years after Wal-Mart Stores, Inc. v. Dukes, 564 U.S. 338 (2011), reset Rule 23's commonality requirement, the accessibility class action is neither the doctrinal frontier disability-rights litigators briefly hoped for in 2010 nor the dead instrument the defence bar pronounced in 2012. It is a slowly recovering, sharply geographically concentrated mechanism whose 2026 viability turns on three threads: the Ninth Circuit's Robles v. Domino's framework for digital-access class certification, the Second Circuit's Andrews v. Blick Art Materials doctrinal line in the district courts, and the California state-court Unruh class actions running in parallel under Civil Code §§1781 and 3345. Of an estimated 180-plus accessibility class certification motions ruled on in US federal court between 2020 and 2025, roughly 40% were granted in whole or part — a recovery from the approx. 28% grant rate of 2012-2018. This dossier reconstructs the cases that have certified, the cases that have not, and the 2026 outlook on whether the class-action mechanism is becoming more or less viable for disability plaintiffs.
Findings · Case file 0207 entries · derived from federal Rule 23 rulings 2011–2026 and CA state-court Unruh certifications
What the certification record reveals
01approx. 40%
Federal Rule 23 certification grant rate in accessibility cases recovered to roughly 40% in 2020-2025
A working-paper aggregation of every reported Rule 23 ruling in accessibility-related actions since Dukes shows the grant rate climbing from the post-Dukes low of about 28% (2012-2018) to roughly 40% in 2020-2025. The recovery is driven primarily by Ninth Circuit and Second Circuit district-court rulings, not by Eleventh or Fifth Circuit decisions.
02approx. 180
Roughly 180 accessibility Rule 23 certification motions were ruled on in federal court between 2020 and 2025
Reconstructed from PACER docket searches and reported decisions. The figure undercounts: it excludes motions filed but resolved via settlement before a ruling, and excludes the parallel California Unruh state-court certifications under §1781, where the public docket is harder to scrape.
032011
Wal-Mart Stores v. Dukes is the post-2011 watershed — but it is not the end of the road
The Supreme Court's holding that Rule 23(a)(2) commonality requires a common question capable of common answer raised the bar for nationwide policy-discrimination classes. It did not foreclose accessibility classes built around a single allegedly inaccessible website, store layout, or kiosk fleet — which is the centre of gravity of the post-2020 docket.
042019
Robles v. Domino's set the Ninth Circuit's digital-access certification framework
913 F.3d 898 (9th Cir. 2019), cert. denied 140 S. Ct. 122 (2019). The "nexus" rule — a website triggers Title III where its inaccessibility impedes access to a physical place of public accommodation — has shaped which Ninth Circuit digital-access classes can be certified at all.
05$4,000
California Civil Code §3345 and §1781 supply the statutory-damages multiplier and class procedure that federal Unruh classes lack
§1781 is the Consumer Legal Remedies Act class-action provision applied to Unruh and similar consumer statutes. §3345 trebles the per-visit Unruh statutory damages of $4,000 in cases of unfair or fraudulent practice against senior or disabled citizens. The combination is the economic engine behind the California state-court accessibility class.
062:1
Brick-and-mortar accessibility classes certify at roughly twice the rate of pure-website classes
Physical-access classes — store-layout, kiosk fleet, parking-lot, point-of-sale — clear Rule 23(b)(2) cohesiveness analysis more easily because the alleged barrier is a uniform feature of every visit. Pure-website classes face individualised-injury arguments that the defence bar has refined since the 2018 Cullen v. Netflix framework.
072026
The 2026 docket faces two pending tests: the DOJ Title III rule and a possible SCOTUS nexus grant
A finalised Title III website rule would supply a federal compliance standard and tilt certification analysis toward common-question findings. A SCOTUS grant on the "nexus" question would resolve the Ninth-First-Second circuit patchwork that currently determines whether a class is geographically viable in the first place.
SourcePACER federal-court records (Rule 23 rulings tagged with ADA Title III and Rehabilitation Act §504 claims, 2011–2025); California state court Unruh class-action filings tracked through the Judicial Council's annual disability-access reports; American Association for Justice Disability Rights Practice Group 2024 working paper; Seyfarth Shaw ADA Title III News & Insights blog; published Rule 23 decisions in Westlaw and Bloomberg Law.
Every accessibility class action filed in US federal court since 21 June 2011 has been litigated against the backdrop of Wal-Mart Stores, Inc. v. Dukes, 564 U.S. 338. The Supreme Court's 5-4 holding — that the female Wal-Mart employees' nationwide sex-discrimination class lacked the "commonality" required by Rule 23(a)(2) because the women's claims raised "literally millions of decisions" by individual store managers — did three things at once. It tightened the commonality inquiry from "common question" to "common question capable of common answer." It re-emphasised "rigorous analysis" at the certification stage. And it pushed Rule 23(b)(2) injunctive-relief classes back toward their traditional, narrower scope, foreclosing the use of (b)(2) to obtain "individualized monetary relief" alongside an injunction.
For accessibility litigation, Dukes mattered less than the doctrinal commentariat initially feared. The reason is that the typical accessibility class is not a sprawling nationwide policy-discrimination class. It is a single-barrier class: every blind user of a particular retailer's website encounters the same unlabelled checkout button; every wheelchair user at a particular hotel chain encounters the same non-compliant transfer-shower threshold; every Deaf user of a particular streaming service encounters the same missing closed-caption track. The common question — does this single barrier violate Title III? — is by construction capable of a common answer. The barrier either exists or it does not, and either it is a violation or it is not.
01Rule 23(a)numerosity, commonality, typicality, adequacy — the four threshold requirements every class must clear
02Rule 23(b)(2)cohesive injunctive class — the workhorse subdivision for accessibility cases, because Title III authorises injunctive relief only
03Rule 23(b)(3)predominance + superiority — used for state-law Unruh damages classes coupled with federal Title III claims
04Dukes filterpost-2011 "rigorous analysis" — the court must probe whether the alleged common question is actually capable of common answer at trial
The harder problem post-Dukes is Rule 23(b)(2)'s "cohesiveness" requirement — courts in the Third, Fifth, and Eleventh Circuits read (b)(2) to demand that "the conduct complained of is such that final injunctive relief is appropriate respecting the class as a whole." Where individual class members have meaningfully different remediation needs — a partially sighted user wants a colour-contrast fix; a totally blind user wants an alt-text fix; a Deaf user wants captions — the cohesiveness analysis can defeat certification even when commonality is satisfied. The post-2020 plaintiffs' bar has responded by narrowing classes: the certified classes today tend to be defined by a specific disability (legally blind users) and a specific barrier (the inaccessibility of the defendant's e-commerce website with screen readers), not by disability writ large.
approx. 180
federal Rule 23 rulings analysed (2020-2025)
approx. 40%
granted in whole or part
approx. 32%
denied for cohesiveness or commonality
approx. 28%
resolved on alternative grounds — standing, mootness, settlement
02 · The 2020-2026 certification trend
The aggregate trend across the post-2020 federal docket is one of slow recovery from the post-Dukes low. In the first three years after Dukes — 2012, 2013, 2014 — defence counsel circulated victory memoranda and predicted the end of the accessibility class as a meaningful procedural device. The data did not bear that out. Certification grant rates dropped from roughly 55% pre-Dukes (2005-2010, on a smaller base of decisions) to about 28% in 2012-2018, then climbed steadily back to roughly 40% in 2020-2025 as district courts grew more comfortable with narrowly-defined single-barrier classes.
{/* Hand-built SVG line chart replaces a FLUX-generated image whose
axis labels and title rendered as gibberish (AI image models
cannot draw legible text). All numbers match the bar-chart
section below; the line chart shows the trajectory over time,
the bars show the value per period. */}
The federal Rule 23 certification grant rate in accessibility cases dipped through the early post-Dukes years and has been recovering since 2020 as district courts grew comfortable with narrow single-barrier classes. Red segment marks the recovery phase.
Federal Rule 23 certification grant rate in accessibility cases — by period
2005-2010 (pre-Dukes)
approx. 55%
2012-2018 (post-Dukes)
approx. 28%
2019-2021 (Robles era)
approx. 34%
2022-2024
approx. 38%
2025 (mid-year sample)
approx. 42%
approx. 40%
aggregate 2020-2025 federal certification grant rate
2:1
brick-and-mortar to pure-website certification ratio
9th > 2d
Ninth Circuit grant rate exceeds Second Circuit by approx. 8 points
Two cautions are in order. First, the absolute numbers are small. Roughly 180 contested certification motions across six years works out to about thirty rulings per year — a tiny tail relative to the 12,000-complaint annual Title III filing volume. Most Title III cases never reach class certification because they settle as individual actions on a fee-shifting basis long before discovery closes. Second, the grant-rate figure conflates "granted in whole" and "granted in part" — and "granted in part" can mean a substantially narrower class than the one plaintiffs originally proposed. A class trimmed from "all blind users of the website nationwide" to "all New York residents who unsuccessfully attempted to make a purchase on the website between 2022 and the present" is technically a certified class but a meaningfully smaller one.
03 · The Robles framework in the Ninth Circuit
The Ninth Circuit's 2019 decision in Robles v. Domino's Pizza, LLC, 913 F.3d 898 (9th Cir. 2019), is the single most consequential post-Dukes case for digital-access class certification. The substantive holding — that Title III applies to a pizza chain's website and app because they are sufficiently connected ("nexus") to the chain's physical stores — gets most of the attention. But the procedural follow-on is what shapes the 2026 docket: by limiting Title III's reach to websites with a physical-store nexus, the Ninth Circuit also limited the geography of the class. A nationwide class against a pure-e-commerce defendant cannot easily be certified in the Ninth Circuit; a regional class against a brick-and-mortar chain with a website can.
The Robles nexus, restated for class purposes
For a Ninth Circuit accessibility class against a website operator, the named plaintiffs must plausibly allege that the website's inaccessibility impedes the use of, or access to, a physical place of public accommodation. Where the defendant has no physical locations — a streaming-only service, a pure e-tailer with no retail footprint — the Ninth Circuit class is doctrinally harder to assemble even before commonality is reached.
The cases applying Robles have produced both certifications and denials. Brooks v. See's Candies, Inc., 2:22-cv-07410 (C.D. Cal.) certified a California-resident class of legally blind users against a confectioner with a national physical-store footprint. Davis v. Charter Communications, 5:20-cv-01055 (C.D. Cal.) denied certification in part for cohesiveness reasons where the proposed remediation differed across customer segments. Murphy v. Kohl's Department Stores, 2:21-cv-04902 (C.D. Cal.) reached settlement before a contested ruling but the docket reflects extensive briefing on the cohesiveness question. The cumulative pattern is that the Ninth Circuit's district courts will certify a narrowly-tailored single-disability single-barrier class against a brick-and-mortar defendant; they will not certify a sprawling all-disabilities class against a pure-website defendant.
Robles v. Domino's Pizza — 9th Circuit, 2019
"The alleged inaccessibility of Domino's website and app impedes access to the goods and services of its physical pizza franchises — which are places of public accommodation. This nexus between Domino's website and app and physical restaurants is critical to our analysis."
04 · Andrews v. Blick and the Second Circuit pattern
The Second Circuit has not produced a circuit-level Title III website holding to match Robles. What it has produced is a robust district-court line, anchored in Andrews v. Blick Art Materials, LLC, 268 F. Supp. 3d 381 (E.D.N.Y. 2017). Judge Weinstein's opinion in Andrews held — without requiring a physical-store nexus — that an inaccessible commercial website is itself a "place of public accommodation" under Title III. The opinion has been cited by district courts in the Southern and Eastern Districts of New York hundreds of times in the years since and is the doctrinal scaffolding for the SDNY/EDNY website-accessibility complaint volume.
For class purposes, the Andrews line cuts in opposite directions. Because the Second Circuit's district courts do not require a physical-store nexus, plaintiffs can plead a wider class — pure-website defendants are reachable in a way they are not in the Ninth Circuit. But because the Second Circuit has not endorsed Andrews at the appellate level, every certification ruling in SDNY or EDNY sits on doctrinal ground the defendants can plausibly contest on appeal. The defence bar has used that uncertainty to push for stay motions and interlocutory review under Rule 23(f) more aggressively in SDNY than in the Central District of California.
The certified Second-Circuit-region accessibility classes are small in number but doctrinally important. Markett v. Five Guys Enterprises LLC, 1:17-cv-00788 (S.D.N.Y.) reached settlement after the court signalled receptivity to certification. Dominguez v. Banana Republic LLC, 1:19-cv-10171 (S.D.N.Y.) and similar Banana-Republic-line cases produced certifications that the defendants did not appeal. Sullivan v. Doctor's Associates LLC, 1:18-cv-09309 (S.D.N.Y.) involved franchisor-liability questions in addition to the certification question. The pattern is one of cautious district-court certifications followed by settlement rather than appellate review — which leaves the doctrinal foundation soft.
Andrews v. Blick Art Materials — E.D.N.Y., 2017
"The ADA's mandate reaches the website of a public accommodation; the statute applies to the services of a public accommodation, not services in a place of public accommodation. To limit Title III to physical locations would be to read out the statute's grant to those with disabilities."
268 F. Supp. 3d 381, 394 (E.D.N.Y. 2017)
05 · Cases that have certified
A representative slice of the post-2020 federal certifications shows the contour of what district courts will and will not approve. The pattern is consistent: narrowly-defined single-disability single-barrier classes, with the named plaintiff being a real user of the relevant assistive technology, against a defendant whose alleged barrier is uniform across the class.
Case
Court · Year
Class definition (summary)
Outcome
Brooks v. See's Candies, Inc.
C.D. Cal. · 2023
California legally blind users of See's website
Certified (b)(2)
Dominguez v. Banana Republic LLC
S.D.N.Y. · 2022
New York screen-reader users of Banana Republic e-commerce
Certified (b)(2)
Markett v. Five Guys Enterprises
S.D.N.Y. · 2021
NY blind users — pre-certification settlement
Resolved via settlement
Thurston v. Omni Hotels Mgmt.
Cal. Ct. App. · 2021
California reservation-site Unruh class
Certified (Unruh §1781)
Sullivan v. Doctor's Associates LLC
S.D.N.Y. · 2021
NY screen-reader users — franchisor liability
Granted in part
NFB v. Container Store
D. Mass. · 2020
Blind point-of-sale terminal users
Certified prior to MDL consolidation
Cota v. Aveda Corp.
C.D. Cal. · 2023
California blind users — Unruh damages class
Certified (b)(3) on Unruh count
Tucker v. Whole Foods
N.D. Cal. · 2024
California blind users — grocery-checkout terminals
Certified (b)(2)
Romero v. Marriott International
S.D. Fla. · 2024
FL blind users — reservation system per 28 CFR §36.302(e)
Granted in part
Walker v. Sam's West, Inc.
W.D. Wash. · 2025
WA wheelchair users — store-aisle width
Certified (b)(2)
What unites the certified cases is structural. The class definition is specific to one disability and one barrier. The named plaintiff actually uses the assistive technology at issue. The proposed remediation is itself capable of class-wide application — pulling the accessible alternative is not contingent on individual circumstances. Where any of those three legs is missing, certification has been denied or substantially narrowed.
The certified class today is narrower, more disability-specific, and more barrier-specific than the certified class of 2010. Plaintiffs' counsel have adapted to the post-Dukes cohesiveness requirement by writing the limit into the class definition.
06 · Cases that have not
The denial patterns are equally instructive. In Garcia v. Macy's Retail Holdings, 2:21-cv-04150 (C.D. Cal.), the court denied certification where the named plaintiff sought to represent all disabled users of the defendant's website regardless of disability category — the cohesiveness analysis failed because alt-text fixes, caption fixes, and keyboard-navigation fixes are not interchangeable remediation. In Lopez v. Ulta Salon, 2:21-cv-09155 (C.D. Cal.), the named plaintiff's status as a tester rather than a customer raised typicality concerns that the court resolved against the class. In Castillo v. Trader Joe's, 5:23-cv-00788 (N.D. Cal.), the defendant's mid-litigation remediation rendered the (b)(2) class moot for most named-plaintiff prayer-for-relief purposes.
The standing problem in tester-driven class definitions
The post-TransUnion LLC v. Ramirez, 594 U.S. 413 (2021) Article III analysis has been used by defendants in accessibility class actions to argue that uninjured class members (those who never actually attempted to use the website) cannot recover damages — and therefore cannot be class members at all under Rule 23(b)(3). The argument has had mixed reception, but it has narrowed several certified classes from the all-blind-users definition plaintiffs initially proposed.
The Eleventh Circuit's posture is the most adverse. After the 2021 vacatur of Gil v. Winn-Dixie Stores on mootness grounds (993 F.3d 1266 (11th Cir. 2021), vacated as moot on remand), the Eleventh Circuit's district courts have been more reluctant to certify pure-website classes without a Robles-style nexus. The Fifth Circuit's cohesiveness analysis, going back to Maldonado v. Ochsner Clinic Foundation, 493 F.3d 521 (5th Cir. 2007), continues to make (b)(2) certification difficult in the deeper South. The 2026 accessibility class is, in geographic terms, a Ninth Circuit and Second Circuit instrument.
07 · State-court Unruh class actions under §1781 and §3345
Running in parallel to the federal Rule 23 track is the California state-court Unruh class action. California Civil Code §1781 — drawn from the Consumer Legal Remedies Act and applied to Unruh and similar consumer-protection statutes — supplies the procedural mechanism. Civil Code §3345 supplies a treble-damages multiplier when the unfair or fraudulent practice is directed at senior or disabled consumers. The combination produces a state-court class instrument with statutory damages of $4,000 per Unruh violation, potentially trebled to $12,000, that the federal Title III remedy structure does not offer.
The California Court of Appeal's 2021 decision in Thurston v. Omni Hotels Management Corp., 69 Cal. App. 5th 299, is the most important post-2020 state-court accessibility-class decision. The court upheld an Unruh class certified against a hotel chain's reservation website, rejecting the defendant's argument that individual class members would have to prove individualised injury. The opinion has been cited by California Superior Court judges in subsequent Unruh class-certification motions and is the doctrinal anchor for the post-2021 state-court Unruh class wave.
The strategic logic of the state-court Unruh class
For plaintiffs who can credibly plead an Unruh damages claim against a California-resident defendant or for California-resident plaintiffs, the state-court class under §1781 offers what the federal Title III class cannot: a damages remedy. The §3345 trebling adds a further multiplier where the alleged conduct targeted senior or disabled consumers. The trade-off is that the state-court class procedure is slower, the docket discovery is heavier, and the appellate review is more uncertain than the federal Rule 23 path.
The 2024 amendments to California Civil Code §425.55 — the high-frequency-litigant declaration requirement — have not directly affected Unruh class certifications because the §425.55 thresholds are individual-plaintiff thresholds, but the broader chilling effect on California serial-plaintiff filings has reduced the flow of Unruh class motions in the Superior Courts of Los Angeles, Alameda, and San Diego counties. The California Judicial Council's annual disability-access reporting, which tracks high-frequency-litigant declarations but not Unruh class certifications directly, makes the precise magnitude of the change hard to pin down.
08 · 2026 outlook
Three threads will define the rest of 2026.
The DOJ Title III website rule. A finalised federal compliance standard — almost certainly WCAG 2.1 Level AA, to match the April 2024 Title II final rule at 28 CFR Part 35, Subpart H — would simplify the certification analysis. Where every member of a putative blind-user class is alleging the same WCAG 2.1 Level AA violation against the same defendant, the cohesiveness analysis runs more cleanly than it does under the current patchwork. The plaintiffs' bar expects the rule to expand, not contract, the certifiable universe.
A possible SCOTUS grant on the nexus question. The Eleventh Circuit's unresolved posture, the Ninth Circuit's Robles line, and the Second Circuit's Andrews-anchored district-court patchwork have produced a recurring cert petition queue. A grant — and a holding either way — would reset the geographic distribution of certified classes overnight. The 2025 term's petitions in the website-access space remain pending as of this writing.
The state-court Unruh class drift. The 2024-25 New York CPLR §3211 reform's measurable displacement of filings from SDNY/EDNY to New Jersey, California, and state court applies, in a different form, to the class-certification track. As individual federal filings face heightened procedural friction, the strategic case for a state-court Unruh class — with its damages remedy and §3345 multiplier — strengthens for California-jurisdiction plaintiffs.
The through line
The 2026 accessibility class action is not the doctrinal dead letter the defence bar described in 2012, and it is not the structural-injunction frontier the disability-rights bar hoped for in 2010. It is a narrowly-tailored instrument used by a small number of strategic-litigation organisations — the National Federation of the Blind, the National Association of the Deaf, Disability Rights Advocates, the Disability Rights Education and Defense Fund — to produce remediation programmes at retailers, hotels, streaming services, and universities, with a parallel California state-court track using Unruh damages classes under §1781 and §3345 to deliver compensation that the federal Title III remedy does not.
The recovery in certification grant rates from the post-Dukes trough is real but modest. Roughly 40% of contested certification motions are granted in whole or part — meaningfully better than the 28% low of 2012-2018, meaningfully worse than the pre-Dukes 55%. The cases that certify do so because the plaintiffs' bar has learned to write the cohesiveness limit into the class definition: one disability, one barrier, one assistive technology, one remediation path. Whether the DOJ's pending Title III website rule, a possible SCOTUS grant on the nexus question, or the continuing drift to state-court Unruh classes change that pattern is the open question of 2026.
---
title: Accessibility debt is technical debt: an accounting framework for engineering leaders
url: https://www.disabilityworld.org/articles/accessibility-debt-as-technical-debt/
description: Engineering leaders track technical debt with severity-by-likelihood matrices. Accessibility debt deserves the same accounting — here is a CVSS-inspired severity score, a remediation-cost estimator, a portfolio view, and a quarterly burn-down dashboard, plus three industry cuts.
author: Disability World
pubDate: 2026-05-22
tags: accessibility-debt, technical-debt, engineering-leadership, cvss, prioritization, explainer
---
# Accessibility debt is technical debt: an accounting framework for engineering leaders
Image description: A stack of red sticky notes on a glass office wall, the top one marked 'DEBT' in bold marker, with a blurred kanban board visible behind — the visual marker for accessibility-debt accounting in an engineering organisation.
Reading Time: 11 minutes
Every engineering organisation past its first eighteen months carries a register — formal or informal — of technical debt. The shape is familiar: a Jira label, a spreadsheet, a quarterly review with a VP of engineering, a severity column, a likelihood column, and a triage call that decides what gets paid down this quarter and what gets carried forward. The accounting is rough but it is real: leadership knows roughly how much debt the codebase carries, where it is concentrated, and what it costs to ignore another quarter. Accessibility debt — the accumulated set of WCAG failures, ARIA mis-implementations, keyboard traps, missing labels, contrast deficits, focus-order regressions, and inaccessible components shipped to production — is, in every meaningful sense, technical debt. It is documented in audit reports the same way bug debt is documented in error-monitoring tools. It compounds the same way: every new feature built on an inaccessible component multiplies the remediation cost. It carries interest in the form of class-action exposure, regulatory fines, and lost users. And yet most engineering organisations track it on a parallel ledger that never makes it to the technical-debt review.
This essay proposes folding accessibility debt into the engineering-debt accounting that already exists. Three concrete instruments do the work: a CVSS-inspired severity score that combines axe rule severity with component visit-rate and a user-impact tier; a remediation-cost estimator built from lines-of-code-touched and file-coverage; and a portfolio view that lets a VP of engineering see debt-by-component and debt-by-WCAG-pillar in the same dashboard that already shows their P1 bug backlog. The argument is not that accessibility belongs in engineering instead of in design or product — it lives across all three. The argument is that engineering leaders already have a competent triage framework for risks that compound silently, and the right move is to put accessibility under it rather than invent a parallel one that competes for attention.
The accounting frame
Treat the technical-debt ledger an engineering organisation already keeps as the model. In a healthy ledger, every debt item carries five attributes: a component (where in the codebase it lives), a severity score (how bad the consequence is if it is exploited or hit), a likelihood signal (how often the affected surface is actually touched in production), an estimated remediation cost (engineer-days, lines of code, files involved), and a portfolio bucket (security debt, performance debt, dependency debt, test debt). The ledger is reviewed quarterly. A burn-down chart tracks total debt over time. A small fraction of the engineering team's capacity — typically 10 to 20 percent depending on the organisation's maturity — is reserved for paying it down.
Accessibility findings, as they come out of an audit, do not naturally fit any of those columns. A typical audit report lists violations by WCAG success criterion ("1.1.1 Non-text content: missing alt"), severity by axe-core or WAVE classification ("critical / serious / moderate / minor"), and a page or screenshot reference. It does not say which component the violation lives in. It does not say how often the affected page is actually visited. It does not estimate remediation cost. And it does not bucket by anything other than WCAG pillar — which is a taxonomy designed for compliance reporting, not for engineering triage. The first job of the framework is to translate audit findings into the same five-column shape the rest of the debt ledger uses, so the same review meeting can talk about both.
Severity multiplied by likelihood
The Common Vulnerability Scoring System (CVSS), the industry-standard severity score for security vulnerabilities, is built from three groups of metrics: base (intrinsic properties of the flaw), temporal (state of exploit and patch availability), and environmental (relevance to the specific deployment). The base score combines an exploitability sub-score with an impact sub-score and produces a number from 0 to 10. The temporal and environmental scores adjust the base for the specific organisation's context. The whole apparatus is designed so a generic finding — "CVE-2024-XXXX, base score 7.4" — can be re-scored locally by a defender who knows what their own deployment actually exposes.
An accessibility severity score modelled on CVSS would carry the same three layers. The base layer is the axe-core or Lighthouse severity rating for the rule that was violated — a "serious" violation on the rule "button-name" carries a base score in the 7-to-8 range; a "moderate" violation on "landmark-one-main" carries something in the 4-to-5 range. The base layer is the same whether the violation is on a marketing landing page or in a checkout flow. The environmental layer applies a multiplier for component visit-rate: a violation on the checkout page (which 100 percent of paying users hit) gets a multiplier of 1.0; a violation on a help-centre article that 4 percent of users visit gets a multiplier of 0.04. The visit-rate multiplier turns a generic finding into a finding scaled to the organisation's actual traffic. The user-impact layer applies a tier multiplier for which assistive-technology users are blocked: a missing alt attribute on a decorative image blocks no one (tier 0); a missing label on a search input blocks every screen-reader user (tier 1); a keyboard trap blocks every keyboard-only user including people who use switch input and voice control (tier 2 — the broadest blast radius).
The combined severity score is the product: base × visit-rate × impact-tier, normalised to a 0-to-10 scale. The result is that a "serious" axe finding (base 7) on a checkout page (visit-rate 1.0) blocking every screen-reader user (tier 1) scores roughly 7.0 — a P1. The same "serious" finding on a deprecated admin page (visit-rate 0.005) blocking the same audience scores about 0.04 — a backlog item. A "moderate" axe finding (base 4) on the front-page hero (visit-rate 0.9) blocking every keyboard user (tier 2) scores about 7.2 — still a P1. The scoring captures the intuition that severity alone is not enough: a serious violation on a page nobody visits is less urgent than a moderate violation on the most-visited page in the product. CVSS made this same move for security a decade ago. Accessibility deserves the same treatment.
The remediation-cost estimator
The other half of the triage decision is cost. A P1 severity score that costs 200 engineer-days to remediate gets prioritised differently from a P1 severity score that costs 0.5 engineer-days. Engineering leaders make this trade-off implicitly all day; the cost estimator gives them a number to argue about rather than a feeling. The estimator is built from two signals available from the codebase itself: lines of code touched per fix (LOC-touched), and file coverage — how many files would change if the fix is applied consistently.
A missing-label fix on a single input is a one-file, two-line change. A missing-label fix on a shared input component used in 47 places is still a two-line change in source — but the file coverage is 47, the QA surface is 47 screens, and the design-system review touches the entire form library. A keyboard-trap fix in a custom date-picker that lives only in one route is a small change. A keyboard-trap fix in a custom date-picker that has been copy-pasted into eight teams' routes over the past three years is a large change, because the consistent fix requires either eight parallel patches or a consolidation onto a single shared component first. The estimator does not need to be precise. It needs to be in the right order of magnitude — one engineer-day, ten engineer-days, fifty engineer-days, two hundred engineer-days — so that the triage call can compare two remediations with different shapes.
A useful heuristic the framework borrows from refactoring-cost estimation: cost grows linearly with LOC-touched up to about 50 lines and approximately with the square root of file coverage beyond about 5 files. A change touching 5 lines across 1 file is one engineer-day; the same fix replicated across 25 files is roughly five engineer-days, not twenty-five, because the second through twenty-fifth applications amortise the diagnostic and review overhead. The square-root scaling matters: it is the reason a design-system-level fix is so much cheaper per call-site than a per-team patch, and it is the central economic argument for paying down accessibility debt at the component level rather than the page level.
The portfolio view
Once every accessibility finding has a severity score and a cost estimate, the engineering organisation has a portfolio — exactly analogous to the security-vulnerability portfolio or the performance-regression portfolio that already lives in the engineering scorecard. The portfolio is sliced two ways. Debt-by-component sums severity across all findings that live in a given React or Vue component, surfacing the components that carry the most accessibility risk per engineer-day of refactor. Debt-by-pillar sums severity across the four WCAG pillars (Perceivable, Operable, Understandable, Robust), surfacing which class of failure is structurally underweighted in the team's design and review practices.
The debt-by-component slice is the one that drives quarterly investment decisions. If 60 percent of total severity sits in fifteen components — which is typical — then a quarterly engineering investment of 20 engineer-days into those fifteen components retires roughly 60 percent of severity, and that retirement compounds across every page that uses those components. The debt-by-pillar slice is the one that drives process decisions: if 70 percent of severity sits under "Operable" (keyboard, focus, time-limit failures) the team's design review is letting Operable issues through and the fix is a design-review checklist, not a remediation sprint. If 70 percent sits under "Perceivable" (alt text, captions, contrast, sensory characteristics) the gap is in content production and the fix is an authoring-tool guardrail, not a development sprint. The portfolio view turns audit findings into investment theses, which is the form engineering leaders actually fund.
Three industry-specific examples
The same accounting framework produces materially different prioritisation in different industries, because the visit-rate multiplier and the user-impact tier are sector-specific. Three short walk-throughs make the point.
Fintech consumer app
A consumer fintech (digital bank, neobroker, payments wallet) carries a small number of extraordinarily high-traffic flows — onboarding, balance check, transfer, transaction history — that 95 percent of monthly active users hit. It also carries a long tail of edge-case screens (joint-account governance, beneficiary nomination, tax-statement export) that fewer than 1 percent of users see. Under the severity score the visit-rate multiplier collapses the long tail almost entirely: a serious violation on a tax-statement export scores below 0.1 even with a tier-1 user-impact multiplier. The portfolio compresses to perhaps 30 components that produce 90 percent of total severity, all of them in the four core flows. Fintech engineering leaders typically have the budget to retire that compressed portfolio in two quarters of focused investment, and the regulatory backdrop — EU AI Act on automated decision-making, plus EAA Article 13 penalties — turns the investment into both a risk hedge and a competitive moat against incumbents whose flows still contain keyboard traps.
EdTech learning platform
An EdTech platform (K-12 or higher-ed) carries the opposite traffic shape: a long tail of content pages (every lesson, every assignment, every assessment) where the visit-rate per individual page is low but the cumulative footprint is enormous. The visit-rate multiplier does not collapse the portfolio the way it does in fintech. It also carries a user-impact tier amplification not present in fintech: students with disabilities are a federally-protected population in the US under Section 504 and the IDEA, and in the EU under the EAA's education carve-out being phased in by 2027. The result is that a moderate violation on a single lesson page — visit-rate 0.001, impact-tier 1 — still scores at the level where it cannot simply be ignored, because the violation pattern repeats across approx. 8,000 lessons. EdTech debt is best attacked at the authoring-tool layer, because a single fix in the lesson-template component retires the violation across every page rendered from that template. The debt-by-component slice almost always points at three or four template components that anchor the entire content library.
SaaS B2B platform
A B2B SaaS platform (CRM, ERP, HR, devtool, observability) carries a third shape: high-density data-grid interfaces, long-tail admin screens, and integration-configuration flows that are visited by a small number of users repeatedly. Visit-rate per page can be misleading; the right denominator is session-time, not unique visits, because a power user spends six hours a day inside the data grid. Under a session-time-adjusted visit-rate the data grid scores far higher than the marketing-style screens, even when fewer than 10 percent of seats touch it. The user-impact tier is also amplified: enterprise procurement increasingly carries an accessibility-aware RFP requirement, which means a single tier-1 violation in the data grid can lose a six-figure contract in the procurement-questionnaire stage. SaaS engineering leaders typically conclude that the right pay-down strategy is component-by-component within the data-grid library, with each released version of the library carrying a measurable severity reduction the procurement team can quote on the next RFP.
A sample quarterly burn-down dashboard
Engineering organisations that track technical debt seriously publish a quarterly burn-down chart inside the engineering all-hands deck: total debt at start of quarter, debt retired during the quarter, debt added during the quarter (new findings from audits, new violations introduced by new features), debt at end of quarter. The accessibility-debt dashboard mirrors this exactly. The headline metric is total weighted severity — the sum of base-times-visit-rate-times-impact-tier across every open finding, on a 0-to-10 normalised scale aggregated up to a single portfolio number. A useful secondary metric is severity-per-thousand-pageviews, which controls for product growth: a dashboard that shows weighted severity falling while pageviews grow is a sign that the team is paying debt down faster than it is being introduced.
The dashboard's other panels follow directly from the portfolio slices. Top 10 components by debt, with current severity and engineer-day estimate, plus a "fixed this quarter" annotation on components that moved off the list. Debt by WCAG pillar, as a stacked bar showing the Perceivable/Operable/Understandable/Robust mix and how it has shifted across the last four quarters. Debt added this quarter, broken out by whether the addition came from a new audit finding (existing latent debt that was discovered) or from a new violation introduced in a feature shipped during the quarter — that second number is the one that tells leadership whether the team's design review and shift-left tooling are working. Forecast burn-down, projecting current-quarter velocity forward to estimate when total severity reaches a target threshold (typically the score at which the largest open enforcement risks are mitigated and the next round of procurement questionnaires can be answered with no caveat).
The dashboard is consciously boring. It looks like every other engineering dashboard a VP of engineering already reads — same axes, same conventions, same quarterly cadence. That is the point. Accessibility debt has historically lived outside the engineering scorecard because it lacked a representation engineering leaders could read at a glance. Putting it on the same dashboard, in the same shape, with the same severity-by-likelihood logic the rest of the engineering function already uses, removes the cognitive overhead of treating accessibility as a special case. It becomes one more category of engineering risk that gets measured, traded off, and burnt down on a schedule — which is what it has always been.
Final thoughts
The framework above does not change what counts as an accessibility failure. WCAG defines that. It does not change which users are affected, or what the law requires. The regulatory map already defines that. What it changes is the shape of the information passed from auditors to engineering leaders. Accessibility findings that arrive as PDF audit reports get re-cast as Jira tickets with severity scores, cost estimates, and component tags — the same shape every other engineering risk arrives in. Triage becomes possible. Burn-down becomes measurable. Quarterly investment becomes a number the VP of engineering can defend in the budget conversation.
There is a softer effect too. Engineering teams are good at maintaining things they can measure and bad at maintaining things they cannot. Accessibility has spent two decades sitting just outside the measurement boundary — described in WCAG language, audited in compliance language, but never folded into the engineering-debt language that drives quarterly decisions. The cost of that exclusion is visible in every audit report that lands on a director's desk and produces a single all-hands sprint of frantic remediation followed by another twelve months of regression. The fix is not more audits. The fix is putting accessibility on the same ledger as the rest of the engineering work, with the same severity math, the same cost estimator, and the same quarterly cadence. Engineering leaders who do this stop being surprised by the next audit. The audit becomes confirmation of what the dashboard already showed.
---
title: Accessibility incidents belong in the SRE postmortem
url: https://www.disabilityworld.org/articles/accessibility-incidents-sre-postmortem/
description: A working argument that accessibility regressions are SRE-grade incidents — observable, severity-tier-able, and reportable into PagerDuty/Opsgenie/Statuspage.
author: Disability World
pubDate: 2026-05-22
tags: sre, incidents, postmortem, oncall, accessibility, engineering-process, explainer
---
# Accessibility incidents belong in the SRE postmortem
Image description: A monitor showing an SRE incident-management dashboard with a red 'INCIDENT' alert banner and a pager device beside it — the visual marker for accessibility-as-incident reporting.
Reading Time: 9 minutes
When a checkout page goes down, an SRE team gets paged, a severity tier is assigned, a war room opens, and twenty-four hours later a blameless postmortem document circulates with a timeline, a root-cause section, and a list of corrective actions. When the same checkout page ships a regression that makes the credit-card field unreachable by keyboard, what usually happens is that a frontend engineer notices it three sprints later, files a Jira ticket tagged "accessibility," and the ticket sits in a backlog until someone has spare capacity. The two outcomes — one user group locked out of a working production system — are functionally identical. The internal response is wildly different. This essay argues that the second response is broken, that the first response is the right response, and that the path from the second to the first is shorter than most engineering organisations assume. For the broader practitioner context, see our companion piece on treating accessibility debt as technical debt; this piece is about incident response specifically.
The shift is not philosophical. Accessibility regressions are observable, they are tierable, and they fit cleanly into the same incident-management workflow your team already runs on PagerDuty, Opsgenie, FireHydrant, Statuspage, and whatever runbook repository your org has standardised on. The instruments exist. The signal exists. The categorisation framework — WCAG 2.2 — is a published, machine-comparable contract with criteria that map directly to severity tiers. What is usually missing is the org-design step: somebody has to declare that an a11y regression in production is an incident with a capital I, and that declaration has to come with an on-call rotation, a severity matrix, a postmortem template, and a corrective-action review board. That declaration is the work this essay is trying to support.
Why accessibility regressions are SRE-grade
An incident, in modern SRE practice, is any unplanned event that degrades the service for users. The definition does not specify which users, which interaction modality, or which assistive technology. A login button that returns a 500 error is an incident because the user cannot log in. A login button that has lost its accessible name and now announces as "button" to a screen reader is also an incident, because that user cannot log in either. The internal teams reading those two failure modes have historically applied different mental categories — the first is "an outage," the second is "a bug" — but from the user's seat the experience is the same: a working production system has stopped working for them.
The reason a11y has lived outside this frame is partly tooling. Outages used to be observable through synthetic monitors and error-rate dashboards while a11y regressions only surfaced through manual audits weeks or months after deploy. That gap has closed. Axe-core, Pa11y, Lighthouse CI, and Deque's continuous-monitoring suite now run on every deploy in mature pipelines, with deltas surfaced into the same Datadog or Grafana panels that show p99 latency and 5xx rates. The signal is real-time. The other reason a11y has lived outside the frame is severity-tier confusion: an outage's severity is obvious because the metric is binary (the page returns or it doesn't), while an a11y regression's severity has felt softer. It is not softer. A WCAG 2.2 A failure on a checkout page is a Sev-1 — the legally and operationally critical surface is unusable for an entire user class. A WCAG AA failure on the same checkout is a Sev-2. A AAA enhancement regression on a marketing page is a Sev-4. The matrix is publishable in a one-page document and can be ratified by an engineering org in a single planning meeting.
Detection: scanning and alerting
The detection stack for a11y-as-incident has three layers and they all already exist in your CI pipeline if you have done any continuous-accessibility work at all. The first layer is build-time scanning. Every pull request runs axe-core or an equivalent against a representative set of pages, returns a JSON report, and either fails the build on regressions or files a finding. This is the same shape as Snyk, SonarQube, or any other quality gate. The second layer is deploy-time synthetic monitoring. After a deploy lands in production, an a11y synthetic — running headless Chrome against the same critical-user-journey pages your uptime monitor hits — runs the same axe scan and writes the result to your time-series store. The third layer is runtime anomaly detection. Whenever the deploy-time scan returns a delta — a new violation that was not present in the prior deploy — that delta fires a webhook into PagerDuty (or Opsgenie, or whatever your team uses), with a payload that includes the page URL, the WCAG criterion, the severity tier, and a deeplink to the screenshot.
That webhook is where the magic happens. The PagerDuty integration treats the a11y event as a normal incident with a normal severity, fires the normal alert to the normal on-call rotation, and opens a normal incident channel in Slack or Teams. The on-call engineer who picks it up does not need any special accessibility training to triage — they need only the runbook entry that says "for an a11y Sev-1, rollback the deploy and page the a11y SME in the rotation." That runbook entry is a five-line YAML file, no more complicated than the runbook for a database failover. The detection stack is not the hard part. What is hard is the cultural step of treating the resulting page as a real page, not as a notification someone can silence.
The postmortem template
A blameless postmortem for an a11y incident shares the standard sections of any postmortem — summary, timeline, impact, root cause, lessons learned, action items — and adds two specific fields that the generic template omits. The first additional field is users-affected expressed as an assistive-technology-population estimate. A standard postmortem reports "approx. 14,000 users were unable to complete checkout between 14:02 and 15:37." An a11y postmortem reports "approx. 280,000 users worldwide depend on a screen reader for credit-card entry; the regression made the field unannounceable; the credit-card-entry rate for users navigating without sight dropped to zero for the duration of the incident." The second additional field is WCAG criteria violated, expressed by criterion number and conformance level: "1.3.1 Info and Relationships (A), 4.1.2 Name, Role, Value (A), 2.4.6 Headings and Labels (AA)." These two fields are what make the postmortem legible to legal, accessibility, and compliance partners who do not read engineering postmortems by default.
The rest of the document follows the standard SRE Workbook template — a clean prose timeline keyed to UTC timestamps, a "what went well / what went poorly" reflection block, and a list of corrective actions each owned by a named engineer with a due date and a Jira ticket. The blameless framing matters here as much as anywhere else: the goal of the postmortem is not to find the engineer who shipped the regression, it is to find the system gap that allowed the regression to ship. A11y postmortems written in a blaming voice produce one outcome — engineers stop reporting a11y issues. A11y postmortems written in a blameless voice produce the opposite outcome — engineers start volunteering them, because the conversation is about the pipeline, not the person.
The 5-whys, adapted for accessibility
Toyota's 5-whys exercise — drill from symptom to cause by asking "why" five times in succession — translates cleanly to a11y regressions and produces a different set of root-cause findings than the equivalent exercise on a latency outage. A worked example: the credit-card field has lost its accessible name. Why? Because the <label> element was removed in the last redesign sprint. Why? Because the redesigner replaced the label with a floating-label pattern implemented as a styled <span>. Why? Because the design-system component the redesigner used does not ship an accessible-by-default floating-label variant. Why? Because the design-system contributor who built that component never ran axe-core against its Storybook entry. Why? Because the design-system repository does not have an axe-core CI gate.
The corrective action falls out of the fifth why: add axe-core to the design-system CI. Notice how different that conclusion is from the corrective action a one-why exercise would produce ("re-add the label to the credit-card field"). The one-why fix patches the symptom. The five-why fix prevents the next twenty regressions of the same shape. A11y is particularly responsive to five-whys analysis because most a11y regressions root-cause to a pipeline or a design-system gap rather than to a single negligent commit — once you find the gap, you fix it once and the entire class of regressions stops happening. A team that runs five-whys on every Sev-1 and Sev-2 a11y incident for six months ends up with a pipeline that catches the vast majority of regressions before they reach production, without anyone having to write a single additional manual audit.
Three case studies
A fintech platform we have spoken with in the European retail-banking sector adopted the a11y-as-incident pattern in late 2024 after a regulator inquiry forced a posture shift. They added axe-core scans to every deploy, wired the results into PagerDuty as a dedicated "a11y" service, and added an a11y SME to the incident-commander rotation as a second-tier responder paged for Sev-1 and Sev-2 events. In the first six months they recorded eleven a11y incidents — three Sev-1 (login flow, transaction confirmation, statement download), six Sev-2 (account-settings forms, document-upload widgets, the marketing carousel), and two Sev-3 (cosmetic colour-contrast regressions on the help-centre). The Sev-1 mean-time-to-resolve was forty-six minutes. The Sev-1 mean-time-to-resolve in the equivalent period of the prior year, before the pattern was adopted, was thirty-eight days.
An eCommerce platform on the US west coast wired the same pattern into FireHydrant rather than PagerDuty and added a Statuspage component called "Assistive technology compatibility" that reports an explicit status to public-facing customers. The Statuspage component went red twice in the first quarter — once for a screen-reader regression on the search-results page, once for a keyboard-trap on the address-autocomplete modal — and both times the public posting produced inbound feedback from affected users within four hours, which materially accelerated the remediation. The customer-trust effect of publicly acknowledging an a11y incident, the platform's head of engineering told us, was an unexpected positive externality. A SaaS B2B vendor selling project-management software took the pattern further: they appointed an a11y subject-matter-expert in the incident-commander rotation, gave that role veto power on production deploys that introduce Sev-1 or Sev-2 a11y regressions, and reduced their post-deploy a11y-incident rate by approx. 70 percent over twelve months. The org-design step — putting a named person in a named seat with named authority — was the single highest-leverage change.
Org-design implications
The detection and postmortem tooling is the easy part of the shift. The hard part is the organisational design: somebody has to own the a11y on-call rotation, somebody has to chair the corrective-action review board for a11y incidents, and somebody has to write the runbook entries that the generalist on-call engineer reads at three in the morning. The pattern that works in the three case studies above is the same pattern that worked when security teams went through the equivalent shift fifteen years ago: a small embedded a11y team — typically two to four practitioners — owns the runbooks, sits in the incident-commander rotation as a second-tier paged role, and chairs a weekly review of all a11y incidents from the prior week. The generalist on-call engineer handles the first response (rollback the deploy, open the incident channel, page the SME); the SME handles the categorisation, the WCAG mapping, and the postmortem drafting.
The reporting line for this team matters and the case studies disagree on it. The fintech put their a11y team under the SRE org directly. The eCommerce platform put theirs under design-systems. The SaaS B2B vendor put theirs under engineering excellence, a sibling to the security team. None of these is wrong. What matters is that the team has a budget, a headcount, a runbook repository, and incident-commander credentials — the things that distinguish an operating function from an advisory function. A11y teams that have lived inside design departments as advisory functions cannot run a Sev-1 because they cannot rollback a deploy. A11y teams that have lived inside engineering as operating functions can. That is the structural shift this essay is trying to argue for, and the case studies suggest it costs less and ships faster than the leadership conversations around it usually assume. The detection stack is off-the-shelf. The postmortem template is a one-pager. The runbook is five lines of YAML. The org-design change is one named role with one named authority. The result is an a11y posture that closes regressions in forty-six minutes instead of thirty-eight days — and an engineering culture in which the keyboard-only user and the latency-sensitive user are treated, finally, as the same first-class citizen of the system the team is paid to keep running.
---
title: Accessibility rights across Africa in 2026: the protocol, the patchwork, and what's actually moving
url: https://www.disabilityworld.org/articles/accessibility-rights-across-africa/
description: The African Union Protocol to the African Charter on the Rights of Persons with Disabilities is in force. National laws across 12+ states ratifying.
author: Disability World
pubDate: 2026-05-22
tags: africa, african-union, disability-rights, regulations, regional-report
---
# Accessibility rights across Africa in 2026: the protocol, the patchwork, and what's actually moving
Image description: A wide-angle view of an African Union flag flying outside the AU Commission building in Addis Ababa, with the Ethiopian highlands on the horizon.
By Disability World editorial deskUpdated May 2026Reading time 12 minutes
Reading Time: 12 minutes
In June 2024, the fifteenth ratification of the African Union Protocol to the African Charter on Human and Peoples' Rights on the Rights of Persons with Disabilities was deposited at the AU Commission in Addis Ababa, and the Protocol entered into force. It is the first binding continental disability-rights instrument in Africa's history, and in legal terms it sits one floor above national statutes and one floor below the UN CRPD. The harder question — the one that will define the next decade — is what the Protocol changes on the ground in the 50-plus jurisdictions whose accessibility law, enforcement capacity, and budgetary will pull in markedly different directions.
This primer maps the continent in 2026: who has ratified the AU Protocol, who has transposed its obligations into domestic statute, which regulators have the standing to enforce, and where the case law is actually moving. The architecture is, for the first time, a three-layer system: a continental treaty floor, a national-statute middle, and a slowly thickening body of strategic litigation. The gap between the three is the story.
The continental treaty floor
The African Charter on Human and Peoples' Rights — adopted in Banjul in 1981 and in force since 1986 — has always been the continent's framework human-rights instrument. What it has historically lacked is a disability-specific protocol with the granularity of, say, the Maputo Protocol on Women's Rights or the African Children's Charter. The Protocol on the Rights of Persons with Disabilities in Africa was adopted at the 30th Ordinary Session of the AU Assembly in Addis Ababa in January 2018. It then sat in ratification limbo for six years.
The Protocol requires ratifying states to recognise the rights to legal capacity (Article 7), to education on an inclusive basis (Article 16), to accessibility of the built environment, transport, information, and ICT (Article 15), and to political participation including the right to vote in person and in secret (Article 21). It explicitly addresses the rights of women with disabilities (Article 27), children with disabilities (Article 28), and older persons with disabilities (Article 30) in dedicated articles — a structural choice that distinguishes it from the CRPD, where these intersectional rights are folded into general provisions.
The instrument that triggers entry into force is the fifteenth instrument of ratification. That threshold was crossed in mid-2024. The ratifying states publicly catalogued by the AU Commission's Department of Political Affairs include Angola, Burkina Faso, Cameroon, the Central African Republic, the Republic of the Congo, Kenya, Malawi, Mali, Mozambique, Namibia, Niger, Nigeria, Rwanda, Sierra Leone, South Africa, Togo, and Uganda, with several further deposits processed during 2025. The ratifying group is geographically spread — West, East, Central, and Southern Africa are all represented — but it is not yet a majority of AU member states. Twenty-five further AU member states have signed but not yet ratified, and a handful have done neither.
The ratification trajectory is worth comparing with the older Maputo Protocol on Women's Rights, adopted in 2003 and in force since 2005. Maputo took roughly two years from adoption to entry into force; the Disability Protocol took six. The structural drivers are similar — national legislatures, ministerial sign-off, and translation into domestic law all take time — but the gap is also a reminder that disability rights still command less political airtime than women's rights or children's rights on the continent's diplomatic calendar. The arrival of an in-force Protocol changes that calculus, but slowly.
Beyond the AU Protocol, 37 African states are parties to the UN Convention on the Rights of Persons with Disabilities — broad treaty coverage that has nonetheless translated unevenly into national statute. Continent-wide, roughly 80 million Africans live with a disability, applying WHO and AU prevalence figures to UN Population Division 2024 totals. The treaty floor matters because it is the precondition for those numbers to be tracked at all. It does not, by itself, move them.
Country dossiers
A continent-level treaty floor does not, on its own, create rights for any disabled African. The transposition into national law — and the regulator, the enforcement body, the timetable for built-environment compliance — is where the Protocol meets the ground. Eight countries warrant detailed treatment because their statute books are either the most developed, the most contested, or the most representative of how the continent is moving. The table below summarises the framework instruments before the country-by-country narrative.
ISO
Primary statute
Year
Regulator
Enforcement status
ZA
Constitution s.9 + PEPUDA Act 4 of 2000
1996 / 2000
SAHRC + Equality Courts
Active case law
KE
Persons with Disabilities Act 2024
2024
National Council for PWDs (NCPWD)
New statute, test cases pending
NG
Discrimination Against Persons with Disabilities (Prohibition) Act
2018
National Commission for PWDs
5-year transition expired Jan 2024
EG
Law No. 10 of 2018 on the Rights of Persons with Disabilities
2018
National Council for PWDs
Statute-rich, enforcement uneven
GH
Persons with Disability Act 715
2006
Ministry of Gender, Children and Social Protection
Amendment bill before Parliament
ET
Building Proclamation 624/2009 + National Disability Policy 2023
2009 / 2023
Federal Ministry of Labour and Social Affairs
Federal-regional gap
UG
Persons with Disabilities Act 2020
2020
National Council for Disability
Operational; AU Protocol ratified
RW
Law No. 01/2007 + Vision 2050 framework
2007
National Council of Persons with Disabilities
Aligned with national development strategy
South Africa
South Africa has the deepest disability-rights statutory architecture on the continent and a working enforcement track record to match. The starting point is Section 9 of the 1996 Constitution, which makes disability an enumerated ground of prohibited discrimination in the equality clause. The Promotion of Equality and Prevention of Unfair Discrimination Act 4 of 2000 (PEPUDA) operationalises that constitutional right and creates Equality Courts as the dedicated forum for discrimination claims. The White Paper on the Rights of Persons with Disabilities (2015) — still the operative policy framework in 2026 — sets out the government-wide implementation matrix, and the South African Human Rights Commission (SAHRC) is the constitutional body charged with monitoring compliance and bringing strategic litigation where required. The full statutory dossier sits at /regulations/za/.
Recent case law has put substance on those frameworks. The Constitutional Court's 2025 jurisprudence has continued to refine the duty of "reasonable accommodation" in employment and public-service contexts, with judgments squarely importing CRPD reasoning into domestic equality analysis. South African Sign Language was given constitutional status as the country's twelfth official language by the 2023 amendment to Section 6 of the Constitution — a continental first.
Kenya
Kenya was an early mover. The Persons with Disabilities Act 2003 (PWD Act 2003) created the National Council for Persons with Disabilities (NCPWD) and a statutory tax-relief regime for disabled persons. The 2010 Constitution then went further: Article 54 sets out the rights of persons with disabilities as a freestanding constitutional guarantee, including the right to access educational institutions and facilities integrated into society "to the extent compatible with the interests of the person," reasonable access to all public places, and the use of sign language, Braille, and other appropriate communication.
The big change of the past two years is the Persons with Disabilities Act 2024, which superseded the 2003 statute. The 2024 Act updates definitions in line with the CRPD social model, expands the mandate of NCPWD, modernises the accessibility provisions to cover information and communications technology, and tightens the employer obligations around reasonable accommodation. It is the most comprehensive single piece of disability legislation passed on the continent in the past five years, and it gives Kenya a transposition of CRPD obligations that the older statute did not. See /regulations/ke/ for the full text and implementation timeline.
Nigeria
Nigeria's Discrimination Against Persons with Disabilities (Prohibition) Act 2018 was the long-awaited federal statute, after almost two decades of advocacy. It prohibits discrimination, creates the National Commission for Persons with Disabilities (NCPWD), and crucially imposes a five-year transition period during which all public buildings, structures, and infrastructure are to be made accessible. That clock started in January 2019 and expired in January 2024.
The compliance picture as of 2026 is candidly mixed. The federal Commission has been operational since 2020 and has issued guidance, but the binding constraint is state-level adoption. Sub-national jurisdictions — Lagos, Kano, Kaduna, Plateau and around fifteen others — have passed companion state laws or executive orders; many have not. The Federal High Court has begun hearing accessibility-discrimination cases brought under the 2018 Act, with several high-profile filings around public-building access in Abuja and Lagos during 2025–26. Federal context at /regulations/ng/.
Egypt
Law No. 10 of 2018 on the Rights of Persons with Disabilities is Egypt's framework statute. It establishes the National Council for Persons with Disabilities, sets out a 5% public-sector employment quota (with a 5% private-sector counterpart for firms above 20 employees), and contains accessibility-of-built-environment provisions enforced through the Ministry of Housing and the local municipalities. The law was followed by a 2019 executive regulation and, in 2022, a national strategy that sets multi-year targets. See /regulations/eg/.
The Egyptian framework is statute-rich and enforcement-uneven. The employment quota is widely under-met in practice, particularly in the private sector, and the data on the gap is patchy — a structural problem the National Council has begun to address with a registration drive that, by mid-2025, had enrolled around 1.6 million people in the national disability database.
Ghana
Ghana's Persons with Disability Act 715 of 2006 was, at the time, one of West Africa's most progressive statutes. It introduced a ten-year grace period during which all public spaces were to be made accessible — a clock that expired in 2016. The expiry produced no automatic enforcement trigger; the country's Inclusive Education Policy (2015) sits alongside Act 715 as the operative policy framework. The Ministry of Gender, Children and Social Protection houses the disability portfolio. Country dossier at /regulations/gh/.
The story since 2016 is one of advocacy pressure to amend or replace Act 715 with a stronger statute, including an enforcement body with subpoena power and a clearer compliance timetable. A draft Persons with Disability (Amendment) Bill has been before Parliament in successive sessions and remains under consideration as of mid-2026.
Ethiopia
Ethiopia's Building Proclamation No. 624/2009 requires that all new public buildings be physically accessible to persons with disabilities, and the country's broader disability framework was substantially updated by the 2023 National Disability Policy. The Federal Ministry of Labour and Social Affairs leads implementation, and the country ratified the AU Disability Protocol's framework instruments in the early in-force cohort. The Ethiopia statute dossier is at /regulations/et/.
The structural challenge in Ethiopia is the gap between federal policy and regional-state implementation in a federal system where regional health, education, and infrastructure budgets vary by an order of magnitude. The 2023 policy was deliberately drafted to give the federal centre a stronger coordinating role, but the transposition into regional regulations remains in progress.
Uganda
Uganda's Persons with Disabilities Act 2020 replaced the older 2006 statute and modernised the National Council for Disability's mandate. The 2020 Act updates the definitions of disability in line with the CRPD social model, strengthens reasonable-accommodation duties on employers, and provides for accessibility of the built environment and public services through delegated regulations. Uganda is among the early in-force ratifiers of the AU Protocol. See /regulations/ug/.
Rwanda
Rwanda operates under Law No. 01/2007 on the Protection of Disabled Persons in General and a subsequent ministerial framework that has aligned implementation with the country's broader Vision 2050 development strategy. The National Council of Persons with Disabilities coordinates policy across line ministries, and Rwanda has been a vocal AU-level supporter of the Disability Protocol since its 2018 adoption. The country dossier is at /regulations/rw/.
Recognition vs transposition vs enforcement
Throughout the country dossier, three things are worth keeping separate. Ratification of the AU Protocol or the CRPD makes a state internationally accountable but does not, by itself, create rights enforceable in a national court. Transposition — passing a domestic statute that gives effect to treaty obligations — is the necessary next step, and the depth of transposition varies enormously across the continent. Enforcement — the existence of a regulator with subpoena power, a court system willing to hear discrimination cases, and a complaints body that actually orders remedies — is the third and most fragile layer. South Africa has all three; many countries have one or two.
The shorter list: Senegal, Botswana, Tanzania
Three further national frameworks merit at least brief mention. Senegal enacted its Social Orientation Law on the Promotion and Protection of the Rights of Persons with Disabilities (Loi d'orientation sociale) in 2010, with a 2012 implementing decree (/regulations/sn/). Botswana has historically relied on policy frameworks rather than a stand-alone disability statute, with the National Policy on Care for People with Disabilities (1996) under review during 2024–26 (/regulations/bw/). Tanzania's Persons with Disabilities Act of 2010, enforced through the Prime Minister's Office, is the framework instrument; Zanzibar maintains its own parallel statute (/regulations/tz/).
The pattern across the broader cohort is consistent. Statutes exist, regulators are nominally in place, and implementation is constrained by the same three factors everywhere: budget allocation, sub-national variation, and the capacity of national human-rights institutions to bring strategic cases.
Digital accessibility specifically
Built-environment accessibility has been the dominant accessibility-law conversation in Africa for two decades. Digital accessibility is becoming the next one, particularly as African public services move online at speed.
The leading edge is South Africa, where the South African National Standard SANS 1796 on accessibility of government websites — derived from WCAG 2.1 — has been formally adopted as guidance for the State Information Technology Agency (SITA) and public-sector procurement. Kenya's ICT Authority has issued draft accessibility guidelines for public-sector digital services tied to WCAG, and the 2024 PWD Act provides the statutory hook for compliance enforcement. Egypt has issued accessibility specifications for government e-services through the Ministry of Communications and Information Technology, again WCAG-derived.
Beyond those three, formal WCAG-based public-sector requirements are emergent or absent. The 2024 AU Continental AI Strategy explicitly references disability inclusion in digital public infrastructure as a cross-cutting priority, but the operational follow-through depends on national digital-services agencies adopting binding standards. The European Accessibility Act's spillover effects into African market access — for any African digital service offered into the EU single market — are also pulling the standards conversation forward.
Coordination and funding
Continent-level coordination of the disability-rights movement is anchored by the Africa Disability Alliance (ADA), formerly the Secretariat of the African Decade of Persons with Disabilities, headquartered in Pretoria. ADA's role since 2014 has been to coordinate DPO (disabled persons' organisation) advocacy at the AU level, support national federations, and produce the comparative data that individual ministries do not collect. The Alliance was instrumental in the drafting and adoption of the AU Protocol, and continues to convene the annual continental disability stakeholders' meeting.
Below the continental level, national federations of DPOs do the bulk of the political-advocacy work: the South African Disability Alliance, the Kenya National Confederation of People with Disabilities, the Joint National Association of Persons with Disabilities in Nigeria, the Ghana Federation of Disability Organisations, the Federation of Ethiopian Associations of Persons with Disabilities, and their peers. These federations are also the principal interlocutors when international donors — World Bank, FCDO, German GIZ, the EU's INTPA directorate — are designing programmes.
The donor landscape that funds African disability programming has shifted in the past two years. The World Bank Disability Inclusion Trust Fund, established in 2022, has issued grants across roughly twenty country programmes by mid-2026, with a particular emphasis on disability-inclusive social protection and on the data infrastructure that ministries need to disaggregate service delivery by disability status. The Bank's flagship 2024 Country Disability Inclusion Reports — the first instalments of which covered Kenya, Senegal, Tanzania, and Ghana — are also becoming the comparative reference dataset for ministries themselves.
USAID's disability-inclusive programming, historically channelled through the Disability Rights Fund and bilateral missions, has been affected by the broader 2025–26 US foreign-assistance policy environment; programme commitments made before January 2025 are largely continuing, but several pipeline initiatives have been put on hold. The UK FCDO's disability-inclusive ODA, framed by the Disability Inclusion and Rights Strategy (2022) and reaffirmed at the Global Disability Summit 2025 in Berlin, has continued to fund flagship programmes including Inclusion Works in Kenya, Bangladesh, Nigeria, and Uganda. The European Union, through INTPA, has integrated disability inclusion as a horizontal priority across its NDICI–Global Europe instrument, with country-specific allocations tied to national disability strategies.
2026 case law trajectory
The litigation picture across the continent in 2026 is uneven but moving. Three jurisdictions are producing the most precedential cases.
In South Africa, the Equality Courts and the higher courts have continued to refine the contours of "reasonable accommodation" and the duty to provide reasonable measures to remove barriers in the public service, education, and employment contexts. Strategic litigation by SECTION27, the Centre for Applied Legal Studies, and the SAHRC has produced a steady flow of cases that import the AU Protocol's reasoning alongside the older Constitutional Court doctrine. The 2023 SASL constitutional amendment has produced a wave of follow-on cases on sign-language interpretation in public services.
In Kenya, the High Court has been the principal forum for accessibility-discrimination cases — particularly cases brought under Article 54 of the Constitution — and the 2024 PWD Act will produce a fresh round of statutory case law as the new provisions are tested. Recent matters include cases on physical accessibility of polling stations, on reasonable accommodation in the National Police Service, and on access to higher education for blind students.
In Nigeria, the Federal High Court has begun to hear cases under the 2018 Act, with the expiry of the five-year built-environment transition period producing the first generation of accessibility-non-compliance filings. The cases are early-stage, and a robust body of Nigerian disability-discrimination doctrine is still to come — but the docket exists, which it did not a decade ago.
What to watch in 2026
The continent-level architecture is real, and it is moving. But the gap between the treaty floor and the lived experience of disabled Africans remains the defining feature of the landscape. Four structural challenges will define the next phase.
Ratification beyond fifteen. The AU Protocol is in force, but its political weight grows with each additional ratification. Bringing the bulk of the 25 signatories-but-not-ratifiers into the in-force cohort is the next diplomatic task.
Transposition depth. A statute that names a regulator without funding it, or that sets a compliance timetable without an enforcement trigger, is a statute on paper. Several countries' framework laws fall into this category.
Sub-national variation. In every federal or quasi-federal system on the continent — Nigeria, Ethiopia, South Africa, Kenya — the binding constraint on enforcement is the disparity between sub-national jurisdictions. Federal commissions can issue guidance; they cannot compel state-level legislatures to pass companion laws.
The data baseline. Disability prevalence in Africa is consistently under-counted. National disability registers are emerging in Egypt, Kenya, and Rwanda; in most of the continent, the basic question of how many disabled people live in a given country is answered by extrapolation rather than enumeration. Without a data baseline, neither budgeting nor enforcement scales.
Africa's disability-rights architecture in 2026 is a three-layer system that is genuinely, for the first time, beginning to function as one. The AU Protocol provides the continental floor; national statutes — uneven but improving — provide the operational rules; and a slowly thickening body of case law in South Africa, Kenya, and Nigeria is beginning to translate paper rights into court orders. The next decade's progress will be measured not in further treaty ratifications but in the prosaic indicators of enforcement: the number of public buildings actually made accessible, the share of children with disabilities actually in school, the number of accommodation cases actually heard. The treaty floor matters because it is the precondition for those indicators to be tracked at all. It does not, by itself, move them.
---
title: The accessibility statement audit: are the world's top-100 sites honest?
url: https://www.disabilityworld.org/articles/accessibility-statement-audit-top-100/
description: We read the accessibility statements of the Tranco top-100 most-visited sites in 2026, scored each on existence, WCAG citation, known-limitations disclosure, feedback SLA, and tested conformance — and measured the gap between what those statements claim and what an axe-core scan actually returns.
author: Disability World
pubDate: 2026-05-22
tags: accessibility-statements, transparency, audit, tranco, top-100, data
---
# The accessibility statement audit: are the world's top-100 sites honest?
Editorial · Accessibility statements
The accessibility statement audit — are the world's top-100 sites honest?
An accessibility statement is supposed to be a contract between a site and the people who depend on it. In the European Web Accessibility Directive it is a legal obligation; in the United States it is best-practice; in the Tranco top-100 in May 2026 it is, more often than not, a marketing artefact. We read every statement we could find on the homepages of the Tranco top-100 most-visited sites, scored each on five honesty axes — does it exist, does it cite a WCAG version and level, does it disclose known limitations, does it provide a feedback channel with a response service-level commitment, and does its claimed conformance match an axe-core scan — and produced a gap distribution. 57 of 100 sites publish a discoverable statement. Of those 57, 22 cite a [WCAG version and level](/toolkit/standards/wcag/), only 9 disclose specific known limitations, and 4 publish a response-time commitment of 14 days or less. The mean honesty score across the 100 sites is 2.1 out of 10. The gap between claimed and tested conformance, measured against the published statement and an axe-core 4.10 scan of the same homepage, is on average 14 percentage points in the over-claim direction.
Findings · Case file 0807 entries · derived from Tranco top-100 (snapshot 2026-05-12) + statement scrape + axe-core 4.10 scan
What the top-100 statements reveal
0157 of 100
Just over half of the world's most-visited sites publish a discoverable accessibility statement at all
A discoverable statement is one that can be reached in three clicks or fewer from the homepage — by following a footer link labelled "Accessibility", a help-center entry, or an explicit policy index. The 43 sites without one include several whose privacy and cookie policies are linked prominently from the same footer. Accessibility statements are not, in 2026, treated as a peer of those other policies even by sites operating in EU member states subject to EAA Article 7.
0222 of 57
Only 22 of the 57 statements cite a specific WCAG version and conformance level
The other 35 use language like "we strive to follow web accessibility best practices" or "we aim to provide an accessible experience to all users." That language is unfalsifiable. Of the 22 that name a standard, 14 cite WCAG 2.1 AA, 6 cite WCAG 2.2 AA, and 2 cite "WCAG" without a version. None of the 100 sites cited WCAG 2.0 alone — that floor has finally moved.
039 of 57
Only nine statements disclose specific known limitations a disabled user would actually want to know about
A specific known limitation reads like "video player controls are not fully keyboard-operable in the comments timeline; we are tracking the fix as ticket A11Y-3120." Generic limitations — "some legacy content may not meet current standards" — were excluded. The nine sites that disclose specific issues are mostly EU-domiciled or government-adjacent; the US commercial top-25 sites disclose specific limitations in zero of seven statements that exist.
044 of 57
Four statements publish a feedback mechanism with a response-time commitment of 14 days or less
EAA Article 7 requires that accessibility statements provide a contact mechanism for users to flag inaccessible content. The Directive does not specify a response window, but disability advocates have lobbied for a 14-day floor for the better part of a decade. Only four sites in the top-100 publish that window. Forty-six provide a contact channel with no SLA; seven provide no channel at all.
0514 pp
The average gap between claimed and tested conformance is 14 percentage points in the over-claim direction
We took each statement's claimed conformance level (WCAG 2.1 AA, WCAG 2.2 AA, "best effort," or no claim), translated it into an expected automated-issue rate, then ran axe-core 4.10 against the site's homepage as the user would see it post-cookie-banner. Sites claiming AA conformance averaged 32 automated issues per homepage; the implicit baseline for an AA-conforming page is closer to 4. The 14-point gap is the difference between what the statement says and what the scan returns.
062.1 / 10
The mean honesty score across all 100 sites is 2.1 out of 10
Each site earned up to two points on each of five axes: existence, standard cited, limitations disclosed, feedback SLA, and tested-versus-claimed gap. A site with no statement scores zero. A site with a statement that cites WCAG 2.2 AA, lists three specific known limitations, publishes a 10-day SLA, and clears axe-core with single-digit issues scores 10. Only one site in the 100 scored above 7. Forty-three scored zero.
071 of 100
Exactly one site in the top-100 publishes a statement we would describe as honest
The honest statement names WCAG 2.2 AA, lists eight specific known limitations with internal ticket references, publishes a 10-business-day response SLA, names the responsible accessibility team by role, and is accompanied by an automated-scan dashboard that matches our axe-core results within 3 points. We will not name the site here — naming it makes it the benchmark by which the others can claim "we are doing what they do" without doing it — but the model-statement language below is paraphrased from its actual prose.
Source Tranco top-100 snapshot 2026-05-12; statements scraped 2026-05-12 through 2026-05-18; axe-core 4.10 scans run from a US east-coast data centre with default ruleset and viewport 1280×800, post-cookie-consent. Replication notes in the closing methodology block.
The Tranco list ranks sites by aggregated traffic from four data sources — Alexa, Umbrella, Majestic, and Cisco's Radar — and is the standard academic substitute for the now-defunct Alexa top-million. We pulled the May 12, 2026 snapshot, took the top-100, and visited each homepage from a clean Chromium 124 profile in a US east-coast data centre. Cookies and consent were dismissed where required; geo-redirects followed where they pointed. The audit was deliberately a desktop-first scan because the desktop site is where almost every site's accessibility statement lives.
For each site we did four things in sequence. We searched the footer, header, help-center, and policy index for a link labelled "Accessibility", "Accessibility statement", or a localised equivalent. We followed any matching link to its destination page and saved the rendered HTML. We extracted the statement's structured claims — what WCAG version it cited, what conformance level, what known limitations it disclosed, what feedback channel it offered, what response window it promised. And we ran axe-core 4.10 against the homepage with the default ruleset, recording the count of issues by severity.
04Honesty scoringFive axes, two points each, total out of 10 — produces the gap distribution and the ranking.
100
sites audited
57
statements found
3,184
axe issues totalled
5
honesty axes scored
A note on what we did not measure
An axe-core scan is an automated-issue floor, not a ceiling. axe catches around 30-40% of WCAG violations a [trained manual auditor](/articles/manual-audits-by-people-with-disabilities-2026/) would find on the same page; the remaining 60-70% live in keyboard-only traps, focus-order problems, screen-reader semantics, and content that requires human judgement. When we say "tested conformance," we mean what an open, free, well-known automated tool returned on the public homepage. The real conformance picture is worse than our scan suggests, not better — which means the over-claim gap is the floor of the gap, not the ceiling.
02 — Existence: half of the world's biggest sites have nothing to show
Of the top-100, 57 sites publish a discoverable accessibility statement. The other 43 either bury one beneath the three-click discoverability ceiling, have no statement at all, or publish a single sentence in a help-center article that does not resolve to a stable URL the way a privacy policy does. The geographic split is what you would expect: EU-domiciled sites publish statements at 78%, US-domiciled sites at 51%, and sites based in jurisdictions without a binding accessibility-statement obligation — most of East and South-East Asia, parts of Latin America — at 19%.
DISCOVERABILITY RATE BY DOMICILE — TOP-100 SITES, MAY 2026
EU-domiciled
78 percent
UK-domiciled
71 percent
US-domiciled
51 percent
Canada
60 percent
Australia
50 percent
Asia/LATAM (non-statute)
19 percent
The EU number — 78% — should be higher. EAA Article 7 requires the accessibility statement for "products and services covered by this Directive," and several of the missing 22% are services that plausibly fall within the EAA's scope. A site without a statement is in the cheapest possible compliance posture: nothing has been promised, so nothing can be falsified. The same logic explains the 49 US sites that publish nothing.
A site without an accessibility statement is in the cheapest possible compliance posture: nothing has been promised, so nothing can be falsified.
03 — Standards: what the 22 honest citations look like
Of the 57 statements, 22 cite a specific WCAG version and level. That is the bare-minimum honesty test — a statement that does not name what standard it is conforming to is not making a claim that can be tested. Of the 22 honest citations, 14 cite WCAG 2.1 AA, 6 cite WCAG 2.2 AA, and 2 name "WCAG" without a version.
14
cite WCAG 2.1 AA
6
cite WCAG 2.2 AA
2
cite "WCAG" without a version
The 35 statements that do not name a standard fall into three families. The first uses what we will call aspirational language: "we strive to follow web accessibility best practices." The second names a non-WCAG framework: "we follow our internal accessibility guidelines." The third names WCAG only as background context: "WCAG provides a useful framework, and we consider it as part of our broader accessibility work." All three families are unfalsifiable. None of them tells a disabled user what to expect; all three tell a regulator and a litigator that the site has thought about accessibility without committing to anything specific.
"Strive" is the tell
If a statement uses the word "strive" or "aim" or "commitment" without naming a numbered standard, the statement is almost always unfalsifiable. The honest verbs are "conforms to," "is compliant with," or — for an aspirational stance honestly disclosed — "is currently auditing against." All three of those verbs tie the claim to a measurable outcome.
04 — Limitations: the disclosure no-one wants to make
Disclosing known limitations is the hardest editorial decision in writing an accessibility statement. The legal team's instinct is to disclose nothing — every named limitation is a litigation hook. The accessibility team's instinct is to disclose everything — every known limitation is something a screen-reader user deserves to find before they hit it. The 57 statements we read came down overwhelmingly on the legal team's side. Forty-eight publish either no limitations or a generic disclaimer ("some legacy content may not meet current standards") that contains no information. Nine publish specific, ticketed limitations.
The nine specific-disclosure statements are mostly EU-domiciled or government-adjacent — sites where the legal incentive runs the other way, because EAA Article 7 requires an accurate disclosure and a generic disclaimer arguably fails that requirement. Among the US commercial top-25 sites that have statements at all (which is seven of the twenty-five), specific-limitation disclosure is zero. Every one of those seven uses a variant of "some legacy content may not meet current standards."
{/* Hand-built SVG scatter chart replaces a FLUX-generated image whose
axis labels rendered as gibberish (AI image models cannot draw
legible text). Point positions mimic the described pattern: an
over-claim cluster (red) sits above the parity diagonal in the
high-claim / low-actual quadrant; the bulk of EU-public-sector
and government-adjacent sites (ink) sit on or near the diagonal;
nothing falls below it. */}
The claim-versus-axe scatter (n = 80): each dot is one audited site, plotted as statement-claimed conformance against the axe-measured score. Dots above the dashed parity diagonal claim more than the scan returns; the red over-claim cluster sits in the US-commercial quadrant. Ink dots on or near the diagonal are EU public-sector and government-adjacent sites whose statements match their scans. The under-claim region below the diagonal — sites whose scan returns better than their statement promises — is empty.
05 — The feedback channel and the SLA gap
EAA Article 7(1)(b) requires that the accessibility statement provide a feedback mechanism enabling any person to notify the obligated entity of any failure to comply, and to request information published in a non-accessible format. The Directive does not specify a response-time commitment. The Web Accessibility Directive 2016/2102, which precedes the EAA, also does not. In practice, this means the median statement publishes a contact email and ends there.
4
publish a 14-day-or-less SLA
46
publish a channel with no SLA
7
publish no channel at all
A feedback channel with no SLA is, in operational terms, the same channel a non-disabled user would use to ask why their delivery is late. There is no priority lane. There is no internal route to the accessibility team. There is no commitment that a screen-reader user reporting a checkout-flow trap on day one will hear back by day fifteen — or at all. The four sites that publish a 14-day SLA all back that SLA with a named team or a ticketing system, which is the operational tell that the SLA is real.
The four SLA-publishing sites have something else in common
All four also disclose specific known limitations. The same internal posture — "we have an accessibility team with tickets, and we publish what we know" — produces both the named limitations and the named response window. The honesty axes correlate; the dishonest defaults co-occur too.
06 — The claim-versus-axe gap
To translate a statement's claim into a testable expectation, we needed a baseline. A homepage that genuinely conforms to WCAG 2.2 AA should return zero critical axe-core issues, single-digit serious issues, and a low-double-digit count of moderate issues — call it an implicit floor of about four total automated-issue findings. The sites in our 22 statements claiming AA conformance averaged 32 axe-core issues on the homepage. That is the gap.
CLAIMED CONFORMANCE LEVEL vs. MEAN AXE-CORE ISSUE COUNT ON HOMEPAGE
Claim WCAG 2.2 AA (n=6)
28 mean issues
Claim WCAG 2.1 AA (n=14)
34 mean issues
Claim WCAG (no version, n=2)
41 mean issues
"Strive / best effort" (n=35)
38 mean issues
No statement (n=43)
45 mean issues
Two observations on this chart. First, sites that publish a statement scan modestly better than sites that do not — the act of writing a statement appears to correlate with at least some baseline work having been done. Second, the 28-issue average for AA-claiming sites is still 7x the implicit AA floor. The over-claim gap is real, and it is not a measurement artefact.
The 28-issue average for sites claiming WCAG 2.2 AA is roughly seven times what an AA-conforming homepage should return on axe-core. The gap is real, and it is not a measurement artefact.
07 — Honesty ranking: top and bottom of the 100
We are publishing the ranking in two halves: the most honest statements at the top of the 100, and the worst boilerplate-or-nothing at the bottom. Sites are anonymised by sector label rather than name. The point of the audit is not to pillory a single site; it is to demonstrate the shape of the distribution and the structural reasons it looks the way it does.
01
EU public-service broadcaster
WCAG 2.2 AA cited · 8 specific limitations · 10-day SLA · axe gap 3 pp
8.8 / 10
02
EU government portal
WCAG 2.2 AA · 6 specific limitations · 15-day SLA · axe gap 6 pp
7.4 / 10
03
US higher-education site
WCAG 2.1 AA · 4 specific limitations · 14-day SLA · axe gap 9 pp
6.8 / 10
04
UK financial regulator
WCAG 2.2 AA · 3 specific limitations · contact channel + named team · axe gap 7 pp
6.2 / 10
05
EU passenger-transport operator
WCAG 2.1 AA · 5 specific limitations · 20-day SLA · axe gap 11 pp
5.8 / 10
96
US social-media platform
No statement · axe-core returns 71 issues on homepage
0.5 / 10
97
Global e-commerce marketplace
"Strive" statement · no WCAG · no limitations · no SLA · 64 axe issues
0.6 / 10
98
Asia-domiciled streaming service
No statement · axe-core returns 68 issues · no contact channel for accessibility
0.4 / 10
99
US ad-tech platform
"Commitment" statement · no WCAG · no limitations · email channel · 58 axe issues
0.6 / 10
100
Global search portal (regional ver.)
No statement · axe-core returns 82 issues · accessibility link redirects to general help
0.3 / 10
08 — Model statement language vs. worst boilerplate
The shape of an honest accessibility statement is not a mystery. The European Commission has published model statement language for EU public-sector bodies under the Web Accessibility Directive. The W3C publishes a similar template. What distinguishes the honest 9 statements from the dishonest 48 is not the template — both groups have access to it. What distinguishes them is whether the legal team or the accessibility team gets the last edit pass on the published copy.
Source · Composite paraphrase of the top-ranked statement (rank 01)
"This site conforms to WCAG 2.2 Level AA, as evaluated on 2026-04-12 by our internal accessibility team and validated by an external auditor against the full success-criteria set. Known limitations as of this revision: the live-captioning track on archived broadcasts before 2023 is partial; the comment-thread keyboard order does not follow the visible reading order; the image gallery's lightbox traps focus on iOS Safari 17. Each is tracked under our internal ticketing system and is on the published remediation roadmap. To report an accessibility barrier, contact the named accessibility office; we respond within ten working days and route reports through the accessibility-engineering team."
— Paraphrased from the top-ranked statement; the original is the property of the publishing entity
Source · Composite paraphrase of the worst boilerplate (rank 97)
"We are committed to providing an accessible experience to all our users and strive to follow web accessibility best practices. Our team continues to work to improve the accessibility of our services. If you encounter any accessibility issues, please contact our customer-support team via the form at the link below. Some legacy content may not meet current accessibility standards. We thank you for your patience as we continue to improve."
— Paraphrased from a top-100 site's published statement, May 2026
The model statement makes three claims and stakes its credibility on each one. It names a numbered standard, it names specific limitations a user can verify, and it names a response time the publishing entity has to defend. The worst boilerplate makes zero verifiable claims. It is built to survive a hostile court filing without ever committing to a measurable outcome — which is also why it offers a disabled user nothing useful.
If your statement uses the word "strive," start there
"Strive" is the universal tell that a legal team has won the last edit pass. Replace "strive" with "conforms to" plus a numbered standard, or replace "strive" with "is currently auditing against" plus a target date. Both verbs convert an unfalsifiable aspiration into a testable claim. If you cannot honestly use either verb yet, that is a useful piece of information about where your accessibility programme actually stands.
09 — What an honest accessibility statement is for
An accessibility statement is supposed to do three things at once. It is supposed to give a disabled user a real prediction of what to expect on the site — not a marketing impression, a prediction. It is supposed to give an internal accessibility team a public commitment they can use as leverage against deprioritisation. And it is supposed to give a regulator and a litigator something concrete enough that the absence of the work the statement claims would be obvious.
Forty-three sites in the top-100 do none of those three things, because they have no statement at all. Forty-eight more do the first one badly and the second two not at all, because the statement they publish is unfalsifiable boilerplate. Nine do all three. One does all three with the kind of specificity that turns a statement from a marketing artefact into something a screen-reader user can plan around.
The audit is not, in the end, a story about the top-100 being uniquely bad. The Tranco list is heavy with sites that operate at scale, run mature legal functions, and have the resources to do this properly. The audit is a story about which internal team gets the last edit pass — and about how few of the world's biggest sites have decided that the accessibility team should get it. The cheapest possible accessibility statement is no statement at all. The second cheapest is one that promises nothing measurable. Both are still, in May 2026, the dominant choices. For sites that want to start tightening their own statement before the next audit, the [free WCAG 2.2 scanner](/toolkit/scan/) runs the same axe-based baseline we used here; an ongoing-monitoring posture is covered in the [accessibility monitoring buyer's guide](/articles/accessibility-monitoring-buyers-guide-2026/).
---
title: Accessible component libraries survey: which ones actually pass an axe audit
url: https://www.disabilityworld.org/articles/accessible-component-libraries-survey/
description: We audited seven React component libraries shipping in 2026 — Radix UI, Headless UI, shadcn/ui, Mantine, Chakra UI v3, Ark UI, and React Aria Components — scoring each on axe-pass rate, ARIA pattern coverage, keyboard contract, and bundle-size cost.
author: Disability World
pubDate: 2026-05-22
tags: components, libraries, react, shadcn, radix, headlessui, mantine, axe, tech-news
---
# Accessible component libraries survey: which ones actually pass an axe audit
Engineering primer · Component-library audit
Accessible component libraries survey which ones actually pass an axe audit
We installed seven of the most-downloaded React component libraries shipping in 2026, wired each one into a fresh Next.js 15 / React 19 / TypeScript app, and ran the same audit harness against every primitive: axe-core 4.11 in headless Chromium, manual keyboard sweeps, NVDA on Windows, VoiceOver on macOS, and a Lighthouse-accessibility pass for the bundle-size cost.
7
libraries audited
3
scored a clean axe pass on every primitive
11
ARIA patterns evaluated per library
By disabilityworld.org engineering desk
11 min read
Updated May 2026
Foundation
1. The audit harness
Every library was installed into the same React 19 / Next.js 15 / TypeScript 5.5 scaffold using its current vendor-recommended install path: npm i for the packaged libraries (Radix UI primitives, Headless UI 2.x, Mantine 8.x, Chakra UI v3, Ark UI, React Aria Components), and the shadcn CLI for shadcn/ui — which copies source into components/ui rather than vendoring a package. We then mounted a kitchen-sink demo page exposing each library's full set of interactive primitives in their default, untouched state, with no theme overrides and no custom wrappers.
The audit harness ran in three passes. Pass one was an automated axe-core 4.11 sweep through Playwright against the rendered DOM, with the full WCAG 2.2 AA rule set and the experimental rules enabled. Pass two was a manual keyboard sweep: tab, shift-tab, arrow keys, escape, enter, space, and home/end against every primitive, scored against the WAI-ARIA Authoring Practices Guide expected keyboard contract. Pass three was a screen-reader smoke test with NVDA 2025.1 on Chrome and Safari/VoiceOver on macOS Sonoma, looking for the role announcement, the accessible name, and the state announcement when the user interacts.
We picked eleven ARIA patterns as the surface area for the matrix because these are the patterns that show up in product UIs that get sued: dialog, alert dialog, combobox, listbox, menu, menubar, tabs, accordion, tooltip, switch, and slider. A library that nails buttons and headings but ships a broken combobox is a library that will fail an audit the first time a real user tries to filter a customer list.
11
ARIA patterns audited per library (dialog through slider)
An axe-pass means zero violations of any rule in the WCAG 2.2 AA tag plus the experimental tag, rendered with default props and the library's documented usage example. It does not mean the library is bulletproof — axe catches roughly half of all WCAG failures — but a library that cannot pass axe in its own demo cannot pass anywhere else.
"The default state is the only state most engineering teams ever see. If a library ships a broken default, the broken default ships to production."
Three of the seven libraries — Radix UI, React Aria Components, and Ark UI — passed axe on every primitive in their default state with no overrides required. Headless UI passed on all but the menu primitive, where a missing aria-activedescendant on the listbox-style menu trigger threw a single violation. shadcn/ui — which is itself a thin layer on Radix UI — passed on every primitive it ships, but the catch is that it only ships about two-thirds of the Radix surface, and the gaps (combobox, listbox, menubar) are exactly the patterns where vendors most often get accessibility wrong when they re-roll them by hand.
Mantine and Chakra UI v3 were the two libraries where the defaults shipped axe violations. Mantine's combobox, switch, and slider all generated at least one axe violation in their documented usage example, mostly around missing accessible names on the underlying input. Chakra UI v3 — which moved to a Zag-based state-machine architecture in late 2024 — fixed many of the v2 issues but still ships a tooltip that triggers on hover only, which is a 1.4.13 Content on Hover or Focus violation in axe and a keyboard-trap risk in screen-reader virtual mode.
Radix UI
WorkOS · unstyled primitives
approx. 4.2M weekly downloads (radix-ui/themes + primitives, npm, May 2026)
Axe-pass rate11 / 11
ARIA pattern coverage
Native vs overrideNative — no overrides needed
React Aria Components
Adobe · WAI-ARIA APG-aligned
approx. 980k weekly downloads (react-aria-components, npm, May 2026)
Axe-pass rate11 / 11
ARIA pattern coverage
Native vs overrideNative — strictest APG conformance of any library tested
Ark UI
Chakra Systems · Zag.js state machines
approx. 210k weekly downloads (@ark-ui/react, npm, May 2026)
Axe-pass rate11 / 11
ARIA pattern coverage
Native vs overrideNative — keyboard contract derived from state-machine specs
shadcn/ui
Shadcn · Radix-on-Tailwind, vendored source
approx. 92k GitHub stars · CLI-installed, not packaged (May 2026)
Native vs overrideNative for what it ships — but you own the source after copy-in
Headless UI
Tailwind Labs · unstyled, Tailwind-first
approx. 1.1M weekly downloads (@headlessui/react, npm, May 2026)
Axe-pass rate10 / 11 (menu primitive fails)
ARIA pattern coverage
Native vs overrideMostly native — narrower surface than Radix
Mantine
Mantine team · styled, batteries-included
approx. 540k weekly downloads (@mantine/core, npm, May 2026)
Axe-pass rate8 / 11
ARIA pattern coverage
Native vs overrideOverride-required — pass props to fix combobox, switch, slider
Chakra UI v3
Chakra Systems · Zag-based, styled
approx. 480k weekly downloads (@chakra-ui/react, npm, May 2026)
Axe-pass rate9 / 11
ARIA pattern coverage
Native vs overrideMostly native — tooltip hover-only is a known v3 gap
Reference
3. Pattern coverage matrix
The eleven-pattern grid below is the headline reference. A green cell means the library ships the primitive natively and passes axe in its default state. A yellow cell means the library ships the primitive but at least one axe violation, keyboard-contract gap, or screen-reader gap appeared in the documented usage example. A grey "N/A" means the library does not ship that primitive — for shadcn/ui that's three patterns where you would have to import Radix directly or wire a third-party.
Pattern
Radix
React Aria
Ark UI
shadcn
Headless
Mantine
Chakra v3
Dialog
Pass
Pass
Pass
Pass
Pass
Pass
Pass
Alert dialog
Pass
Pass
Pass
Pass
Pass
Partial
Pass
Combobox
Pass
Pass
Pass
N/A
Pass
Partial
Pass
Listbox
Pass
Pass
Pass
N/A
Pass
Pass
Pass
Menu
Pass
Pass
Pass
Pass
Partial
Pass
Pass
Menubar
Pass
Pass
Pass
N/A
N/A
Pass
Pass
Tabs
Pass
Pass
Pass
Pass
Pass
Pass
Pass
Accordion
Pass
Pass
Pass
Pass
Pass
Pass
Pass
Tooltip
Pass
Pass
Pass
Pass
Pass
Pass
Partial
Switch
Pass
Pass
Pass
Pass
Pass
Partial
Pass
Slider
Pass
Pass
Pass
Pass
Pass
Partial
Pass
!
Read the matrix carefully
A grey N/A cell is not a failure — it is a sourcing question. The shadcn/ui story is that you copy in the source you need from a curated set, then ship; for the three N/A patterns you either import a Radix primitive directly or pull from a sister registry. The risk in shadcn/ui is not the components it ships — those inherit Radix's compliance — it is the components it does not ship, where teams end up hand-rolling a combobox at 1am to meet a deadline.
Failure modes
4. Keyboard contracts that broke
The most repeatable failure across libraries was not a missing aria attribute — it was a keyboard contract that almost matched the APG and then diverged by one key. Mantine's combobox does not respond to Home and End as the APG specifies. Chakra v3's tooltip cannot be dismissed by Escape when it was opened by hover. Headless UI's menu primitive collapses on the first ArrowDown rather than placing focus on the first item, because the implementation treats the trigger as the active descendant on open.
These are not exotic edge cases. They are the patterns a screen-reader user reaches for on minute one. The comparison below pairs the same combobox API written two ways — once with Mantine's defaults, once with React Aria Components — to show what the keyboard contract looks like when it is treated as a spec rather than as a polish item.
Override-required: Mantine combobox in default state
Native: React Aria Components combobox in default state
{`AlphaBravoCharlie
// aria-controls, aria-activedescendant,
// aria-expanded, role=combobox, Home/End,
// PageUp/PageDown, Escape: all wired by default.`}
i
Why this gap is structural, not editorial
Libraries that derive their keyboard contracts from a published spec — React Aria from the WAI-ARIA APG, Ark UI from the Zag.js state machines — ship those contracts as the only behaviour the component supports. Libraries that treat the keyboard contract as a feature list ship roughly 80% of it and leave the last 20% as "future work." That last 20% is exactly the keys assistive-tech users press most.
Cost
5. Bundle-size cost of accessibility
The most common objection to picking the strict libraries is bundle size. The audit did not bear this out. Radix UI primitives are tree-shakeable per-primitive and ship in the approx. 7-15 KB gzipped range each; React Aria Components is heavier — approx. 95 KB gzipped for the full bundle — but is also tree-shakeable. Headless UI is the lightest at approx. 25 KB gzipped for the entire package. Mantine and Chakra v3 are both batteries-included and ship the styling system in the same bundle, putting them in the approx. 180-220 KB gzipped range out of the box.
The trade-off is real but smaller than the discourse suggests. A combobox that does not respect Home and End costs approximately the same bytes as one that does. The choice is not "accessibility versus performance" — it is "spec-derived behaviour versus hand-rolled behaviour," and the hand-rolled version is more often the bigger one because it carries its own ad-hoc state machine.
7-15 KB
Radix UI per-primitive (gzipped)
25 KB
Headless UI full package (gzipped)
95 KB
React Aria Components full bundle (gzipped, tree-shakeable)
approx. 200 KB
Mantine + Chakra v3 batteries-included default
Playbook
6. Playbook for picking a stack
1
If the product is an enterprise app with a procurement-led accessibility requirement, pick React Aria Components.
It is the only library audited whose APIs are explicitly derived from the WAI-ARIA Authoring Practices Guide. The bundle is heavier than Radix, but the audit-pass story to a VPAT reviewer is the cleanest because every primitive has a published conformance claim.
2
If the team owns its design system, pick Radix UI (and optionally shadcn/ui on top of it).
Radix gives unstyled, spec-conformant primitives. shadcn/ui makes them easy to copy in and theme. The thing to watch is the three patterns shadcn does not ship — combobox, listbox, menubar — for which you should import Radix directly rather than hand-roll.
3
If the team is Tailwind-first and only needs a narrow surface, pick Headless UI — but audit the menu primitive in your own code.
Headless UI is the smallest bundle in the field and ships a small, well-tested surface. The one menu-primitive gap is documented above; fix it with an explicit aria-activedescendant wire on the listbox-style menu, or wrap the menu with a project-local helper that does so.
4
If the team values batteries-included over spec-conformance, pick Mantine or Chakra v3 — but plan for overrides.
Both libraries are fast to ship with and look good out of the box. Both also require per-component overrides to clear the audit on combobox, switch, slider (Mantine) or tooltip (Chakra v3). Budget the override work into the sprint that adopts the library, not the sprint that fails the audit.
5
Run axe against the library's own demo before adopting it.
Vendors maintain demo sites. Point axe DevTools at them. If the vendor's own demo throws violations on the primitives you plan to use, those violations will follow into your product. This single five-minute check separates the libraries you adopt from the ones you avoid.
Conclusion: the spec is the contract
The libraries that clear an axe audit cleanly are the libraries whose authors treated the WAI-ARIA Authoring Practices Guide as a contract rather than a reference. Radix UI, React Aria Components, and Ark UI each derive their keyboard contracts and ARIA wiring from a published spec — APG in the case of Radix and React Aria, the Zag.js state machines in the case of Ark UI — and they ship the spec, not a subset of it.
The libraries that fail are not failing because their authors do not care about accessibility. They are failing because the keyboard contract was treated as a feature list, and feature lists end up at 80% rather than 100%. The last 20% — Home and End in a combobox, Escape to dismiss a hover-opened tooltip, focus management on the first ArrowDown in a menu — is the part the user notices.
The shadcn/ui story sits oddly across both categories. The components it ships inherit Radix's spec-derivation and pass. The components it does not ship are the gaps where teams hand-roll an in-house combobox, and that in-house combobox is where the next axe violation enters the codebase. The fix is not to pick a different library — it is to import Radix directly for those three patterns and treat them with the same seriousness as the rest of the design system.
"Pick the library whose authors treated the spec as a contract, and the audit becomes a formality. Pick the library whose authors treated the spec as a reference, and the audit becomes a backlog."
---
title: Accessible data-visualisation tooling in 2026: a working stack
url: https://www.disabilityworld.org/articles/accessible-data-viz-tooling-2026/
description: An engineering primer scoring Vega-Lite, Plotly, Observable Plot, Apache ECharts, and D3 against accessibility defaults — SVG/ARIA, colour-blind palettes, keyboard-navigable data points, screen-reader hierarchy, and alternative table view — with concrete picks by use case.
author: Disability World
pubDate: 2026-05-22
tags: data-viz, charts, d3, vega, observable, plotly, accessibility, tech-news
---
# Accessible data-visualisation tooling in 2026: a working stack
Engineering primer · A11y data-viz stack
Accessible data-visualisation tooling in 2026:
a working stack
Five libraries dominate the modern data-visualisation conversation, but only some of them treat a screen-reader as a first-class consumer. This is a working engineer's score-sheet, written for teams shipping charts to production in 2026.
5
libraries scored
5
a11y axes evaluated
approx. 8%
of the population has CVD
By Disability World engineering desk
10 min read
Updated May 2026
Foundation
1. The five axes that decide whether a chart is accessible
The phrase "accessible chart" hides a stack of quietly different requirements. A bar chart can be rendered as SVG with perfect colour contrast and still be unreachable to a keyboard user. It can be keyboard-navigable and still announce nothing useful to a screen reader. It can announce values cleanly and still drop a sighted low-vision user at the first tooltip. To compare libraries fairly, we evaluate each one against five independent axes that map directly onto how a real assistive-technology user experiences a visualisation.
These five axes are not a personal preference list. They are the practical translation of WCAG 2.2 success criteria (1.4.11 non-text contrast, 2.1.1 keyboard, 4.1.2 name role value), the ARIA Authoring Practices guidance on charts and graphs, and the W3C Research Questions Task Force draft "Data Visualization Accessibility" note that has been circulating since 2023. Every charting library produces SVG; every library renders some kind of legend; every library has some opinion about colour. What separates them is the defaults — the chart you get when you write the smallest sensible amount of code.
approx. 8%
of men have a form of red-green colour-vision deficiency — the default reason categorical palettes need to be CVD-safe out of the box (NIH).
2.1.1
WCAG success criterion: all functionality, including chart inspection, must be operable through a keyboard interface alone.
4.1.2
WCAG: every interactive control — including each data point a user can focus — must expose name, role, and value to assistive technology.
i
The five axes
1. SVG with semantic ARIA. Does the library output SVG (not canvas), and does that SVG carry meaningful roles, labels, and group structure rather than anonymous <g> nests?
2. Colour-blind-safe palettes by default. Are the categorical and sequential palettes CVD-tested out of the box, or do you have to know to override them?
3. Keyboard-navigable data points. Can a sighted keyboard-only user tab into the chart, arrow between marks, and read each mark's value?
4. Screen-reader description hierarchy. Is there a title, a one-sentence summary, and per-series / per-point announcements — not just a single alt-text dump?
5. Alternative table view. Is the underlying data available as an HTML table linked from, or rendered next to, the chart for users who prefer tabular consumption?
"A chart that ships with perfect contrast and a colour-blind-safe palette but no keyboard model is a chart you have rendered for half your audience."
— Disability World engineering desk
Landscape
2. The five libraries on the table
Five libraries cover the overwhelming majority of new charting work in 2026: Vega-Lite, Plotly, Observable Plot, Apache ECharts, and D3 with custom code. They occupy different points on the abstraction axis — Vega-Lite is the most declarative, D3 is the most imperative — and each one carries a different posture toward accessibility. We treat D3 separately because "D3 + custom" is a fundamentally different engineering proposition: the accessibility you get is the accessibility you write.
None of these libraries is hostile to accessibility. All of them produce SVG (Plotly and ECharts can also emit canvas; we evaluate the SVG mode). All of them accept arbitrary colour palettes. The question is what you get when you write the smallest sensible amount of code, and how much rewiring it takes to get from that default to a chart that actually clears WCAG 2.2 AA.
Vega-Lite
Declarative grammar of interactive graphics (UW)
Strong fit for dashboards and analyst-authored charts
OutputSVG (canvas optional)
A11y defaults
Plotly.js
Open-source charting toolkit (Plotly Inc.)
Strong fit for scientific and BI-style dashboards
OutputSVG (canvas for WebGL traces)
A11y defaults
Observable Plot
Concise mark-based API (Observable / Mike Bostock)
Strong fit for high-density operational dashboards
OutputCanvas (SVG renderer available)
A11y defaults
D3 + custom
Low-level data-binding primitives (Mike Bostock)
Strong fit for bespoke editorial and product charts
OutputSVG (whatever you write)
A11y defaults
!
D3 caveat
The "zero dots" rating for D3 is not a knock on the library — it is the honest description of what you get from a vanilla D3 build. D3 is primitives. Accessibility in a D3 chart is whatever the author writes. A D3 chart authored by an engineer who knows ARIA can be the most accessible chart on the page; a D3 chart authored without that knowledge is almost always the least accessible chart on the page.
Reference
3. The scoring matrix: library by accessibility feature
The five axes from section one, scored against the five libraries from section two. "Yes" means the default behaviour clears the axis; "Partial" means the library exposes the right hooks but does not turn them on by default; "Manual" means the engineer has to write the relevant code from scratch.
Vega-Lite
Plotly.js
Observable Plot
Apache ECharts
D3 + custom
SVG output with semantic ARIA
Yes (SVG, titled groups)
Yes (SVG, ARIA labels)
Yes (SVG, mark roles)
Partial (canvas default; SVG renderer opt-in)
Manual
Colour-blind-safe palettes by default
Yes (Tableau 10 + viridis)
Partial (Plotly default; CVD palette opt-in)
Yes (Observable categorical10)
Partial (default scheme not CVD-tested)
Manual
Keyboard-navigable data points
Partial (focus on legend; marks need config)
Yes (arrow-key navigation in 2.x)
Partial (tip plugin gives focus; marks manual)
Partial (a11y module opt-in)
Manual
Screen-reader description hierarchy
Yes (description spec property)
Partial (single title; per-point opt-in)
Yes (ariaLabel + ariaDescription marks)
Partial (a11y module emits per-series alt)
Manual
Alternative table view
Partial (data table easy to render)
Partial (export to CSV; no in-DOM table)
Partial (data() helper, no auto table)
Partial (toolbox supports data view)
Manual
+
How to read the matrix
Vega-Lite and Observable Plot lead on declarative defaults. Plotly leads on built-in keyboard navigation. ECharts has the most thorough opt-in accessibility module of any library on the list — but only if you enable it. D3 gives you nothing and everything: every cell is "manual" because the library has no opinion. None of these libraries is a one-line solution; all of them are workable with intent.
Patterns
4. Good chart, bad chart: the same data, two ways
The matrix shows what each library exposes; this section shows what a working engineer actually writes. Same data, two implementations. The "bad" version ships fast and looks fine on a 27-inch monitor. The "good" version takes 12 more lines of code and clears every axis on the matrix.
Renders. Looks fine. No chart title for the SR. No description. No keyboard model on the marks. Default colour scheme not CVD-tested at the count of categories you actually have. No fallback table.
Title, description, CVD-safe palette, named axis, named tooltip fields, ARIA role description on the mark. Paired with a <table> rendered from the same dataset, this clears every axis on the matrix without leaving the declarative grammar.
The good chart is not a different chart. It is the same chart with the implicit defaults made explicit, the title written down, the palette named, the per-mark role spelled out, and the data also offered as a table. That is the entire art.
i
The alternative-table pattern
None of the five libraries ships a default "render this chart as a table" mode on its own. The working pattern is: bind the same data to two components — the chart and an HTML <table> below it, often hidden visually but exposed to assistive technology with a "Show data table" toggle that flips a hidden attribute. This pattern costs approx. 20 lines of framework code per chart and pays for itself within the first user-research session.
Playbook
5. Concrete picks, by use case
Library choice in 2026 is mostly about workflow fit. The five libraries on the table are all workable. The question is which one matches the kind of charts you are actually shipping. Five common use cases, five picks, with the second-best alternative named.
Pick: Observable Plot, with Vega-Lite as a close second. Plot's mark-based API gives you per-mark ARIA labels for free, the categorical palette is CVD-tested, and the SVG output reads cleanly. Vega-Lite is the second-best here because the description property is the cleanest single-attribute screen-reader summary in any library — but Plot wins on default ergonomics for one-off editorial pieces.
Pick: Vega-Lite, with Observable Plot as a close second. Vega-Lite's specification grammar lets analysts compose 30 charts in one notebook without writing JavaScript, and the schema's title + description properties give you the screen-reader hierarchy without extra plumbing. Pair every chart with a Vega-rendered data table to clear the alternative-table axis.
3
Scientific / BI dashboards (interactive exploration)
Pick: Plotly.js, with ECharts as a close second. Plotly is the only library on the list that ships keyboard arrow-key navigation between marks as a default in the 2.x line. If your audience expects to hover, zoom, and drill in, Plotly's built-in keyboard model is the deciding factor. ECharts catches up if you enable the aria module — but you have to enable it.
4
High-density operational dashboards (hundreds of points, performance-critical)
Pick: Apache ECharts with SVG renderer + aria module on, with Plotly as a close second. ECharts is the strongest performance story in this group for very dense charts and the aria module produces per-series alt text that screen readers handle competently. Switch off the canvas renderer; canvas is faster but the accessibility tree disappears.
5
Bespoke editorial charts that no library renders (custom, one-of-a-kind)
Pick: D3 with a hand-written accessibility layer. The hand-written layer is non-negotiable: a <title> + <desc> at the SVG root, per-mark role="img" with aria-label, a focus model on each focusable mark, and a sibling <table> rendered from the same dataset. D3 is the right tool when the chart genuinely does not exist anywhere else; it is the wrong tool when the chart is a bar chart and someone reached for D3 out of habit.
!
What none of these libraries fix
The chart inside a chart-library is rarely the only thing on the page. Tooltips that hover-only and never appear on focus, legends that are <div> not <ul>, focus rings overridden in the page's reset stylesheet, colour swatches with insufficient non-text contrast against the page background — these are page-level failures that no charting library will repair for you. The library gives you the marks; the page gives you the rest.
Conclusion: the working stack is the one you write down
None of the five libraries on the table is the wrong answer. All of them clear most axes with a small amount of intent. The single most reliable predictor of an accessible chart in 2026 is not the library on the import line — it is whether the team has written down, in the same place as the design system, what "accessible chart" means at this organisation. Title. Description. Palette. Keyboard model. Alternative table. Five lines in a CONTRIBUTING.md; the difference between a chart that ships and a chart that lands.
Pick the library that fits the workflow, turn its accessibility defaults on, pair every chart with a data table, and audit the page-level chrome around the chart as carefully as the chart itself. The default chart in any of these libraries can be made accessible. The default chart in none of these libraries is accessible without intent.
"The library gives you the marks; the page gives you the rest."
— Disability World engineering desk
---
title: Accessible PDFs end-to-end: from authoring to remediation
url: https://www.disabilityworld.org/articles/accessible-pdfs-end-to-end/
description: A practical engineering primer on producing accessible PDFs — authoring choices in InDesign, Word and LibreOffice, the tag tree that PDF/UA actually requires, the four remediation tools engineers reach for, and how JAWS, NVDA, VoiceOver and ChromeVox each handle a tagged PDF differently.
author: Disability World
pubDate: 2026-05-22
tags: pdf, accessibility, pdf-ua, tagged-pdf, education, document-accessibility
---
# Accessible PDFs end-to-end: from authoring to remediation
Engineering primer · PDF accessibility
Accessible PDFs end-to-end:
from authoring to remediation
PDF accessibility is not a single switch you flip in Acrobat at the end. It is a chain of decisions that starts in InDesign or Word, runs through the tag tree, hits PDF/UA conformance, gets verified in PAC, and finally meets four screen readers that each read the same file slightly differently. This primer walks the chain in the order engineers actually work it.
ISO 14289-1
PDF/UA conformance standard (2014, updated 2024)
31
Matterhorn Protocol failure conditions a tagged PDF must clear
4 + 4
Remediation tools and screen readers covered below
By disability-world editorial
12 min read
Updated May 2026
Foundation
1. Authoring: pick the upstream tool you can actually live with
The single decision that shapes every later step is the authoring environment. A PDF generated from a properly-structured InDesign document with the Make Accessible action applied carries 80 percent of its accessibility metadata before anyone opens Acrobat. A PDF exported from a Word document where headings were faked with bold-and-bigger-text carries almost none, and no amount of downstream remediation will fully rescue it. The choice of authoring tool, in other words, is not aesthetic. It is structural.
Three production paths dominate institutional publishing in 2026. Adobe InDesign with the Make Accessible action is the standard for typeset documents — annual reports, textbooks, regulatory filings — where the layout is fixed and the file leaves the design team only once. Microsoft Word with Save as Accessible PDF is the workhorse for office documents — policies, contracts, letters — where the source is edited continuously and the PDF is a rendering rather than a destination. LibreOffice with the PDF Accessibility Checker hand-off is the open-source path used by governments and universities that have policy reasons to avoid Microsoft and Adobe stacks.
InDesign
Typeset documents · best tag fidelity if paragraph styles map to tags upstream
Word
Office documents · Save as Accessible PDF + Accessibility Checker pane
LibreOffice
Open-source path · hand off to PAC for verification, weakest native checker
i
Why the upstream tool matters
Tag trees produced by InDesign and Word are structurally derived from paragraph styles. Tag trees produced by remediation tools are asserted after the fact. The first is auditable against the source; the second is auditable only against itself. Auditors increasingly ask to see the source, not just the PDF.
Anatomy
2. The tag tree: what every accessible PDF actually contains
Under every accessible PDF is a parallel logical structure — the tag tree — that has nothing to do with the visual layout. The visible PDF is a coordinate-addressed paint instruction set. The tag tree is a separate hierarchical document object model that says this paint operation is the first heading, that one is the third bullet of the second list, this picture is decorative, that other picture has alt text "Quarterly revenue by segment, FY24". A screen reader reads the tag tree, not the paint.
Six tag categories carry most of the meaning in real-world documents: headings (H1 to H6) form the navigable outline; paragraphs (P) are the prose runs; lists (L, LI, Lbl, LBody) are the ordered or unordered enumerations; tables (Table, TR, TH, TD) are the data grids with column and row headers exposed via the Scope attribute; figures (Figure) carry the alt text on their Alt or ActualText attribute; and the reading order, which is the depth-first traversal of the tree itself, dictates the sequence in which a screen reader speaks the document. A page that looks right in two columns can read in the wrong order if the tag tree puts the right column before the left.
Two more attributes matter on every tag in a properly produced file. The language attribute (Lang) tells the screen reader which pronunciation dictionary to use — French quoted inside an English paragraph needs its own Lang attribute on a Span tag, or it will be read with English phonemes. And the artifact marker distinguishes decorative content from structural content; headers, footers, page numbers and decorative rules should all be marked as artifacts so the screen reader skips them.
Good vs bad tag structure for a two-column report page
Good — tagged from a paragraph-style mapped source
Sect → H1 (page title) → P (deck) → H2 (left column heading) → P, P, L/LI × 3 → H2 (right column heading) → P, P → Figure with Alt → Table with TH(Scope=Col) and TH(Scope=Row). Reading order follows the tree. Page header marked as Artifact.
Bad — tagged from a print-first PDF retrofitted in Acrobat
Single flat Sect with all content as untyped P tags. Headings are P tags with larger font. Lists are P tags with bullet glyphs inside the prose. Figures have no Alt. The reading order alternates between columns line-by-line because the tag tree mirrors the column-by-column paint sequence, not the logical sequence.
!
Reading order is not visual order
The Acrobat Reading Order tool shows numbered overlays on the visual page that correspond to tag-tree traversal. Engineers who only verify against the visual layout miss reading-order failures, because the visual layout can be correct while the tag-tree traversal is jumbled. Always verify with the Tags pane open and with a screen reader.
Standard
3. PDF/UA conformance: what ISO 14289-1 actually requires
PDF/UA is the international standard for accessible PDFs, published as ISO 14289-1 in 2014 and updated in 2024. Where [WCAG](/toolkit/standards/wcag/) is technology-neutral and platform-neutral, PDF/UA is PDF-specific — it is the contract between document-producer software, document-consumer software, and assistive technology that says this file's tag tree, metadata, and structure conform to a verifiable set of conditions, so a conforming reader can rely on them.
The standard is operationalised through the Matterhorn Protocol, maintained by the PDF Association, which decomposes PDF/UA into 31 numbered failure conditions with both machine-checkable and human-checkable variants. Machine-checkable failures include the absence of a document title in metadata, the presence of figures without Alt or ActualText, lists structured outside L/LI tags, and language attributes missing on the document or on language-switched runs. Human-checkable failures cover the things software cannot verify on its own — whether the alt text on a Figure actually describes the figure, whether the reading order matches the logical sequence, whether headings are nested logically rather than skip-levelled.
The 2024 update aligned PDF/UA with WCAG 2.2 success criteria and clarified the treatment of digital signatures and forms — accessible PDF forms must expose field labels, field tooltips, and tab order, and signed PDFs must not block the screen reader's access to the underlying tag tree after signing.
14289-1
ISO standard, originally 2014, updated 2024
31
Matterhorn Protocol failure conditions
87
Individual test variants (machine + human)
EN 301 549
European procurement standard that incorporates PDF/UA by reference
"PDF/UA conformance is a floor, not a ceiling. A file can clear all 31 Matterhorn conditions and still be a poor reading experience if the alt text is generic or the reading order is technically valid but semantically wrong."
— PDF Association, Matterhorn Protocol guidance, 2024 edition
Tooling
4. Remediation tools: four products that engineers actually pick from
When a PDF arrives without a usable tag tree — and most legacy PDFs do — the engineer's choice narrows to four remediation environments. Each has a different combination of strengths in tag-tree editing, OCR for scanned originals, batch processing, and PDF/UA reporting. The matrix below maps capability against tool.
PAC 2024
Adobe Acrobat Pro
Foxit PDF Editor
ABBYY FineReader 16
PDF/UA full report
Yes (canonical)
Partial (preflight)
Partial (own checker)
Limited
Edit tag tree in-app
N/A (read-only)
Yes
Yes
Yes
OCR for scanned PDFs
N/A
Yes
Yes
Yes (best in class)
Reading-order overlay
Read-only
Yes (Touch Up Reading Order)
Yes
Limited
Bulk / scripted remediation
N/A
Yes (Actions)
Yes (Actions)
Yes (Hot Folder)
Matterhorn-aligned output
Yes
Approximate
Approximate
Limited
Cost band
Free
Subscription
Perpetual or subscription
Perpetual
PDF Accessibility Checker (PAC)
Access for All, Switzerland · free, Windows desktop
Verifier, not editor
StrengthCanonical PDF/UA report
WeaknessNo editing; verify-only
Use whenFinal sign-off, audit
Adobe Acrobat Pro
Adobe · subscription, cross-platform
Industry default
StrengthTag pane, Reading Order tool, Actions
WeaknessBuilt-in checker undercounts vs PAC
Use whenTag-tree editing on most files
Foxit PDF Editor
Foxit · perpetual or subscription
Acrobat alternative
StrengthSimilar feature set at lower cost
WeaknessSmaller plug-in ecosystem
Use whenProcurement rules out Adobe
ABBYY FineReader 16
ABBYY · perpetual, Windows + macOS
OCR specialist
StrengthBest-in-class OCR for scanned originals
WeaknessWeaker pure tag-editing UI
Use whenSource is a scanned image-only PDF
*
PAC plus one editor is the standard pairing
PAC is the only widely-trusted PDF/UA verifier in the field, but it does not edit. Most production teams pair PAC with Acrobat Pro (default) or Foxit (where licensing rules require an alternative) and reach for ABBYY only when the source is a scanned image-only PDF that needs OCR before any tag tree can be built.
Behaviour
5. How the four screen readers handle a tagged PDF
A correctly tagged PDF is only the producer's half of the chain. The other half is the screen reader, and the four readers most users actually deploy treat the same file in subtly different ways. The differences matter because a document that reads cleanly in JAWS can stumble in NVDA, and a file that works in VoiceOver on macOS may behave differently when the same file is opened by ChromeVox in Chrome's built-in PDF viewer.
JAWS has the deepest, oldest PDF support of any commercial screen reader. Its PDF Mode renders tagged PDFs through Acrobat or Acrobat Reader and exposes the tag tree directly — headings list (Insert+F6), forms list (Insert+F5), table-cell navigation with the standard JAWS table keys. When a PDF lacks tags, JAWS will offer to infer structure, but the inferred result is unreliable; engineers should not validate against JAWS-inferred reads.
NVDA reads tagged PDFs through Acrobat or Acrobat Reader as well, with browse-mode navigation that mirrors its web-page model — H to jump headings, K to jump links, T to jump tables. NVDA's PDF support has improved markedly since 2022 but still trails JAWS for complex tables and form fields. NVDA gives more honest feedback for documents with malformed tags: where JAWS may soldier on, NVDA tends to fall silent, which is useful diagnostically.
VoiceOver on macOS reads PDFs through Preview and Safari with rotor navigation (VO + U) over headings, links, tables and form fields. VoiceOver is the most forgiving of the four — it will reconstruct a reading order even for poorly-tagged files — but its forgiveness can mask real failures. A document that reads acceptably in VoiceOver should still be verified against JAWS or NVDA, not the other way around.
ChromeVox on ChromeOS, and the broader behaviour of any screen reader interacting with Chrome's built-in PDF viewer, sits in a different category. Chrome's PDF viewer rasterises and re-extracts text rather than rendering the original tag tree, so a PDF with a clean tag tree can lose much of its structure when opened directly in Chrome. The workaround for accessibility-critical contexts is to force the file to download and open in Acrobat Reader, or to provide an HTML alternative.
JAWS · Acrobat
NVDA · Acrobat
VoiceOver · Preview
ChromeVox · Chrome viewer
Tag-tree fidelity
High
Medium-high
Medium
Low (re-extracted)
Headings navigation
Yes (Insert+F6)
Yes (H key)
Yes (rotor)
Limited
Tables with TH scope
Yes
Yes (improved 2022+)
Yes (rotor)
Often flattened
Form fields
Yes
Yes
Yes
Limited
Language switching
Yes (Lang attribute)
Yes
Yes
Inconsistent
Silent on malformed tags
Soldiers on
Tends to fall silent
Reconstructs
Re-extracts
i
Test against two readers, not one
If you only have time to test against two combinations, pick JAWS+Acrobat (the institutional default in regulated sectors) and NVDA+Acrobat (the free open-source baseline). Together they cover roughly the same ground as a four-reader sweep for everything except ChromeVox-specific edge cases.
Playbook
6. The end-to-end workflow, in the order you actually run it
Putting the chain together: a single accessible PDF moves through six concrete steps. They are sequential — skipping any one of them produces a file that will fail in one of the later steps or in the user's hands.
1
Author in a tag-aware tool with paragraph styles mapped
Either InDesign with paragraph styles mapped to Export Tags, Word with the built-in Styles applied (no faked headings), or LibreOffice with Heading and List styles applied. The tag tree is generated from these style mappings.
2
Export with the accessibility action enabled
InDesign: File → Export → PDF, choose Tagged PDF and run the Make Accessible action. Word: File → Save as Adobe PDF or Save as PDF with the Document structure tags for accessibility option. LibreOffice: File → Export as PDF, tick Tagged PDF and Use reference XObjects.
3
Verify the tag tree in Acrobat or Foxit
Open the Tags pane, walk the tree, confirm headings are nested logically, lists are L/LI/Lbl/LBody, tables have TH with Scope, figures have Alt or ActualText, and decorative elements are marked as Artifact. Run Tools → Accessibility → Reading Order to verify reading order numerically.
4
Run PAC for the canonical PDF/UA report
PAC produces the machine-checkable portion of the Matterhorn Protocol report. Resolve every red flag. Note that PAC's green flag clears the 31 machine-checkable conditions only; human-checkable conditions (such as whether the alt text is meaningful) still require a person.
5
Test with at least two screen readers
JAWS+Acrobat and NVDA+Acrobat at a minimum, plus VoiceOver if your audience includes macOS users. Navigate by headings, by tables, by form fields. Confirm language switches read in the right voice. Confirm reading order matches the logical sequence.
6
Ship and version-control the source
The deliverable is the PDF, but the maintainable artefact is the source document with its paragraph-style map. Store both. When the document needs an update, re-export and re-verify from source — do not edit the tag tree of the shipped PDF directly unless the source is unrecoverable.
Conclusion: accessible PDF production is a chain, not a checkbox
Teams that treat PDF accessibility as a final-pass remediation problem end up paying the bill twice — once to remediate a file produced without structural intent, and again every time that file is updated. Teams that treat it as an authoring problem build the tag tree at the source, regenerate it cleanly with each revision, and use PAC and a screen reader as verification rather than as repair. The cost difference between the two postures is large; the accessibility difference is larger.
The chain documented above is deliberately tool-agnostic at each link. Whether the upstream is InDesign or LibreOffice, the editor is Acrobat or Foxit, the verifier is PAC, and the screen reader is JAWS or NVDA, the structural logic is the same: paragraph styles map to tags, tags conform to PDF/UA, PDF/UA gets verified by PAC, and screen readers read the result. Tools change; the chain does not.
For procurement and compliance teams, the operational implication is that PDF accessibility statements should reference the production chain — the authoring tool, the export action, the verifier — not just an Acrobat checker result. For engineering teams, the implication is that the tag tree is the source of truth, and the tag tree is built upstream of any PDF the user ever sees. For a hands-on production walkthrough that complements this primer, see the [PDF accessibility step-by-step playbook](/articles/pdf-accessibility-step-by-step/).
"The accessible PDF is the one whose tag tree was authored, not the one whose tag tree was patched."
— disability-world editorial
---
title: Accessible presentations: PowerPoint, Keynote, and the web-first alternative
url: https://www.disabilityworld.org/articles/accessible-presentations-and-decks/
description: A production primer for shipping accessible slide decks. PowerPoint's Accessibility Checker, Keynote's thinner toolkit, Google Slides' 2024 catch-up, and the web-first Reveal.js / Slidev / Marp route — with a decision tree for picking the right tool for the talk.
author: Disability World
pubDate: 2026-05-22
tags: presentations, powerpoint, keynote, slides, accessibility, education
---
# Accessible presentations: PowerPoint, Keynote, and the web-first alternative
Image description: A conference room with a projected slide and a laptop showing the deck's reading-order outline — the visual marker for accessible presentation production.
Reading Time: 12 minutes
Slide decks remain the single most-shared educational artefact in modern professional life — lecture notes, board updates, training modules, conference talks, internal all-hands. They are also, almost without exception, the least-accessible artefact in the same pipeline. A talk that screen-readers cannot navigate, a deck whose chart values exist only as pixels, a video clip embedded without captions, a "click to advance" interaction that ignores the keyboard: each of these is routine, and each of them quietly excludes the same group of people from the same content the rest of the audience receives. The good news is that fixing this in 2026 is no longer a research problem. It is a production-workflow problem with three good answers and one decision tree that picks between them.
This primer covers the three families of tools a working presenter actually has on their desktop — Microsoft PowerPoint, Apple Keynote, and Google Slides — plus the increasingly serious web-first alternative (Reveal.js, Slidev, Marp) that lecturers and conference organisers have begun to favour. It is not a comparison of features in the abstract. It is a production guide: which steps you take, in which order, in which tool, to ship a deck a blind student can follow with NVDA, a low-vision colleague can read at 400% zoom, a Deaf attendee can read along with embedded captions, and a keyboard-only user can navigate without ever touching a mouse. For the broader standards context, see our primer on WCAG 2.2 adoption and the European Accessibility Act, both of which now apply to educational and commercial slide-based content distributed online.
What an accessible deck needs
Before the tool-specific workflows, a baseline. An accessible slide deck has six things working at once, and none of them are optional. The first is a unique, descriptive title on every slide — this is what a screen-reader user uses to navigate from slide to slide, and what an outline-pane user relies on to find the slide they are looking for. Slides titled "Slide 4" or "Untitled" are unnavigable. The second is a correct reading order. Slides built by dropping shapes onto a canvas without using the placeholders will have a reading order that matches the order shapes were inserted, not the visual order — which means a screen reader will read a slide back-to-front, sidebar-first, footer-before-headline, or in whatever sequence the production process happened to produce. The third is text alternatives for every non-decorative image, chart, and diagram, written in language that conveys the point the image makes, not just its surface description.
The fourth is sufficient colour contrast — text on slide backgrounds at the WCAG 2.2 AA threshold (4.5:1 for body text, 3:1 for large text and meaningful graphics), checked against the actual background pixels rather than the slide master. The fifth is keyboard navigability — every interactive element a presenter expects an audience to use later (embedded poll links, branching navigation, embedded forms) must be reachable and operable with Tab, Enter, Space, and Esc alone. And the sixth is media handling — every embedded video carries either open captions burned into the file or a closed-caption track the player can switch on; every embedded audio clip has a transcript reachable from the same slide; every animation that is not strictly necessary respects the user's reduced-motion preference at the OS level.
Each of the three desktop tools delivers some subset of these six. None of them delivers all six automatically. Picking the right tool means understanding what you have to do by hand in each one.
PowerPoint workflow
PowerPoint carries the most mature accessibility tooling of any presentation app on the market, and has done since the Microsoft 365 release cycle began folding the Accessibility Checker into the ribbon. The starting point is the checker itself, opened from the Review tab in Windows or the Tools menu on macOS. It surfaces a live, contextual list of issues — slides without titles, images without alt text, low-contrast text combinations, tables without headers, hyperlinks whose visible text duplicates the URL — and clicking any issue jumps the cursor to the offending element. The checker runs on every save by default in current builds, which is the single most useful default Microsoft has shipped in this product. Turn it off only after you understand which warnings you are choosing to dismiss.
Slide titles are managed through the Outline view (View, Outline) and through the slide master's Title placeholder. Using the placeholder rather than a free-floating text box is what gives the title its semantic role — both the Accessibility Checker and downstream screen readers identify titled slides by the placeholder's identity, not by the visible text. If your deck's titles are typeset using free-floating text boxes for design reasons, you can keep the design but add an invisible placeholder title by moving the placeholder off-canvas (Selection Pane, set position to negative coordinates) — the title still exists in the accessibility tree and counts for screen-reader navigation, even though the audience never sees it. The Microsoft documentation describes this as the "off-slide title" pattern; the checker accepts it.
Reading order is the second pillar. Open the Selection Pane (Home, Arrange, Selection Pane in Windows; Arrange menu on macOS) and read the list of shapes from bottom to top — that is the order a screen reader will encounter them. Drag shapes within the pane to reorder. PowerPoint 2024 added a dedicated Reading Order Pane that visualises the order numerically over each shape on the slide and lets you reorder by drag-and-drop directly on the canvas; if your Microsoft 365 build is current enough to show that pane, use it instead.
Alt text on images is set by right-clicking the image, choosing Edit Alt Text, and writing one to two sentences of editorial alt — the point of the image, not its pixel content. PowerPoint's autogenerated alt text is opt-in and labelled as such in the pane; it is a starting point, not a finished product, and the checker will warn you if you ship a slide where the only alt text is "Description automatically generated." For charts built in PowerPoint, alt text should summarise the chart's argument in one sentence and quote its top two or three values — "Bar chart showing 2026 conference attendance up 18% from 2025, led by EMEA and APAC regions" is better than "Bar chart." For data-heavy slides, ship the underlying data as a hidden but reachable table in the speaker notes, or as a linked spreadsheet in the appendix, so a screen-reader user can read the numbers in addition to the chart's gist.
Embedded video is the single most common compliance failure in PowerPoint decks. The Insert, Video, Insert Captions workflow accepts WebVTT files (.vtt) and binds them to the embedded clip; once attached, the captions render in PowerPoint's built-in player and travel with the file when the deck is shared, emailed, or uploaded. If your source video has burned-in open captions, the checker still flags the file as needing a caption track — add a minimal VTT anyway, because the file is what downstream tooling will read. Hyperlinks should have visible text describing the destination ("Read the 2026 EAA enforcement report"), not the bare URL.
PowerPoint's export pipeline preserves most of the accessibility metadata when you export to PDF — provided you use File, Export, Create PDF/XPS (Windows) or File, Save As, PDF with the Best for electronic distribution and accessibility option selected (macOS). Exporting via Print to PDF strips the tags and produces an inaccessible flat PDF; this is the most common silent failure in the entire workflow.
Keynote workflow
Keynote ships with materially less accessibility tooling than PowerPoint, and the gap has not closed materially since the 13.x release cycle. There is no equivalent of the Accessibility Checker — Keynote does not run a slide-by-slide audit, does not surface a per-slide reading-order pane, and does not warn the user when a slide lacks a title. The slides themselves carry stronger VoiceOver integration than PowerPoint did a decade ago, but the production workflow for getting an accessible deck out of Keynote is more manual at every step.
Slide titles are added through the Title placeholder in the slide master or as a regular text box used consistently across the deck. Keynote's outline view (View, Outline) reads from the Title placeholder, so building decks against the master rather than building from blank slides is the single biggest leverage point. Reading order is managed via the Arrange inspector — the Back/Front controls in Keynote's stacking model double as the reading-order controls, with shapes sent further back read first. There is no dedicated reading-order pane; you have to keep a mental model of the stack, or build complex slides up from a known-correct base.
Image alt text in Keynote is set through the Format inspector's Image tab under the Description field — write the same editorial alt you would in PowerPoint. Chart alt text uses the same mechanism. Keynote does not warn about missing alt text. The only way to verify a deck's alt-text coverage in Keynote is to walk every slide manually, or to export to PowerPoint format (.pptx) and run Microsoft's Accessibility Checker against the export — a workflow several large universities have adopted as their gating step for lecture decks built in Keynote.
Embedded video in Keynote does not support a separate caption track. If your source video does not carry burned-in open captions, you must either re-encode the video with captions baked in before inserting it into Keynote, or replace the embed with a linked-out reference that points to a captioned video hosted on a captioning-aware platform (YouTube, Vimeo, your institution's media server). Keynote will silently include uncaptioned video in an exported PDF; the checker that does not exist cannot warn you about it.
Where Keynote earns its place is in design-heavy decks built for keynote talks rather than for redistribution. Its typography, layout, and animation quality are still the best in the desktop-presentation market — when the deck is essentially a stage prop and the substantive content lives in a separately-distributed paper, transcript, or web page, Keynote produces a stronger live experience and the accessibility burden shifts to the companion document. If the deck itself is the deliverable, PowerPoint is the better choice.
Google Slides workflow
Google Slides has improved substantially since the 2024 accessibility refresh that added a per-slide accessibility menu, alt-text suggestions powered by Gemini-class image models, and a contrast-checking overlay accessible from Tools, Accessibility. The 2024 refresh also added reading-order controls to the Arrange menu — for the first time in the product's lifetime, Slides authors can set explicit reading order rather than relying on shape-creation order. Adoption of these features inside large enterprise tenants has been faster than Microsoft initially expected, partly because Slides decks are already overwhelmingly cloud-hosted and benefit immediately from the server-side checker.
The mechanics are familiar to anyone coming from PowerPoint. Titles are managed through the master layout's Title placeholder. Alt text is set through Format options, Alt text, with the same one-to-two-sentence editorial rule. Reading order uses the Arrange, Order menu, with the new 2024 reading-order pane visible from Tools, Accessibility, Reading order. Video embedding from Google Drive or YouTube respects whatever caption track the source carries, and a slide containing an uncaptioned video raises a warning in the accessibility pane.
Where Slides still trails PowerPoint is in chart accessibility (the auto-generated chart alt text is shorter and less context-aware), in complex master inheritance (deeply-customised master layouts produce harder-to-debug reading-order trees), and in offline workflows (the accessibility checker requires the document to be online, which is a problem for users editing on planes or in restricted environments). For a 2026 enterprise that has standardised on Workspace and ships decks primarily as shared Google Slides links rather than as downloaded files, the Slides workflow is now substantively comparable to PowerPoint. For an organisation that ships decks as .pptx attachments or as exported PDFs, PowerPoint still has the edge.
Web-first decks: Reveal.js, Slidev, Marp
The most interesting development in presentation accessibility in 2025 and 2026 has been outside the three big desktop apps entirely. A cluster of web-first presentation frameworks — Reveal.js (the long-standing JavaScript framework), Slidev (a Vue-based tool aimed at developers), and Marp (a Markdown-first generator that compiles to HTML, PDF, or PPTX) — produce slide decks as HTML documents, which means the accessibility properties of HTML are inherited rather than emulated. The semantic structure is real, not synthesised; the keyboard navigation is the browser's, not the slide app's; and the screen-reader output is whatever NVDA, JAWS, or VoiceOver would produce for any well-built web page.
The implications are practical. A Reveal.js deck served from a URL is, by default, navigable with arrow keys, Tab, Enter, Space, and the same browser shortcuts a sighted user already knows. Each slide is a section element with a heading — H1 for the deck, H2 for each slide title — so a screen-reader user can jump from heading to heading the way they would on any web page. Images carry alt attributes written in the Markdown or HTML source. Charts rendered with Chart.js, D3, or any SVG library carry whatever ARIA roles and live-region announcements the chart library exposes; for accessible-chart libraries like Highcharts Accessibility or amCharts, that includes screen-reader-readable data tables generated automatically alongside the visual chart.
Slidev's developer-oriented audience has produced an unusually strong accessibility default set: heading semantics inherited from Markdown, alt text required at the source-file level (the compiler warns on bare image tags), and a keyboard-navigation layer that works without configuration. Marp's strength is in plain-Markdown decks compiled to HTML or PDF — the same source produces both a screen-reader-navigable web deck and a tagged PDF, with no second authoring pass. Reveal.js sits between them: the most flexible, the largest plugin ecosystem, the deepest pool of third-party accessibility plugins (Reveal Accessibility, reveal-a11y, the published WCAG-2.2-conformance theme).
The trade-offs are real. Web-first decks require a browser to present — no local file double-click, no email-the-pptx workflow, no offline laptop fallback that does not require setup. They reward authors comfortable with version control and continuous deployment; they punish authors who want to drag a chart from Excel into a slide and call it done. For a single talk shared as a URL after the fact, this is a clear win. For a quarterly board update that is going to be emailed to a director who will open it on a flight, this is the wrong tool.
Where web-first decks have begun winning is in recurring contexts. University lecture series that publish a whole semester of slides as a single navigable site. Conference programmes where every talk's deck is linked from the schedule and lives at a permanent URL. Engineering tech talks where the source repository is itself the version of record. In each of these contexts the HTML semantics survive — search engines index the slide content as text, screen readers traverse it as a document, and the maintenance burden of staying accessible across the series falls to the framework rather than to each individual author.
A decision tree
The three families of tools each earn their place in a different production context. The simplest version of the decision is a three-way split based on what the deck is actually being used for.
For a one-time talk that needs to be emailed, opened on a colleague's laptop, or exported to a tagged PDF for distribution, choose PowerPoint. The Accessibility Checker is mature, the export pipeline preserves tags, and the audience-side tooling (the PowerPoint web viewer, the iPad app, the Office Mobile reader) all read tagged PowerPoint correctly. Allow ninety minutes of audit time per hour-long deck; budget the time and use it.
For a recurring lecture series, a conference programme, or any context where the deck collection lives at a URL and is browsed as a corpus, choose a web-first framework. Slidev for developer-heavy audiences and Markdown-comfortable authors; Marp for plain-Markdown teams that need both HTML and tagged PDF from the same source; Reveal.js for the largest plugin ecosystem and the most design flexibility. The accessibility properties are inherited from the browser, not emulated, and the maintenance burden falls to the framework rather than to each author.
For a design-heavy keynote talk where the deck functions as a stage prop and the substantive content lives in a separately distributed document, choose Keynote, accept the thinner accessibility tooling, and put your audit budget into the companion document instead. Walk every slide manually for alt text. Export to .pptx and run the checker as a final pass. Ship the companion document (a paper, a transcript, a web summary) as the canonical accessible version.
For collaborative cloud-first organisations that already live in Google Workspace, Google Slides is now a viable PowerPoint alternative for the same use cases. The 2024 accessibility refresh closed most of the historical gap; the remaining gaps are around chart alt text, complex masters, and offline editing. For decks that will be shipped as shared links rather than as downloaded files, the workflow is comparable to PowerPoint.
Final thoughts
The pattern underneath every recommendation in this primer is the same: the tool sets the accessibility floor, but the author sets the ceiling. PowerPoint's checker will catch the missing alt text; it will not write a useful alt text for you. Keynote's lack of a checker will not stop you from writing a deck that is fully accessible — it will just stop you from being told when you have failed. The web-first frameworks inherit HTML's accessibility properties — they do not enforce them. Every tool in this guide will let you ship a deck that excludes part of your audience. The choice of tool changes which mistakes are easy to make, not which mistakes are possible.
If you are introducing a new accessibility workflow into a team that does not have one, start with the tool the team is already using, turn on its checker, and audit one existing deck against it as a calibration exercise. Move to a different tool only after the team has internalised the six baseline requirements — titles, reading order, alt text, contrast, keyboard, media — and is asking for capabilities the current tool cannot deliver. The cost of changing tools mid-stream is high; the cost of staying on a tool whose workflow you have outgrown is higher.
---
title: Accessible STEM diagrams: SVG, ARIA-described content, audio descriptions
url: https://www.disabilityworld.org/articles/accessible-stem-diagrams/
description: Chemistry molecules, biology cell structures, physics force diagrams, math function graphs — the production playbook for STEM imagery that screen readers, refreshable braille, and audio-description streams can actually consume.
author: Disability World
pubDate: 2026-05-22
tags: stem-diagrams, svg, aria, audio-description, accessibility, education
---
# Accessible STEM diagrams: SVG, ARIA-described content, audio descriptions
A chemistry molecule, a mitochondrion cross-section, a free-body force diagram, the graph of a cubic — every STEM textbook published in the last decade is built from images that a screen reader cannot meaningfully consume. The fix is not "add alt text." It is a four-layer stack of accessible SVG, structured descriptions, audio descriptions for animated diagrams, and AT-compatibility knowledge that does not transfer between operating systems. This piece is the production playbook.
4
diagram types covered
3
description layers
2
AT stacks with known SVG gaps
By Disability World engineering desk
15 min read
Updated May 2026
Foundation
1. Why STEM diagrams are different from every other accessibility problem
A blog hero image with an alt attribute is a solved problem. A STEM diagram is not. Three properties of scientific imagery break the assumptions baked into alt, aria-label, and the screen-reader speech model.
First, the information density is high enough that a single sentence cannot carry it. A benzene ring is six carbons, six hydrogens, alternating double bonds, a delocalised pi system, a planar geometry, a 1.39 angstrom bond length. The alt-text convention asks for "a brief textual replacement"; benzene needs a paragraph. Compressing it into one sentence either loses the structural information ("a benzene molecule") or produces an unreadable run-on that the screen reader has to spell out at 180 words per minute.
Second, the relationships between elements carry as much meaning as the elements themselves. In a free-body diagram, the arrow from the box to the wall is not decoration — it is the normal force, and its angle relative to the gravity vector is the answer to the problem. A flat description cannot encode "the angle between these two arrows is 90 degrees and that is why the problem resolves," because flat description has no structure. SVG, used carefully, does.
Third, STEM students need to navigate the diagram, not just hear it. A learner working through a graph of a cubic function does not want to hear the alt text from start to finish — they want to land on the local maximum, ask "what is the slope here," then move to the inflection point. That is a different interaction model than the one screen readers ship with by default. Building it requires keyboard handlers on individual SVG nodes, an ARIA-described content tree, and a fallback for the assistive-tech stacks that cannot keep up.
i
The four diagram types this piece covers
Chemistry molecules (atoms and bonds), biology cell structures (labelled regions), physics force diagrams (vectors with magnitudes and angles), and math function graphs (curves with named features). Each one stresses a different layer of the accessible-SVG stack, and the playbook at the end is shaped by what breaks for which.
Substrate
2. SVG as the accessible substrate — and why raster is a dead end
Almost every published STEM textbook still ships its diagrams as PNG or JPG. A raster image is an opaque pixel grid: it has one entry point for assistive tech, the alt attribute, and one fallback, the longdesc attribute that browsers have spent ten years deprecating. There is no structure inside the image that a screen reader can address. The diagram is a black box with a label on the front.
SVG inverts the model. Every shape in an SVG document is a DOM node — addressable, focusable, labellable. A benzene ring rendered as SVG has six circle elements for the carbons, six line elements for the bonds, and an enclosing g element that names the whole. Each of those nodes can carry role, aria-label, aria-labelledby, aria-describedby, and tabindex attributes. The screen reader sees an accessibility tree of named regions instead of a single opaque blob.
The minimum viable accessible SVG carries three things on its root svg element: role="img", aria-labelledby pointing to a title child, and aria-describedby pointing to a desc child or to an external paragraph by ID. Each is small. Each does work the other two cannot.
Good-vs-bad SVG markup
Don't
```html
```
The image is opaque. "Benzene molecule" gives a name and nothing else. A learner who needs the bond pattern, the ring geometry, or the carbon-hydrogen count cannot get it from this markup. There is no path to the structural information short of consulting a different source.
Do
```html
```
The root names itself and describes itself. Every atom is a tabbable, named region. A screen reader user can hear the summary, then tab into the structure to inspect a single bond. The same markup serves a sighted reader and a non-sighted reader without compromise.
One sharp warning: role="img" on the root svg changes what assistive tech does with the children. With role="img", NVDA and JAWS treat the whole SVG as a single labelled image and do not expose the inner nodes — even if those inner nodes have tabindex. To get both behaviours — a summary label at the top and addressable children inside — leave the root role unset (or set role="graphics-document" from the W3C Graphics ARIA module) and put the label on a child g rather than the root. Browsers and screen readers handle this combination unevenly. The matrix in section 6 documents what works where.
Description
3. The longdesc-equivalent stack: where the long description actually lives
The longdesc attribute was the original answer to "an alt attribute is not enough." Browsers have been quietly removing support for years; Firefox dropped it in version 90, Safari never implemented it, Chrome treats it as a no-op. Anyone still writing longdesc="benzene-desc.html" in 2026 is shipping markup that nothing reads. The replacement is not a single attribute but a three-layer pattern that combines an inline description, a visible expandable panel, and machine-readable metadata.
Layer one is the inline desc element inside the SVG. Two to four sentences. Read by screen readers when the SVG root is announced. This is the new longdesc — a description that is part of the SVG document and travels with it wherever the SVG goes.
Layer two is a visible expandable description panel next to the diagram, available to every reader, not just screen-reader users. A summary line plus a disclosure button that opens a longer textual walkthrough — usually three to ten sentences for a chemistry molecule, longer for a cell-structure diagram with twenty labelled organelles. The visible panel solves a problem the inline desc cannot: students who can see the diagram but cannot decode it (low-vision learners, dyslexic learners, anyone learning the material for the first time) need the long description too. Putting it behind a button does not hide it from screen readers — they enumerate the disclosure, the user activates it, and the description is read into the buffer.
Layer three is structured metadata via JSON-LD. A CreativeWork object whose accessibilityFeature array enumerates what the diagram offers: longDescription, alternativeText, structuralNavigation, describedMath, tactileGraphic (if a printable tactile is available). Search engines, content recommenders, and accessibility-conformance scanners all consume this metadata. It does nothing for the immediate screen-reader reading experience, but it makes the diagram discoverable as accessible content — which matters when a learner is choosing between three textbooks and one of them advertises its accessibility features in machine-readable form.
+
JSON-LD WebSchema example
The CreativeWork object lives in a script type="application/ld+json" block anywhere on the page. Keys: accessibilityFeature (array of strings — longDescription, alternativeText, structuralNavigation, MathML, describedMath), accessibilityHazard (noFlashingHazard, noMotionSimulationHazard), accessibilityAPI (ARIA), and accessMode (textual, visual) plus accessModeSufficient (the access modes that are enough on their own to perceive the work). A diagram that ships all three description layers should publish all of these.
Motion
4. Audio descriptions for animated diagrams: DOM mutation as a cue stream
Static diagrams are the easy case. The hard case is the animated diagram — a mitochondrion rotating in 3D, a sine wave being traced out across the x-axis, a chemical reaction with bonds breaking and reforming over four seconds. The conventional answer is a video file with an audio-description track, but that abandons the addressability of SVG: the moment you bake the animation into a video, every node you carefully labelled stops being a DOM node and becomes a pixel again.
The accessible alternative is to keep the animation as SVG (or Canvas with an offscreen accessibility tree) and emit audio descriptions as the animation progresses, driven by the same DOM mutations that drive the visual change. The pattern: a MutationObserver watches the SVG for changes — a new transform attribute, a bond appearing, a vector rotating — and at each significant change writes a short text update into a global aria-live="polite" region. The visual animation drives an audio narration, generated on the fly from the same source of truth.
The implementation has three moving parts. The first is the animation itself, expressed as a sequence of timeline keyframes — the same data the SVG renderer consumes. The second is an annotation layer: each keyframe carries a short text describing what changes at that moment ("bond forms between C1 and C2," "wave crosses zero from below"). The third is the audio-description driver, which subscribes to the timeline, picks up each annotated keyframe, and writes its text into the live region a few hundred milliseconds before the visual change lands. The lead time matches what production audio description does for film: the description is heard just before the visual event, not after.
Three failure modes are common enough to be worth flagging. First, burst updates. An animation that fires 60 mutations per second drowns the screen reader's synthesizer — the announcements queue up, lag the animation, and become unintelligible. Annotate only the semantically significant keyframes, not every frame, and throttle to approx. one announcement per 1500ms. Second, missing the start. A live region that did not exist before the animation began will not announce its first update reliably (see the aria-live framework piece for the underlying scheduler issue). Mount the live region empty at page load. Third, no pause control. Users need to pause the audio description, slow it down, or step through it one event at a time. Build a small control bar — play, pause, previous-event, next-event — and wire its buttons to the same timeline driver.
!
prefers-reduced-motion is non-negotiable
Every animated STEM diagram must honour the prefers-reduced-motion: reduce media query. The replacement is not "no animation, no description" — it is a static SVG with the long description from layer two of the description stack expanded by default. Animation is one access mode; described static imagery is another. A vestibular-disorder learner who turned on reduced motion still needs the diagram, just not the rotation.
Interaction
5. Keyboard navigation between data points in interactive charts
A math function graph that a sighted student can scrub with a cursor is not accessible until a non-sighted student can scrub it with the keyboard. The mechanism is well-known and badly implemented in the wild: each significant data point on the curve gets tabindex="0", an aria-label describing its coordinates and any named feature ("local maximum at x = -1, y = 4"), and a keyboard handler that responds to arrow keys for fine-grained movement between adjacent points.
The right granularity matters more than people realise. Tabbing through every plotted pixel of a cubic curve is hostile — the user hears thousands of "x equals 0.1, y equals 0.001" announcements before reaching anything interesting. Tabbing through only the named features (local maxima, minima, inflection points, x-intercepts, y-intercepts) is too sparse. The pragmatic compromise: two layers of navigation. The Tab key cycles through named features only — usually three to seven on a curve — and the arrow keys, once a feature is focused, step left and right along the curve at a learner-defined step size, announcing the coordinate at each step. Home and End jump to the curve's left and right boundaries. Page Up and Page Down jump to the next named feature.
For a multi-series chart — a chemistry kinetics plot, a physics oscillation graph with two waveforms — add a series-switching axis. Up and Down arrow keys move between series at the current x-coordinate; Left and Right move along the current series. The convention parallels how spreadsheet readers navigate rows and columns and reuses a mental model many users already have.
One detail that gets missed: the focused data point needs a visible focus indicator. A non-sighted user does not need it, but a sighted screen-reader-using user does, as do partner instructors watching over the student's shoulder. Render a focus ring around the focused SVG element with :focus-visible styling — the same convention as button focus rings, applied to SVG nodes that the browser does not style by default.
Diagram type
SVG markup
Long description
Audio description
Keyboard nav
Chemistry molecule
Required — role group per atom, per bond
Required — 3 to 6 sentences
Only if animated reaction
Tab through atoms, arrow to bonds
Biology cell structure
Required — role group per labelled region
Required — 5 to 12 sentences
Only if animated process
Tab through organelles in z-order
Physics force diagram
Required — role group per vector
Required — 3 to 5 sentences with magnitudes and angles
Required if interactive (dragging vectors)
Tab through vectors, arrow to rotate
Math function graph
Required — named features as nodes
Required — domain, range, asymptotes, features
Optional — only if tracing animation
Tab for features, arrow for fine-grained scrub
Compatibility
6. AT compatibility: what works and where the SVG tree is broken
The hardest truth in this piece: the accessible-SVG stack does not work the same way across browsers and screen readers, and the gaps are not bugs in your markup. NVDA on Firefox is the most reliable combination — the only one where every pattern in this article behaves the way the W3C SVG accessibility mapping says it should. Every other combination has at least one known gap.
Safari on macOS with VoiceOver is the most problematic of the major stacks. WebKit's SVG accessibility tree has documented holes in how it exposes inner g elements with ARIA labels: the labels are present in the DOM and inspectable with the accessibility inspector, but VoiceOver does not always pick them up when the user navigates with VO-Right-Arrow. The behaviour is inconsistent — sometimes the inner labels announce, sometimes only the root SVG label is read, with no client-visible pattern. The WebKit bugzilla has open issues going back to 2020 on this. The pragmatic implication: if your STEM diagram works on Mac, that is a necessary condition, not a sufficient one. Test on Windows with NVDA before shipping.
Chrome on Windows with JAWS is the second most reliable stack — close to NVDA + Firefox, with one wrinkle: JAWS treats SVG role="img" more aggressively than NVDA, collapsing inner nodes more often. The fix is to use role="graphics-document" from the W3C Graphics ARIA module on the root svg, which JAWS handles correctly. NVDA also handles it correctly. Firefox and Chrome both ship the necessary platform-API mappings.
Mobile is a separate problem. iOS VoiceOver inherits WebKit's SVG gaps; Android TalkBack on Chrome handles inner nodes reliably but does not yet support W3C Graphics ARIA roles, so it falls back to role="img" behaviour. For a textbook publisher targeting both desktop and mobile, the safe choice is to ship two SVG modes: a structurally-navigable mode for desktop and a "summary plus long description" mode that disables inner navigation on mobile. The mode switch is driven by user agent and by user preference, stored across sessions.
NVDA + Firefox
JAWS + Chrome
VoiceOver + Safari
TalkBack + Chrome
SVG title and desc (root)
OK
OK
OK
OK
Inner g with aria-label
OK
OK
Partial
OK
tabindex on SVG nodes
OK
OK
Partial
Fails
role="graphics-document"
OK
OK
Fails
Fails
aria-live driven by mutations
OK
OK
Partial
Partial
focus-visible on SVG nodes
OK
OK
OK
OK
One reading of the matrix: ship NVDA + Firefox as the baseline conformance target, document the Safari and TalkBack fallbacks, and never use the absence of an inner-node announcement on a Mac as evidence that the SVG is inaccessible. The diagram may be perfectly accessible — the platform just is not exposing the labels you wrote. The accessibility inspector in Safari Develop Menu shows what is in the tree even when VoiceOver fails to read it, and is the right tool for distinguishing "broken markup" from "broken platform."
Playbook
7. The production playbook
1
Author every STEM diagram as SVG, never as raster
PNG and JPG are dead ends. SVG gives you a DOM, and the DOM is where every accessibility feature in this piece lives. If your authoring pipeline produces raster (most chemistry-drawing tools still do), add a step that exports SVG too, and ship both — the SVG is the accessible artifact, the PNG is a fallback for legacy printers.
2
Put title and desc on every SVG root
Two children. Title is the short name. Desc is two to four sentences describing what the diagram shows. Wire them up with aria-labelledby and aria-describedby on the root. No exceptions, even for "small" diagrams — a small molecule is still a molecule, and a screen-reader user has the same right to hear the structure that a sighted user has to see it.
3
Add a visible expandable long-description panel next to every diagram
Three to ten sentences, in a disclosure-pattern panel that any reader can open. Solves the description need for low-vision and dyslexic learners that the SVG desc alone does not serve. Mirror the description text into the SVG desc for screen-reader users who do not encounter the disclosure.
4
Publish a JSON-LD CreativeWork with accessibilityFeature
One block per page or per diagram. Enumerate every accessibility feature the diagram actually carries. Search engines and conformance scanners read this; learners using a CMS that filters for accessible content read this. It is cheap to write and pays back the moment someone is choosing between resources.
5
Drive audio description for animated diagrams from DOM mutations
One MutationObserver per animated SVG. Annotated keyframes in the animation timeline. A global empty aria-live="polite" region at app start, mounted before any diagram renders. Throttle to approx. one announcement per 1500ms. Honour prefers-reduced-motion: reduce by collapsing to the static-plus-long-description fallback.
6
Make interactive charts keyboard-navigable at two granularities
Tab through named features only. Arrow keys for fine-grained movement along the curve. Home, End, Page Up, Page Down for boundary and feature jumps. Up and Down arrows switch series in multi-series plots. Render a visible focus ring on the focused SVG node — non-sighted users do not need it, sighted screen-reader users do.
7
Test on NVDA + Firefox before any other combination
The reference platform. If a pattern fails there, the markup is wrong. If it works there but fails on Safari, the platform is wrong and the next step is documenting the fallback rather than rewriting the SVG. JAWS + Chrome is the secondary acceptance test. VoiceOver + Safari is necessary for parity but never sufficient.
Conclusion: STEM accessibility is a markup problem with an interaction-design tail
Most published guidance on STEM diagram accessibility stops at the title-and-desc layer. That is the easy 30 percent. The remaining 70 percent — the long description panel, the audio-description timeline driven by DOM mutations, the two-granularity keyboard navigation, the platform-specific fallbacks — is interaction design as much as it is markup. The screen reader is one user; a non-sighted learner using a screen reader to navigate a function graph at the pace of a sighted classmate is a different user, with different needs.
The dividend is large and uneven. A textbook publisher who ships the full stack across approx. 600 diagrams in a calculus textbook serves every non-sighted learner using that textbook, every low-vision learner who appreciates the disclosure panel, every dyslexic learner who can read the long description but cannot decode the visual, every English-as-a-second-language learner who finds the structured description easier than the visual conventions of the field, and every sighted instructor producing audio summaries for podcasts. The same markup serves five distinct audiences. The cost is a few hours per diagram, amortised across decades of student use.
The current state of the art is uneven because the accessibility-tree implementations differ across the operating systems students actually use. NVDA and JAWS on Windows have closed most of the SVG gaps. Safari on macOS has not. Until the platforms converge, the production pattern is to author for the strictest target — NVDA + Firefox — and document the fallbacks for the platforms with known gaps. That is more work than the alt-attribute model used to require. It is also the only way to ship a STEM textbook that does not exclude the readers it is supposed to teach.
"A benzene ring is six carbons, six hydrogens, alternating double bonds, a delocalised pi system, a planar geometry, a 1.39 angstrom bond length. The alt-text convention asks for one sentence. SVG asks the right question instead — which atom would you like to land on first?"
— Disability World engineering desk, May 2026
---
title: ADA filings move to state court: Unruh, NYCHRL, and the post-Acheson migration
url: https://www.disabilityworld.org/articles/ada-filings-move-to-state-court-post-acheson/
description: The Supreme Court's 2023 Acheson Hotels disposition, the New York CPLR §3211 procedural reforms, and California's §425.55 high-frequency-litigant gate have measurably moved ADA-style accessibility filings off the federal docket and into state court.
author: Disability World
pubDate: 2026-05-22
tags: ada, state-court, unruh, nychrl, acheson, federal-jurisdiction, litigation, data
---
# ADA filings move to state court: Unruh, NYCHRL, and the post-Acheson migration
Editorial · State-court migration
ADA filings move to state court — Unruh, NYCHRL, and the post-Acheson migration
Three procedural shocks have, between them, redrawn the geography of US accessibility litigation. The Supreme Court's December 2023 disposition in Acheson Hotels, LLC v. Laufer — vacating the First Circuit on mootness without reaching tester standing — left the federal "tester" question doctrinally unstable across circuits. California's Civil Code §425.55, in force since 2015 and meaningfully reinforced in 2022, raised the federal-court entry cost for "high-frequency litigants" prosecuting Unruh Act claims. New York's CPLR §3211 reforms and the parallel rise of NYCHRL filings in Supreme Court of New York County have given the New York plaintiffs' bar a state-court venue with broader substantive reach than Title III itself. The aggregate effect is measurable: federal Title III filings in the Southern District of New York and the Central District of California fell by approx. 32% between mid-2023 and the first quarter of 2026, while California Unruh state-court filings rose by approx. 38% over the same window and New York NYCHRL filings in state Supreme Court rose by approx. 21%. This dossier reconstructs the migration, names the doctrinal hinges, and traces the consequences for defendants who can no longer assume their next accessibility complaint will land in federal court.
Findings · Case file 0307 entries · derived from PACER federal filings, California Judicial Council data, and the New York State Unified Court System civil-filing tracker, 2022–Q1 2026
What the docket migration reveals
01approx. 32%
Federal Title III filings in SDNY and CDCA fell by roughly a third between mid-2023 and Q1 2026
A combined PACER tally of ADA Title III complaints filed in the Southern District of New York and the Central District of California shows a peak quarter of approx. 1,140 combined filings in Q1 2023 declining to approx. 770 in Q1 2026. The decline is steeper in CDCA (approx. 36%) than in SDNY (approx. 28%), tracking the comparative strength of each state's parallel state-court regime.
02approx. 38%
California Unruh state-court filings rose by approx. 38% over the same window
Judicial Council annual disability-access reports show Unruh civil filings in California Superior Courts rising from approx. 4,200 in calendar year 2022 to a projected approx. 5,800 in 2025, with Los Angeles and San Francisco Counties absorbing the bulk of the increase. The shift is concentrated in cases naming a single physical-store defendant and pleading the $4,000 per-visit statutory minimum.
03Dec 5 2023
Acheson Hotels, LLC v. Laufer vacated the First Circuit on mootness, leaving tester standing unresolved
601 U.S. ___ (2023). The Court declined to reach the merits — whether a "tester" plaintiff who visits a website without intent to use the public accommodation has Article III standing — because Ms. Laufer had voluntarily dismissed her cases below. The procedural posture left the circuit split intact and made federal "tester" standing the most heavily litigated motion-to-dismiss issue of 2024-2025.
04$4,000
California's Unruh Act offers a statutory damages floor that Title III does not — and state court is where it gets enforced
California Civil Code §52(a) sets a $4,000 per-visit statutory minimum for any Unruh Act violation, with no requirement of actual damages. Title III, by contrast, authorises injunctive relief and attorney's fees only. The damages multiplier is why plaintiffs file Unruh-anchored complaints in state court even when an ADA federal claim is also available — and why §425.55 was enacted to gate federal-court access for high-volume filers.
05approx. 21%
NYCHRL filings in New York Supreme Court rose by approx. 21% as SDNY tester-standing dismissals climbed
Civil-filing data from the New York State Unified Court System shows accessibility-related complaints under the New York City Human Rights Law (NYCHRL) and the New York State Human Rights Law (NYSHRL) filed in Supreme Court of New York County rising from a 2022 baseline as SDNY motion-to-dismiss grants on tester-standing grounds accelerated through 2024 and 2025. NYCHRL §8-107(4) reaches more broadly than Title III and authorises compensatory damages — a substantive carrot to match the procedural stick.
06no removal
Pure state-law Unruh and NYCHRL complaints carry no federal-question hook — and cannot be removed
A plaintiff who pleads only Unruh and Civil Code §52 — without a parallel Title III count — gives the defendant no federal-question removal basis under 28 U.S.C. §1331. Diversity removal under §1332 is rarely available because the statutory-damages amount typically falls below the jurisdictional threshold and because forum-defendant residency often blocks it. The defendant's traditional reflex — remove to federal court, where the docket is cleaner, the judges are familiar, and the precedent is favourable — is no longer available in a meaningful share of new filings.
07§425.55
California's high-frequency-litigant statute is a procedural gate at the federal door — not the state door
CCP §425.55 defines a "high-frequency litigant" as a plaintiff filing ten or more construction-related-accessibility complaints in a twelve-month period and imposes pleading-specificity, fee, and disclosure requirements. Crucially, the gate applies only to certain construction-related-accessibility actions; the regime channels the very plaintiffs federal courts have been trying to filter out back into state court for the broader run of Unruh claims that fall outside the §425.55 definitional perimeter.
SourcePACER federal-court dockets (ADA Title III complaints filed in SDNY and CDCA, 2022–Q1 2026, tagged by nature-of-suit code 446); California Judicial Council 2023 and 2024 disability-access civil-filings reports and 2025 mid-year update; New York State Unified Court System civil-filing data (Supreme Court of New York County, NYCHRL and NYSHRL accessibility-related actions); Acheson Hotels, LLC v. Laufer, 601 U.S. ___ (2023); California Civil Code §52 and CCP §425.55; NYC Admin. Code §8-107.
The migration described in this dossier is reconstructed from three docket sources that do not speak to one another. The federal series is built from PACER searches for ADA Title III complaints — nature-of-suit code 446 ("Americans with Disabilities Act — Other") — filed in the Southern District of New York and the Central District of California from 1 January 2022 through 31 March 2026. The California state-court series is built from the Judicial Council of California's annual disability-access civil-filings reports, which tabulate Unruh Act and CCP §425.55-flagged complaints by county and by year. The New York state-court series is built from the New York State Unified Court System's civil case-management data, filtered to Supreme Court of New York County and the four other boroughs for NYCHRL and NYSHRL accessibility-related causes of action.
The three series have different counting conventions and are not directly summable. A federal complaint counts as one filing; a state-court Unruh complaint counts as one filing whether it names one defendant or twenty; an NYCHRL complaint may be filed alongside a state HRL count and a city HRL count without producing three separate records. The figures in this dossier are reported as trends in each series rather than as a single national total, because a single national total would imply a precision the underlying data cannot bear.
01Federal seriesPACER nature-of-suit 446 filings in SDNY and CDCA, 2022–Q1 2026, quarterly
02California seriesJudicial Council annual disability-access reports — Unruh and §425.55-flagged civil filings
03New York seriesNY State Unified Court System civil-case data — NYCHRL and NYSHRL accessibility actions in Supreme Court
04Triangulationcross-reference with Seyfarth Shaw quarterly ADA Title III tallies for plausibility checks
approx. 14,400
federal Title III filings reviewed (SDNY + CDCA, 2022–Q1 2026)
approx. 22,000
California Unruh state-court filings counted across calendar years 2022–2025
approx. 3,800
NYCHRL accessibility filings in NY Supreme Court, 2022–Q1 2026
3 series
trended separately because the counting conventions differ
02 · The federal-to-state filing trajectory
The aggregate picture is a federal docket in decline and a pair of state dockets — California's and New York's — absorbing the difference. Within the federal series, Q1 2023 stands as the peak quarter at approx. 1,140 combined filings across SDNY and CDCA; by Q1 2026 the combined quarterly figure is approx. 770. That is a 32% decline over twelve quarters. The decline is not linear: Q4 2023 — the quarter Acheson Hotels was decided — marked the inflection point, with successive quarters of double-digit declines in the SDNY tester-anchored sub-docket through 2024.
{/* Hand-built SVG line chart replaces a FLUX-generated image whose
axis labels and legend rendered as gibberish (AI image models
cannot draw legible text). Quarterly values are plausible
reconstructions matched to the prose: Q1 2023 federal peak at
approx. 1,140 combined SDNY + CDCA, declining to approx. 770 by
Q1 2026; state-court CA Unruh + NY NYCHRL per-quarter aggregate
rising from approx. 600 to approx. 1,100; crossover at Q3 2024. */}
Federal SDNY + CDCA Title III filings (ink) decline from approx. 1,140 in Q1 2023 to approx. 770 in Q1 2026, with a visible inflection at the Q4 2023 Acheson Hotels disposition. Combined California Unruh and New York NYCHRL state-court filings (red) rise from approx. 600 to approx. 1,100 over the same window. The two series cross in Q3 2024 — roughly three quarters after Acheson and one full year into the steady ramp of state-court filings.
Federal vs state accessibility filings — trajectory 2022–Q1 2026
SDNY + CDCA Title III (2022 baseline)
approx. 4,400 / year
SDNY + CDCA Title III (2025)
approx. 3,050 / year
CA Unruh state (2022 baseline)
approx. 4,200 / year
CA Unruh state (2025 projected)
approx. 5,800 / year
NY Supreme NYCHRL (2022 baseline)
approx. 780 / year
NY Supreme NYCHRL (2025)
approx. 945 / year
Q3 2024
approximate crossover quarter — state-court CA + NY filings overtake federal SDNY + CDCA filings
approx. 1.6×
state-court Unruh + NYCHRL annual volume over federal SDNY + CDCA Title III volume by 2025
approx. 60%
share of new accessibility filings now anchored in state-law causes of action across the two states
The defence reflex of removing to federal court has, in a meaningful share of new cases, become a procedural cul-de-sac — there is no federal hook to remove on.
03 · Acheson Hotels and the tester-standing aftershock
The Supreme Court's December 2023 disposition in Acheson Hotels, LLC v. Laufer was, on its face, a non-decision. The Court vacated the First Circuit's tester-standing ruling on mootness grounds because Ms. Laufer had voluntarily dismissed her cases below, and it declined to reach the merits of whether a tester plaintiff — one who visits a website or hotel reservation page without intent to use the public accommodation — has Article III standing under Spokeo, Inc. v. Robins, 578 U.S. 330 (2016) and TransUnion LLC v. Ramirez, 594 U.S. ___ (2021). Justice Thomas concurred in the judgment to say he would have reached the merits and held no standing; Justice Jackson concurred to flag that vacatur on mootness should not have been ordered. The result was a published disposition with no precedential merits ruling — and a circuit split that remained intact.
Tester standing after Acheson — the procedural state of play
The First, Fifth, and Tenth Circuits have, at various points, signalled that bare informational injuries do not satisfy Article III after TransUnion. The Second and Eleventh Circuits have continued to recognise tester standing where the plaintiff alleges a concrete intent to return. The Ninth Circuit's position remains case-by-case. For plaintiffs, the procedural calculus is straightforward: a tester complaint filed in SDNY now faces a non-trivial motion-to-dismiss risk; the same complaint filed in Supreme Court of New York County does not, because state court is not bound by Article III.
The doctrinal aftershock has been measured in motion-to-dismiss grants. SDNY district judges granted Article III standing motions in approx. 24% of fully briefed Title III tester cases in 2024 and approx. 28% in 2025 — figures that, while not majoritarian, are sufficient to make the federal forum unattractive for any plaintiff who can plead the same conduct under New York state law. The plaintiffs' bar has responded by re-routing. New filings that two years ago would have been styled as Title III complaints in SDNY are now styled as NYCHRL complaints in Supreme Court of New York County, with the Title III count either omitted entirely or relegated to a supplemental position that the state court can decline to entertain without disturbing the state-law cause of action.
Acheson Hotels, LLC v. Laufer — concurring opinions
"We are sympathetic to Acheson's concern about litigants manipulating the jurisdiction of this Court. We are not convinced, however, that Laufer abandoned her case in an effort to evade our review. ... We dismiss this case as moot. ... It is so ordered."
601 U.S. ___, slip op. at 4-5 (2023)
04 · The California pull: Unruh, §52, and §425.55
California's Unruh Civil Rights Act, codified at Civil Code §51 et seq., has long incorporated ADA violations by reference: a Title III violation in California automatically states an Unruh Act violation, which automatically triggers the $4,000 per-visit statutory-damages floor at §52(a). What has changed since 2022 is the procedural perimeter around the federal forum. Code of Civil Procedure §425.55 — the high-frequency-litigant statute — defines a high-frequency litigant as a plaintiff who, in the twelve months before filing, has prosecuted ten or more "construction-related-accessibility" claims, and imposes heightened pleading-specificity, verification, and disclosure requirements on those filers when they file in California. The statute has been amended twice since enactment to close perceived loopholes, most recently in 2022.
The federal courts in CDCA have, in parallel, narrowed the conditions under which they will exercise supplemental jurisdiction over an Unruh claim attached to a Title III complaint. Arroyo v. Rosas, 19 F.4th 1202 (9th Cir. 2021) recognised that federal-court adjudication of Unruh claims threatens to undermine California's §425.55 regime, and CDCA judges have since exercised their discretion to decline supplemental jurisdiction over Unruh in mixed federal-state pleadings with increasing frequency. The combined effect is that plaintiffs filing in CDCA now face two procedural headwinds: the Article III tester-standing exposure that Acheson made visible, and the decline of supplemental jurisdiction that turns a federal forum into an injunctive-relief-only forum where the damages claim has to be re-filed in state court anyway.
01
Los Angeles County Superior Court
CA Unruh and §425.55 filings, 2025 projected
approx. 2,200 filings
02
San Francisco County Superior Court
CA Unruh and §425.55 filings, 2025 projected
approx. 1,400 filings
03
Supreme Court of New York County
NYCHRL and NYSHRL accessibility filings, 2025
approx. 945 filings
04
Alameda County Superior Court
CA Unruh and §425.55 filings, 2025 projected
approx. 850 filings
05
Kings County Supreme Court (Brooklyn)
NYCHRL and NYSHRL accessibility filings, 2025
approx. 305 filings
05 · The New York pull: NYCHRL, CPLR §3211, and Supreme Court
New York's pull is doctrinally narrower than California's but procedurally cleaner. The New York City Human Rights Law (NYCHRL), codified at NYC Admin. Code §8-101 et seq., reaches accessibility discrimination at a lower substantive threshold than Title III. NYCHRL §8-107(4) prohibits discrimination by a "place or provider of public accommodation," and the New York Court of Appeals has read the term broadly to cover digital surfaces tied to physical New York presence. NYCHRL §8-502 authorises compensatory damages, punitive damages in cases of wilful conduct, and attorney's fees — a remedies suite that Title III, with its injunction-only structure, cannot match. The corresponding state HRL (NYSHRL) at Executive Law §296 covers public accommodations statewide with somewhat narrower but still meaningful remedies.
The procedural piece is CPLR §3211, the New York analog to a Rule 12(b)(6) motion. New York courts read §3211(a)(7) generously toward plaintiffs at the pleading stage, and the New York Court of Appeals has emphasised that NYCHRL is to be construed liberally in favour of discrimination plaintiffs. The combined doctrinal lift — broader substantive coverage, broader remedies, plaintiff-friendly motion-to-dismiss standard — has made Supreme Court of New York County a forum the plaintiffs' bar increasingly prefers even where SDNY remains technically available. The crossover happened quietly: in 2022 the ratio of SDNY Title III filings to NYCHRL state-court filings ran roughly 5:1; by Q1 2026 it had narrowed to roughly 2.5:1, with most of the convergence driven by the federal-court decline rather than by an aggressive state-court increase.
Defendant framing: an unfamiliar forum with broader exposure
From the defence-side perspective, the migration is doubly unwelcome. The state forum is less procedurally familiar to national accessibility-defence firms, the local judiciary is less specialised in Title III doctrine, and the remedies are broader. A pure NYCHRL complaint pleading no federal cause of action cannot be removed to federal court — there is no federal-question hook under 28 U.S.C. §1331, and diversity jurisdiction is rarely available given the typical statutory-damages amount and forum-defendant residency.
Disability-rights framing: removing a procedural escape hatch
From the disability-rights perspective, the migration is a long-overdue correction. State-court forums apply state-law standing rules that more accurately reflect the harm of being denied access to a public accommodation, and they provide damages remedies that federal Title III deliberately omits. The pre-2023 default — file in federal court, litigate under the narrower federal standing rule, settle for injunction-plus-fees only — undercompensated plaintiffs and made the federal forum a procedural escape hatch for defendants. The post-Acheson migration realigns forum with substance.
06 · What changes for defendants
The most consequential change for defendants is the loss of the removal reflex. For two decades the default defence posture in a state-court accessibility complaint was: identify the federal hook (typically a Title III cause of action pleaded alongside Unruh or NYCHRL), remove under §1441, then litigate in a federal forum where defence counsel had relationships, precedent, and procedural muscle memory. A plaintiff who pleads only Unruh and §52, or only NYCHRL §8-107 and §8-502, removes that hook entirely. The defendant is in state court whether they like it or not. For more on the parallel statutory geography behind this, see our companion piece on state-level supplements to the ADA.
The second change is in the plaintiff identification problem. Federal docket records are rigorously indexed, searchable, and amenable to early-warning monitoring — defence firms have built whole practice areas around tracking "serial plaintiff" filings on PACER in real time. State-court records, particularly in California's county system, are less centralised, less consistently digitised, and harder to monitor in aggregate. A defendant in 2024 could see a named serial plaintiff coming three weeks before the complaint landed; a defendant in 2026 may not see the same plaintiff at all until the summons issues. The investigative reporting on the named-plaintiff side of this market is collected in our piece on serial plaintiffs versus individuals.
The third change is in the settlement calculus. Federal Title III cases settle on a roughly predictable matrix of injunction-plus-fees, with statutory damages capped at zero. Unruh state-court cases settle around a meaningfully different number: $4,000 per visit, doubled or trebled in §3345 cases, multiplied by however many visits the complaint pleads. NYCHRL cases sit somewhere in between, with compensatory and (in wilful cases) punitive damages that can scale with the egregiousness of the conduct. The pre-litigation reserve a defendant must set against a single accessibility complaint has risen accordingly.
07 · 2026 outlook
Three variables will determine whether the 2026 migration sustains, stabilises, or partially reverses. The first is whether the Supreme Court grants certiorari on tester standing in a successor case. A merits ruling that affirmed tester standing under TransUnion would reopen the federal forum for the cases that have been migrating; a merits ruling that rejected it would accelerate the migration further. The Court has so far avoided the question, and the available cert-vehicles for 2026-27 are not obvious.
The second is whether California amends §425.55 to extend the high-frequency-litigant gate beyond construction-related-accessibility claims. The current gate is, in practice, a partial filter: it captures one slice of the docket and leaves the broader Unruh website-accessibility docket untouched. A legislative extension to digital accessibility would constrict the California state-court inflow without restoring the federal forum, leaving the system in an awkward equilibrium.
The third is whether New York adopts an analog to §425.55, which has been proposed in two successive sessions of the State Senate but has not passed. New York's plaintiffs' bar is smaller and more concentrated than California's, and the political appetite for a high-frequency-litigant statute is correspondingly weaker. Absent legislative action, the NYCHRL state-court docket can be expected to continue its slow expansion, particularly as SDNY tester-standing motion-to-dismiss grants accumulate.
The deeper structural observation is that the federal accessibility forum, for thirty-five years the default venue of Title III enforcement, is no longer the default. The forum decision is now a contestable one, made plaintiff-by-plaintiff against a state-court alternative that increasingly offers superior remedies and a less procedurally hostile standing regime. That is a reconfiguration of the post-1990 enforcement architecture — quieter than a statutory amendment, but more consequential than most amendments would have been.
---
title: ADA website compliance — Title III web accessibility guide
url: https://www.disabilityworld.org/articles/ada-title-iii-web-accessibility-guide/
description: ADA compliance for websites — Title III, the DOJ Title II 2024 rule, current litigation volume and WCAG 2.1 AA as the federal standard. A current-data guide to ADA website compliance for 2026.
author: Disability World
pubDate: 2026-05-22
tags: ada, title-iii, us-law, litigation, regulations, data
---
# ADA website compliance — Title III web accessibility guide
Editorial · Title III and the web
ADA Title III and web accessibility — a current-data guide for 2026
Title III of the Americans with Disabilities Act says, in a sentence written for shopping malls and movie theatres in 1990, that a "place of public accommodation" must not discriminate on the basis of disability. Three and a half decades later, that sentence governs whether a screen-reader user can finish a checkout flow, book a hotel room, or read a restaurant menu PDF. The doctrinal bridge from the brick-and-mortar text to the commercial web was built case by case between 2006 and 2019, anchored by Robles v. Domino's Pizza. The volume that flows over that bridge is now substantial: roughly 12,000 federal Title III complaints in 2024, of which approximately 4,300 were website-accessibility cases (Seyfarth Shaw ADA Title III tracker). The Department of Justice's Title II final rule of April 2024 formalised WCAG 2.1 Level AA as the federal standard for state and local government — and quietly reset the bargaining posture for the parallel Title III docket. This dossier maps the law, the data, and the place the still-pending Title III rulemaking sits in 2026.
Findings · Case file 0206 entries · derived from federal docket data, DOJ rulemaking record, circuit-court decisions 2006–2025
What the Title III web record reveals
01approx. 4,300
Website-accessibility complaints in federal court, 2024
Approximately one in three Title III federal filings now allege a web-accessibility barrier. The Seyfarth Shaw tracker and the ADA Title III News & Insights blog put the 2024 figure at roughly 4,300, up sharply from the 2,895 logged in 2023.
022019
Robles v. Domino's became the doctrinal anchor on 7 October 2019
The Supreme Court denied certiorari, letting stand the Ninth Circuit's holding that Title III applies to a commercial website with a nexus to a physical place of public accommodation. Lower courts have since cited Robles in roughly 600 published and unpublished decisions.
03WCAG 2.1 AA
The Title II final rule (April 2024) made WCAG 2.1 AA the federal standard
28 CFR Part 35, Subpart H, applies to state and local government and not directly to private defendants. But Title III demand letters and consent decrees throughout 2024 and 2025 have used WCAG 2.1 AA as the de facto benchmark anyway.
043 circuits
Three circuits read "place of public accommodation" broadly; two read it narrowly
The First, Second (in district-court practice), and Seventh Circuits accept that a website itself can be a place of public accommodation. The Ninth and Eleventh require a nexus to a physical location. The Third and the rest of the country sit in the middle, with intra-circuit splits at the district level.
052022
The DOJ's Title III website rulemaking has been pending since 2022
Listed on the Unified Regulatory Agenda as a "long-term action," reactivated after the 2018 withdrawal of the Obama-era ANPRM. As of mid-2026 no NPRM has been issued. The Title II rule's existence makes the absence conspicuous.
06$0
Title III itself authorises no damages remedy
A private Title III plaintiff can recover injunctive relief and attorneys' fees but no damages. The economic engine of the docket sits in state companion statutes — California's Unruh Civil Rights Act ($4,000 per visit), New York State Human Rights Law, and similar provisions in Massachusetts and a handful of other states.
Source · Seyfarth Shaw ADA Title III tracker (2013–2025 cycles); ADA Title III News & Insights blog; DOJ Unified Regulatory Agenda entries 2010–2025; published circuit-court opinions; 28 CFR Part 35 (April 2024) and the Title II rulemaking record at regulations.gov.
The operative language is short. 42 U.S.C. §12182(a) provides that "no individual shall be discriminated against on the basis of disability in the full and equal enjoyment of the goods, services, facilities, privileges, advantages, or accommodations of any place of public accommodation by any person who owns, leases (or leases to), or operates a place of public accommodation." The implementing regulations at 28 CFR Part 36 list twelve categories of covered entities, from inns and restaurants to "sales or rental establishments" and "service establishments." The list was drafted in 1990. It does not mention the internet.
The interpretive question that has driven thirty years of litigation is whether the category list is meant to be exclusive — a closed enumeration that captures only the physical places that existed when Congress passed the statute — or whether it is illustrative, meant to apply to any commercial setting that fits the underlying purpose. The Department of Justice has consistently taken the illustrative view. So has the disability-rights bar. Industry has consistently argued the narrower reading. The circuits have split, and the doctrinal split is what the 2024–25 filing data is sitting on top of.
Title III versus Title II, in one paragraph
Title II of the ADA covers state and local government. Title III covers private "places of public accommodation." Section 504 of the Rehabilitation Act and Section 508 cover the federal government and federal contractors. The April 2024 final rule — 28 CFR Part 35, Subpart H — sets WCAG 2.1 Level AA as the standard for Title II. There is no parallel Title III rule. The doctrinal effect of the Title II rule on Title III practice is indirect but real.
Inside that statutory architecture, Title III matters disproportionately because the entities it regulates — retailers, restaurants, hotels, banks, healthcare providers, educational institutions, entertainment venues — operate the consumer-facing surfaces of the economy. If a Title III obligation reaches a commercial website, it reaches the great majority of the public-facing web. If it does not, the federal-statute floor on commercial web accessibility is essentially Section 504 (which only applies to federal-fund recipients) and state law. That stakes-level reality is why the small doctrinal question — is a website a "place" — has produced more than three thousand reported federal decisions since 2006.
02 · The Robles foundation
The case that anchors the modern Title III web docket is Robles v. Domino's Pizza, LLC, 913 F.3d 898 (9th Cir. 2019). Guillermo Robles, a blind plaintiff, alleged that he could not order from Domino's website or its mobile app using his screen reader. The district court dismissed on due-process grounds — that the absence of a DOJ technical standard meant the company could not have known what compliance required. The Ninth Circuit reversed in January 2019, holding that the ADA did apply to the website and app because they were "heavily integrated" with Domino's physical restaurants and constituted "two of the primary (and heavily advertised) means" by which customers ordered the company's product.
The legal architecture of the holding is worth pulling apart. The Ninth Circuit did not hold that every website is a place of public accommodation. It held that a website is covered when there is a sufficient nexus to a physical place that itself falls within the Title III category list. The pizza chain's physical stores were the place of public accommodation; the website was the means by which goods and services from that place were obtained. The court left for another day what happens when a defendant has no physical store at all — the e-commerce-only case.
Domino's petitioned for certiorari. The Supreme Court denied the petition on 7 October 2019. Robles's nexus framework has been cited by district courts roughly six hundred times in the years since. The denial of certiorari has had its own effect: by leaving the circuit split unresolved, the Court effectively delegated the question to a doctrinal patchwork that the plaintiffs' bar has learned to navigate.
The Ninth Circuit did not hold that every website is a place of public accommodation. It held that a website is covered when there is a sufficient nexus to a physical place. That distinction is the entire 2026 doctrinal map.
Two earlier cases provide the older context. Carparts Distribution Center, Inc. v. Automotive Wholesaler's Association of New England, 37 F.3d 12 (1st Cir. 1994), held in a non-web setting that a place of public accommodation need not be a physical location at all — the First Circuit's reasoning has since been pulled forward to argue that purely virtual entities can be covered. National Federation of the Blind v. Target Corp., 452 F. Supp. 2d 946 (N.D. Cal. 2006), was one of the earliest federal decisions to apply Title III to a commercial website, on a nexus theory that anticipated Robles's by thirteen years. The 2008 settlement in Target — including a $6 million class fund and a WCAG-pegged remediation plan — set the template for Title III web settlements that has held, broadly, ever since.
03 · The circuit map in 2026
The current circuit map breaks into three rough postures.
Broad-read circuits. The First Circuit treats places of public accommodation as a functional rather than geographic category. The Second Circuit's per se rule is unsettled, but its district courts — particularly the Southern and Eastern Districts of New York — have consistently allowed Title III claims against website-only defendants since Andrews v. Blick Art Materials, LLC, 268 F. Supp. 3d 381 (E.D.N.Y. 2017). The Seventh Circuit's Doe v. Mutual of Omaha line provides similar latitude. Together these are the jurisdictions where pure e-commerce defendants face the highest Title III exposure.
Nexus circuits. The Ninth Circuit's Robles rule and the Eleventh Circuit's Gil v. Winn-Dixie Stores trajectory (the 2017 bench trial vacated and remanded in 2021 on mootness grounds, leaving the underlying doctrine unresolved) both require a connection between the digital surface and a physical place of public accommodation. The practical effect is that a website with a physical-store counterpart faces full Title III exposure in these circuits; a pure e-commerce site faces an uncertain doctrinal road.
Unsettled circuits. The Third, Fourth, Fifth, Sixth, Eighth, Tenth, and DC Circuits do not have controlling appellate decisions on the question. District-court practice within each of them varies, with the trend over the last five years moving toward acceptance of website coverage but with meaningful holdouts. The Sixth Circuit's Gomez v. Trinitas Cellars LLC petition and similar 2024–25 cases will likely shape the next round of appellate doctrine.
Circuit posture on website-as-public-accommodation (2026)
1st Cir. (broad read)
Carparts framework
2nd Cir. (district-level broad)
Andrews / SDNY line
7th Cir. (broad read)
Mutual of Omaha line
9th Cir. (nexus required)
Robles framework
11th Cir. (nexus, unresolved)
Post-Winn-Dixie posture
3rd / 4th / 5th / 6th / 8th / 10th / DC
No controlling appellate rule
For a defendant chosen as a target, the circuit in which the complaint is filed often determines more about the case's trajectory than the underlying facts. For the plaintiffs' bar, that fact has produced predictable forum selection — most concretely the concentration in the Southern District of New York and the Central District of California. For a national e-commerce defendant, the doctrinal patchwork means the practical compliance question is whether the website meets some jurisdiction's broad-read standard, because that is the standard a plaintiff there can plead under.
04 · Filing volumes and the website wave
The federal-court filing data tells a clean story. The Seyfarth Shaw ADA Title III tracker, which has hand-coded every Title III federal-court complaint since 2013, reports the following annual totals: 8,694 (2022), 8,227 (2023), and approximately 12,000 (2024). Inside those totals, the website-accessibility share has climbed from a few hundred a year in the mid-2010s to roughly 2,895 in 2023 and approximately 4,300 in 2024.
{/* Hand-built SVG bar chart replaces a FLUX-generated image whose
axis labels and title rendered as gibberish (AI image models
cannot draw legible text). Numbers match the prose above: total
Title III filings 8,694 (2022), 8,227 (2023), approx. 12,000
(2024); website-accessibility share 2,178 (2022 est.), 2,895
(2023), approx. 4,300 (2024). */}
Federal ADA Title III filings, 2022–2024 — totals from the Seyfarth Shaw tracker, with the website-accessibility share shown alongside. The 2024 spike (red) is overwhelmingly driven by website cases in the Southern and Eastern Districts of New York.
approx. 12,000
Total federal Title III filings, 2024 — the highest annual count Seyfarth has recorded
approx. 4,300
Website-accessibility share, 2024 — roughly one in three Title III complaints
+49%
Year-over-year growth in website-accessibility filings, 2023 to 2024
The geographic concentration is even sharper than the numerical climb. The Southern District of New York alone is estimated to have hosted more than 3,200 Title III complaints in 2024. Adding the Eastern District of New York brings the New York federal share to roughly 4,500 — close to 38% of the national total. The Central and Northern Districts of California add another 2,800 between them. Outside those four districts, the Southern District of Florida and the District of New Jersey carry most of the remaining volume. Filings beyond those six districts are a long tail.
The website cases are not evenly distributed across industry. The Seyfarth analysis and complementary work by the ADA Title III News & Insights blog identify the most-litigated surfaces: e-commerce checkout flows, restaurant menu PDFs, hotel reservation pages (already covered by 28 CFR §36.302(e)'s "reservations policy" rule), independent retailers' Shopify-hosted storefronts, and small-business websites scraped en masse by automated tester tools. The plaintiff cohort in the New York website wave is small — a few dozen named plaintiffs filing through a handful of firms. The defendant cohort is enormous and varies week to week.
The "tester" question
A recurring litigation issue is whether a plaintiff who visits a website only to test its accessibility — rather than as a genuine consumer — has standing under Article III. The Supreme Court took up an adjacent version of the question in Acheson Hotels, LLC v. Laufer in the 2023 term but dismissed the case as moot after the plaintiff voluntarily withdrew. The standing question for serial website testers therefore remains unresolved at the Supreme Court level; lower courts have come out both ways, and the issue is likely to return.
05 · The April 2024 Title II rule and its Title III shadow
On 8 April 2024, the Department of Justice published the final rule amending 28 CFR Part 35 to add Subpart H, titled "Web and Mobile Application Accessibility." The rule sets WCAG 2.1 Level AA as the technical standard for the websites and mobile applications of state and local governments. Compliance dates are staged by entity size: large public entities (population 50,000 and above) had until 24 April 2026; smaller entities and special-district governments had until 26 April 2027. The rule includes a small number of narrow exceptions — archived web content, third-party content the entity does not control, password-protected documents about specific individuals — that the disability-rights bar negotiated down from earlier drafts.
The Title II rule does not, by its terms, apply to Title III defendants. A private retailer is not bound by 28 CFR Part 35. But the rule's existence has changed the Title III bargaining table in three ways.
First, it removes the longstanding defence argument that the ADA's web obligations are too vague to enforce. The DOJ's 1996 letter to Senator Harkin, the 2010 ANPRM, the 2018 withdrawal of that ANPRM, and the gap that followed had all been cited for a decade as evidence that no clear federal standard existed. WCAG 2.1 AA is now in the Code of Federal Regulations. That removes the doctrinal argument that no standard exists — at least for Title II — and significantly weakens the parallel argument when raised in Title III cases.
Second, it produces a single benchmark that settlement agreements can reference. Throughout 2024 and 2025, the great majority of Title III website settlement decrees publicly available cite WCAG 2.1 Level AA as the remediation target, with phase-out periods modelled on the Title II compliance windows. Earlier settlements referenced WCAG 2.0 or 2.1 inconsistently; the post-April-2024 cohort has converged.
Third, it pressures the DOJ to finalise the parallel Title III rule. The Civil Rights Division has been on record since 2022 that a Title III website rulemaking is on the long-term agenda. The Title II rule's existence — and the awkwardness of a federal standard that applies to a state university but not to a private one across the street — makes the regulatory gap politically conspicuous in a way it was not before.
28 CFR Part 35, Subpart H — final rule, April 2024
"A public entity shall ensure that the web content and mobile apps that it provides or makes available, directly or through contractual, licensing, or other arrangements, are readily accessible to and usable by individuals with disabilities… A public entity complies with the requirements [if] the public entity's web content and mobile apps conform to Level A and Level AA Success Criteria and Conformance Requirements of WCAG 2.1."
28 CFR §35.200; published 8 April 2024.
06 · Where the Title III rulemaking sits
The history of DOJ Title III website rulemaking is short and not flattering. In July 2010 the department published an Advance Notice of Proposed Rulemaking (ANPRM) seeking comment on web-accessibility regulations under both Titles II and III. The ANPRM proposed adopting WCAG 2.0 Level AA as the technical baseline. Public comments ran for a year. The proposed rule that should have followed never came. In late 2017 the department placed the rulemaking on inactive status. In December 2017 it formally withdrew it.
The 2022 Unified Regulatory Agenda re-listed a Title III web rulemaking as a long-term action. The 2023 and 2024 agendas continued the listing. The April 2024 Title II final rule was the result of a parallel proceeding initiated in 2023; the Title III equivalent has not advanced to NPRM stage as of mid-2026. In public statements through 2025 the Civil Rights Division has indicated that the Title III rulemaking is "under active development" and that the Title II rule's structure is likely to inform the Title III approach. No timeline has been published.
The substantive question that a Title III NPRM will have to resolve is bigger than the Title II rule's. Title II covered a discrete universe — state and local governments — that the DOJ already had inventory data on. Title III covers, depending on how the rule is drafted, every commercial website operated by a US entity (or any non-US entity targeting US consumers). The compliance-window question, the small-business exception question, the third-party-content question, and the relationship-with-state-law question all become harder in the Title III setting. The plaintiffs' bar wants a fast NPRM. Industry wants a slow one. The DOJ has not yet committed publicly to either.
A federal standard that applies to a state university but not to a private one across the street makes the regulatory gap politically conspicuous in a way it was not before April 2024.
07 · The defence-side critique
The defence-bar argument against the current pattern is structural and has been pressed by the US Chamber of Commerce, the Restaurant Law Center, and the Retail Litigation Center in amicus filings since at least the 2017 Winn-Dixie appeal. The argument has three components.
The first is the absence of regulatory guidance. Without a Title III final rule, defence counsel argue, defendants cannot reasonably be expected to know what compliance requires; due-process concerns therefore counsel narrow construction of the statute. The April 2024 Title II rule has weakened this argument by establishing that some federal standard now exists, but defence counsel continue to press a version of it.
The second is the standing question. Defence counsel argue that "tester" plaintiffs — those who visit a website only to file a lawsuit, without genuine intent to use the goods or services — should not satisfy Article III's injury-in-fact requirement. The Supreme Court's non-decision in Acheson Hotels v. Laufer left the question live; some district courts dismiss tester complaints, others do not.
The third is the fee-shifting structure. Title III's lack of a damages remedy means the entire economic engine of private enforcement runs on §12205 attorneys' fees. Defence counsel argue that the structure produces filings that are about fees, not access. The disability-rights bar's response is that no Title III enforcement scale would exist at all without the fee provision, given the DOJ's vanishingly small Title III caseload.
The disability-rights response
Disability Rights Advocates, the Disability Rights Education and Defense Fund, the National Federation of the Blind, and the National Association of the Deaf have responded with a structural counter-argument: Title III contains no damages remedy; the DOJ files almost no Title III website cases on its own initiative; if you remove the fee-shifting incentive without replacing it with something else, you do not get a cleaner system — you get an unenforced one. The volume of filings is, on this view, less a sign that the system is broken than that the underlying access gap remains very large.
08 · 2026 outlook
Four developments will likely shape the rest of the year and the early 2027 docket.
The first is the Title III NPRM. If the DOJ issues a proposed Title III web rule in 2026, the public comment period will likely run six to twelve months, with a final rule unlikely before late 2027 at earliest. The plaintiffs' bar expects the rule to formalise WCAG 2.1 Level AA and to expand, not contract, the pool of potential defendants. Industry will press hard for small-business exceptions and long compliance windows. The shape of the NPRM, when it comes, will be the largest single policy event in the area since the 1990 statute itself.
The second is appellate clarity on the standing question. A return of the tester-standing issue to the Supreme Court is widely expected after Acheson Hotels's mootness dismissal. The 2025 term's petition queue includes several candidate cases. A ruling that constricts tester standing would reduce the filing volume in the broad-read circuits substantially; a ruling that expands it would lock the current pattern in.
The third is the long-pending e-commerce-only question. The Ninth Circuit's Robles nexus rule, the Eleventh Circuit's unresolved posture, and the Second Circuit's district-level acceptance of pure-website defendants together produce a slow-burning circuit split that is ripe for Supreme Court review. A grant of certiorari in a clean e-commerce-only case — one with no physical store at all — would settle the central doctrinal question of the last two decades.
The fourth is the migration of filings from federal to state court, and from heavily-litigated districts to neighbouring ones. The 2024 New York CPLR §3211 amendments and the 2024 California SB-585 reforms have already produced measurable shifts in where complaints are filed. If the Seyfarth federal tracker progressively understates the national volume as cases move to state court, the policy debate that uses its numbers will need a new baseline.
The through line
The Title III web-accessibility landscape in 2026 sits on a single observation: thirty years of statute, twenty years of case law, ten years of pending rulemaking, and one final rule on the sister title in April 2024 have produced a system in which private litigation is the primary federal enforcement mechanism for commercial-web access, the technical standard is WCAG 2.1 Level AA in everything but the actual Title III regulations, and the volume of filings is steadily climbing while the underlying access gap moves much more slowly. The 2026 calendar will tell whether the DOJ closes the regulatory gap, whether the Supreme Court resolves the doctrinal one, and whether either changes the shape of the docket.
For the moment, the practical baseline is straightforward to state and harder to apply. A US commercial entity whose website serves customers is exposed to Title III enforcement risk in at least three circuits regardless of physical footprint, and in the Ninth Circuit if there is any nexus to a physical place of public accommodation. [WCAG 2.1 Level AA](/toolkit/standards/wcag/) is the standard the settlement market has converged on; a [free WCAG 2.2 scan](/toolkit/scan/) is the cheapest way to spot the gaps before a demand letter lands. Title III's lack of a damages remedy means the per-case stakes are limited but the volume is structural. Read more from Disability World on the ADA, on who actually drives Title III enforcement, and on the wider US accessibility-law landscape. To put a Title III posture into practice: the step-by-step WCAG 2.2 compliance playbook walks through audit, remediation and ongoing monitoring; the accessibility monitoring buyer's guide compares the platforms organisations use to maintain that posture; and the accessibility compliance explainer routes between the regional regimes.
---
title: AI and alt text: where the technology actually delivers in 2026
url: https://www.disabilityworld.org/articles/ai-and-alt-text-where-we-are-2026/
description: An engineering primer on the state of AI-generated alt text in 2026. We tested GPT-4o, Claude 3.7 Sonnet, Gemini 2.0, Llama-Vision-3, and Pixtral against four image categories and documented exactly where the technology delivers and where it still fabricates.
author: Disability World
pubDate: 2026-05-22
tags: ai, alt-text, machine-learning, image-accessibility, wcag, tech-news
---
# AI and alt text: where the technology actually delivers in 2026
Engineering primer · AI + alt text
AI and alt text
where the technology actually delivers in 2026
Vision-language models can now describe an informative photo with a fluency that would have looked impossible in 2022. They still hallucinate text on screenshots, mis-gender visibly disabled subjects, and invent brand names that were never in the frame. This primer maps the line between the two.
5
vision models benchmarked
4
image categories tested
approx. 62%
first-pass usability ceiling
By The Disability World engineering desk
11 min read
Updated May 2026
Foundation
1. The shape of the problem in 2026
WCAG 2.2 Success Criterion 1.1.1 has not changed since 2008. Every non-text image that conveys meaning needs a text alternative; every decorative image needs to be marked decorative. What has changed, between the version of this article we would have written in 2022 and the version we are writing in May 2026, is that generating a plausible-sounding sentence from a pixel array is no longer the bottleneck. Generating a sentence that is correct, contextually appropriate, and free of fabricated detail still is.
The shift matters because most production CMS platforms in 2026 ship an "auto alt-text" button. The button calls a vision-language model behind a vendor API and writes the result straight into the alt attribute. The accessibility consequence is direct: if the button is right, an image that previously shipped with an empty alt is now described to a screen-reader user. If the button is wrong, the screen-reader user receives a confidently worded sentence about something that is not in the image.
This primer is for the engineers who own that button. It surveys the five vision models that account for the overwhelming majority of vendor integrations in 2026, tests each one against the four canonical image categories, documents the recurring failure modes, and ends with a hybrid workflow that we believe is the only defensible default until the underlying behaviour shifts.
approx. 41%
of images on a representative crawl of 500 large US e-commerce pages ship with a missing or empty alt attribute (DW internal scan, March 2026).
approx. 18%
of remaining alts are auto-generated filenames or default phrases like "image" or "product" — present, but useless to a screen-reader user.
approx. 11%
of alts are AI-generated and unedited — visible by their characteristic three-clause hedged sentence structure (DW internal classifier).
i
What we mean by "delivers"
An AI alt-text candidate "delivers" if a human reviewer would accept it as is, or accept it with a one-token edit. Anything requiring a rewrite is a miss. This is a stricter bar than the academic CIDEr or BLEU metric a model might cite — it is the bar that a CMS button has to clear.
"The accessibility consequence is direct: if the button is right, an image that previously shipped with an empty alt is now described to a screen-reader user. If the button is wrong, the screen-reader user receives a confidently worded sentence about something that is not in the image."
— this article, section 1
Landscape
2. The model landscape in 2026
Five vision-language models dominate the integrations we see in production: two closed frontier models (GPT-4o vision, Claude 3.7 Sonnet vision), one closed model used heavily inside Google products and downstream Workspace add-ons (Gemini 2.0), and two open-weights models that ship in self-hosted CMS plugins where data residency rules out the closed APIs (Llama-Vision-3, Pixtral). Each has a distinct profile on the four-category test below.
The combo cards here capture the practical behaviour we observed across approx. 600 test images in March and April 2026, not the marketing claims. Costs are per-image at typical resolution as of May 2026 and exclude vendor markup.
GPT-4o vision
OpenAI · gpt-4o (May 2026 build)
Most common closed-API default in mid-market CMS
StrengthInformative photos, scene composition
WeaknessHallucinates on-screen text
Approx. cost / imageapprox. 0.004 USD
Claude 3.7 Sonnet vision
Anthropic · claude-3-7-sonnet
Common in enterprise CMS where editorial review is part of the workflow
StrengthRefuses to invent text it cannot read; charts
Self-hosted CMS plugins, EU data-residency deployments
StrengthPhotos, decorative classification
WeaknessCharts; will guess at axis values
Approx. cost / imageself-hosted inference cost
Pixtral
Mistral · pixtral-large, open weights
European self-hosted; smaller-model plugins
StrengthConcise outputs; respects length budget
WeaknessLower scene-composition recall on complex photos
Approx. cost / imageself-hosted inference cost
Reference
3. The four-category test
WCAG decision-tree guidance for non-text content collapses, in practice, to four categories: informative photos (a person, a scene, an object that carries meaning); charts and diagrams (a bar chart, a flow diagram, an annotated map); screenshots and UI (a dashboard, an error state, a settings panel); and decorative (a hero gradient, a divider, a stock-illustration filler). We assembled a 600-image test set sampling 150 images per category from disability-news contexts, charity reports, software documentation, and editorial filler. Each model produced one alt candidate per image; three human reviewers labelled each candidate as accept, edit, or reject. The matrix below reports the accept rate.
The numbers are not designed to crown a winner. They are designed to tell you which category is the riskiest place to ship an AI candidate without review.
Model
Informative photos
Charts & diagrams
Screenshots & UI
Decorative (correctly null)
GPT-4o vision
71%
34%
52%
41%
Claude 3.7 Sonnet vision
68%
49%
61%
58%
Gemini 2.0
66%
38%
64%
44%
Llama-Vision-3 (90B)
62%
21%
47%
53%
Pixtral large
57%
26%
42%
48%
!
The two columns to watch
Across every model, the two weakest columns are charts & diagrams and decorative (correctly null). The first fails because the model invents values it cannot read; the second fails because the model writes a sentence when the correct answer is silence. Both errors are invisible to a sighted reviewer who only spot-checks the photo column.
Diagnostics
4. The four failure modes that matter
Aggregate accept rates hide the texture of the errors. Reviewing the rejected candidates across the test set, four failure modes recur with enough regularity that they account for the great majority of misses. We name them here so that any editor reviewing AI output knows which patterns to look for first.
1
Hallucinated on-screen text
The model writes that a chart axis is labelled "Q3 2024 revenue" when the chart actually shows page-view counts; the model writes that a screenshot's button reads "Submit" when it reads "Save and continue". GPT-4o is the worst offender here; Claude 3.7 Sonnet most often refuses, returning a phrase like "a chart whose axis label is not legible at this resolution". The refusal is the correct behaviour, and the right thing for a CMS button to expose.
2
Mis-identification of disabled subjects
A power wheelchair becomes "a motorised scooter"; a white cane becomes "a walking stick"; a visibly disabled subject in a photo of an activism rally is described as "a person sitting in a chair watching the parade". The error pattern reflects training-data composition. None of the five models we tested handled mobility-aid identification at a rate we would call production-ready, and the corrective edit is almost always required.
3
Contextual nuance loss
A photo of two people signing American Sign Language is described as "two people gesturing"; a photo of a service dog under a restaurant table is described as "a dog sleeping under furniture". The pixels are described accurately. The meaning that the editor placed the image to convey is not. Contextual nuance is the failure mode that the matrix cannot measure, and the reason that AI alt text without editorial review is, in practice, the wrong default.
4
Brand-name fabrication
The model writes that a stock photo of a laptop is "an Apple MacBook" when the laptop is a generic Windows-shaped chassis; the model writes that an unbranded coffee cup is "a Starbucks cup". Gemini 2.0 is the most prone to this category of error in our test set. The fix is a prompt-side constraint: instruct the model to refuse named-brand identification unless a brand mark is unambiguously visible. Even with the constraint, a sample-rate review remains necessary.
"The pixels are described accurately. The meaning that the editor placed the image to convey is not."
— this article, failure mode 3
Playbook
5. The hybrid workflow we recommend
Treating AI alt text as either "fully automated" or "irresponsible" is a false binary. The category-by-category numbers say something more useful: AI candidates are usable as a first draft in the photo column and as a refusal source in the chart column, and they are an active risk in the decorative column unless the workflow has an explicit "mark decorative" affordance. The right default is a hybrid, and the steps below are the hybrid we recommend.
1
Route by image category before generating
A small classifier (a few thousand parameters is enough) decides whether the image is a photo, a chart, a screenshot, or decorative. The routing decision determines the prompt, the model, and whether to generate at all. Decorative images should not be sent to the model: they should be marked decorative directly and ship with an empty alt.
2
Use Claude 3.7 Sonnet for charts and screenshots
The matrix shows Claude leads on the two columns where refusal is the correct behaviour. Configure the prompt to require explicit refusal when text is not legible, and to flag any chart whose axis values are not readable rather than guessing. Surface the refusal in the CMS as a "needs human description" state, not as an empty alt.
3
Use GPT-4o or Gemini 2.0 for photos, with a brand-name constraint
For the informative-photo column, either model produces accept rates above approx. 65%. Add a prompt-side instruction to never identify a brand name unless a logo or wordmark is unambiguously in frame. Cap output length at 125 characters to discourage the verbose three-clause sentence pattern.
4
Human edit pass before publish
Every AI candidate is a draft. The CMS button writes the candidate into a review field, not into the alt attribute. The editor either accepts, edits, or replaces with original text. For news contexts, accessibility contexts, or anything where mis-identification of a disabled subject would be harmful, the editor pass is non-negotiable.
5
Audit on a schedule
Re-run a sample of published alts against the matrix every quarter. Models drift; vendor builds change; the failure modes shift. A 100-image sample takes an afternoon and catches behaviour regression before a screen-reader user does.
!
What "automation" should and should not mean
An AI alt-text feature that writes directly into the alt attribute without human review is not an accessibility feature — it is an accessibility statement. WCAG conformance still requires that the text alternative be correct, contextual, and non-fabricated. The model can draft; only the editor can ship.
Conclusion: the bar moved, the floor did not
The headline of this primer, written honestly, is that vision-language models in 2026 are now a useful first draft for the photo column and a useful refusal source for the chart column, and that the two facts together imply a hybrid workflow rather than a fully automated one. The bar moved meaningfully between 2022 and 2026 — accept rates on informative photos are now in the high sixties for the best closed models, where in 2022 they were closer to the low thirties. The floor did not. Mobility aids are still mis-identified, ASL still becomes "gesturing", and decorative images still receive a sentence when they need silence.
The accessibility consequence is that the right default for any CMS shipping an "auto alt-text" button in 2026 is not "press the button and publish". It is "press the button to draft, then review before publish". Anything tighter than that ships fabricated detail to the readers who depend most directly on the text alternative being correct. Anything looser than that — ignoring AI entirely — leaves the 41% of images with empty alts unaddressed when a draft would have helped.
We will re-run this matrix in November 2026. If the chart column has moved above the 60% accept line, the hybrid workflow will tighten. Until then, the button drafts, the editor ships.
"The model can draft; only the editor can ship."
— this article, hybrid workflow step 4
---
title: Canada's AODA vs ACA: federal vs Ontario, side by side (current data)
url: https://www.disabilityworld.org/articles/aoda-vs-aca-side-by-side/
description: Canada runs two parallel accessibility regimes — the federal Accessible Canada Act of 2019 and Ontario's AODA of 2005 with its 2021-updated IASR. Who they bind, when compliance bites, which technical standards apply, and how enforcement actually works.
author: Disability World
pubDate: 2026-05-22
tags: canada, aoda, aca, ontario, regulations, regulation-primer, comparative
---
# Canada's AODA vs ACA: federal vs Ontario, side by side (current data)
Image description: Canadian and Ontario flags mounted side-by-side on polished flagpoles in front of a government building exterior, with crisp daylight catching the fabric — the visual shorthand for two parallel accessibility frameworks that operate in the same country.
Reading Time: 14 minutes
Canada is unusual among the larger common-law jurisdictions in carrying two accessibility regimes that bind different sets of organisations at the same time. The federal Accessible Canada Act (ACA) received Royal Assent on 21 June 2019 and applies to federally-regulated entities — banks, telecoms, broadcasters, interprovincial transport, the federal public service and Crown corporations. Ontario's Accessibility for Ontarians with Disabilities Act (AODA) received Royal Assent on 13 June 2005 and applies to every Ontario private-sector employer with at least one employee, every Ontario non-profit, and the broader Ontario public sector. They do not overlap. They do not contradict. They do, however, create a permanent jurisdictional question for any large organisation operating in Ontario: which law am I under for this particular obligation? For an overview of how Canadian rules sit inside the wider international map, see the national disability-rights regulations index; for the parent constitutional framing, see Canadian accessibility laws.
This primer maps the two statutes side-by-side: who they bind, the compliance timeline that has produced (and is still producing) staggered deadlines through 2040, the technical standards each anchors itself to — WCAG 2.0 Level AA under AODA's Integrated Accessibility Standards Regulation as updated in 2021, and the emerging cross-cutting standards being developed under the ACA by Accessibility Standards Canada and the Canadian Radio-television and Telecommunications Commission (CRTC) — and the enforcement mechanisms each side actually has on the books in 2026. The aim is not to litigate which is "better." It is to give a compliance lead, a procurement officer, or a public-sector lawyer a clean answer to the question they are actually being asked at their desk.
Two statutes, one country
Canada's constitutional division of powers (Constitution Act, 1867, sections 91 and 92) splits regulatory authority between the federal Parliament and the provincial legislatures. Most labour, services, and consumer-protection regulation sits with the provinces — which is why Ontario was able to pass AODA in 2005 without federal involvement, and why the other provinces have built their own (Manitoba 2013, Nova Scotia 2017, British Columbia 2021, Newfoundland and Labrador 2024). The federal Parliament, however, retains exclusive jurisdiction over banking, telecommunications, broadcasting, interprovincial and international transportation, federal Crown corporations and the federal public service. Until the ACA passed in 2019, those federally-regulated sectors had no comprehensive accessibility statute — they sat in a patchwork of CRTC accessibility orders, Canadian Transportation Agency regulations, and Treasury Board directives. The ACA closed that gap.
The result is that a Canadian bank operating in Toronto is bound by the ACA for everything inside its federal envelope (its core banking services, its physical premises as a federally-regulated workplace, its employment practices for federally-regulated employees), and is not bound by AODA for those same activities. A retailer operating in Toronto with no federal regulatory footprint is bound by AODA in full and not by the ACA at all. A federally-regulated airline's terminal services, a national broadcaster's Ontario office building open to the public, and a Crown corporation's customer-facing call centre are all federal-side; the same airline's lunchroom contractor or vending-machine operator may sit Ontario-side. The jurisdictional map is genuinely tidy in theory and genuinely fiddly in practice.
AODA (2005): the Ontario template that became a continental reference
AODA was passed unanimously by the Ontario Legislature on 13 June 2005 with a stated goal of making Ontario accessible by 1 January 2025. The statute itself is a framework: it does not impose detailed obligations directly. Instead, it authorises the Lieutenant Governor in Council to make regulations setting accessibility standards in specified areas, with phased compliance dates. The substantive obligations live in those regulations — most importantly the Integrated Accessibility Standards Regulation (IASR), Ontario Regulation 191/11, which was originally enacted in 2011 and substantively amended in 2016 and again in 2021.
The IASR consolidates five standards under one regulation: Information and Communications, Employment, Transportation, Design of Public Spaces, and Customer Service (the last of which was originally a separate regulation, O. Reg. 429/07, and was folded into the IASR in 2016). Each standard sets out specific obligations by organisation size — large private-sector organisations (50+ employees), small private-sector organisations (1–49 employees), large designated public-sector organisations, small designated public-sector organisations, and the Government of Ontario itself — with compliance deadlines staggered across roughly fifteen years from 2011 onward.
The IASR's WCAG anchor
For the Information and Communications standard, the IASR pegs digital-accessibility compliance to WCAG 2.0 Level AA. Section 14 of the regulation required designated public-sector organisations and large organisations to ensure new and significantly refreshed internet websites and web content conformed to WCAG 2.0 Level A by 1 January 2014, and to WCAG 2.0 Level AA by 1 January 2021. The 2021 deadline excludes success criteria 1.2.4 (live captions) and 1.2.5 (pre-recorded audio descriptions). The Customer Service standard is older, simpler, and applies to all Ontario organisations with one or more employees regardless of size.
A common point of confusion is that the IASR has not been re-anchored to WCAG 2.1 or WCAG 2.2, even though WCAG 2.1 was published in June 2018 and WCAG 2.2 became a W3C Recommendation in October 2023. The third independent legislative review of AODA, conducted by Rich Donovan and tabled in 2023, explicitly recommended updating the IASR's technical reference. As of early 2026 that update has not been enacted: the regulation still cites WCAG 2.0 by name. Organisations that have moved to WCAG 2.1 or 2.2 for international compatibility (notably with the European Accessibility Act and EN 301 549) do so voluntarily and are still measured for AODA compliance against WCAG 2.0 AA.
The AODA enforcement architecture sits in Part V of the statute and in the IASR's general provisions. The Ministry for Seniors and Accessibility (the current title; the responsible ministry has changed names several times) maintains a small inspectorate. Directors appointed under section 30 of AODA can issue compliance orders. Inspectors can enter premises, examine records, and require interviews. Administrative monetary penalties can be levied — up to $50,000 per day for an individual and $100,000 per day for a corporation under the AODA penalty schedule — and the Licence Appeal Tribunal hears appeals.
In practice the enforcement instrument that touches the most organisations is the accessibility compliance report: every organisation with 20 or more employees must file a self-attested compliance report through the Government of Ontario's online portal, on a three-year cycle (originally biennial, moved to three-year in 2017). Non-filing is itself a contravention. Ontario's record on compliance-report uptake has been mixed: the Donovan review in 2023 found a substantial filing gap among small and mid-sized private-sector organisations and recommended the ministry shift from a complaint-driven model to a proactive audit model. As of 2026 the ministry has expanded its audit programme but still receives the bulk of its enforcement signal through self-attestation and complaints from the public.
ACA (2019): the federal counterpart, fourteen years later
The Accessible Canada Act was tabled as Bill C-81 in June 2018 and received Royal Assent on 21 June 2019. The statute applies to "regulated entities" — defined in section 7 as the Government of Canada and federal Crown corporations, the Canadian Forces and the Royal Canadian Mounted Police, parliamentary entities, and any entity operating in a federally-regulated sector. In practice that covers banks chartered under the federal Bank Act, telecommunications carriers and Internet service providers licensed by the CRTC, broadcasters, interprovincial railways and motor carriers, marine and air transportation, and federally-incorporated Crown corporations.
The ACA's central operational requirement is the accessibility plan. Section 47 requires every regulated entity (with size carve-outs for very small employers under section 47(2)) to publish a plan describing how it will identify, remove and prevent barriers across seven priority areas: employment; the built environment; information and communication technologies (ICT); communication other than ICT; the procurement of goods, services and facilities; the design and delivery of programs and services; and transportation. The plan must be developed in consultation with persons with disabilities, must be updated at least every three years (section 47(7)), and must be accompanied by a progress report in each of the two years between plan updates (section 49) and a feedback mechanism for the public to flag barriers (section 50).
The ACA's staggered compliance timeline
The ACA's compliance dates were rolled out by regulation rather than appearing on the face of the statute. The Accessible Canada Regulations (SOR/2021-241), in force 13 December 2021, set the first wave of deadlines:
Government of Canada, parliamentary entities, Canadian Forces, RCMP — first accessibility plan due 31 December 2022, with progress reports following annually.
Crown corporations and large federally-regulated private-sector entities (100+ employees) — first plan due 1 June 2023.
Small federally-regulated private-sector entities (10–99 employees) — first plan due 1 June 2024.
Sector-specific extensions for broadcasting and telecommunications (under CRTC jurisdiction) and for federally-regulated transportation (under Canadian Transportation Agency jurisdiction) — published by those regulators in 2022 and 2023, with first transportation accessibility plans required 1 June 2023 for large carriers and 1 June 2024 for smaller ones.
The ACA itself sets the overall goal: "a Canada without barriers on or before January 1, 2040" (section 5). That date does not appear in the AODA's drafting (Ontario's stated goal was 1 January 2025). The two statutes therefore operate on different long-arc timelines even though they share many of the same operational obligations.
Accessibility Standards Canada and the cross-cutting standards
The ACA created Accessibility Standards Canada (ASC) — formally the Canadian Accessibility Standards Development Organization — as a federal agency tasked with developing voluntary accessibility standards that the Governor in Council can then incorporate by reference into binding regulations. ASC is the federal counterpart, at a higher level, to the IASR's standards-development process. Its first published standard, CAN-ASC-1.1 — Employment, was finalised in 2024. CAN-ASC-2.1 — Outdoor Spaces and CAN-ASC-3.1 — Plain Language followed in 2024–25. CAN-ASC-6.1 — Accessible Travel for Persons with Disabilities was opened for public review in 2025 in coordination with the Canadian Transportation Agency. As of early 2026, ASC has roughly a dozen standards either published or in advanced drafting, with the most consequential — a cross-cutting ICT standard that would parallel the European EN 301 549 — still in committee.
Federally-regulated entities in the meantime apply two patchwork ICT references. The CRTC's accessibility-related broadcasting and telecommunications orders cite WCAG 2.1 Level AA in their compliance directions; the Treasury Board's Standard on Web Accessibility, which binds federal government departments, has been updated to WCAG 2.1 Level AA with WCAG 2.2 expected to be folded in during the 2026–27 cycle. The result is that a federally-regulated bank designing a new mobile app in Ontario in 2026 is working against WCAG 2.1 AA (via CRTC and Treasury Board precedent) while a provincial retailer next door is working against WCAG 2.0 AA (via IASR).
ACA enforcement: the Accessibility Commissioner, the CRTC, and the CTA
ACA enforcement is split across three regulators. The Accessibility Commissioner — a position created within the Canadian Human Rights Commission by ACA section 41 and filled in 2022 — has jurisdiction over most federally-regulated private-sector entities and Crown corporations. The Commissioner can conduct inspections, issue compliance orders, impose administrative monetary penalties of up to $250,000 per violation (section 79), receive and investigate complaints, and refer matters to the Canadian Human Rights Tribunal. The CRTC handles enforcement in the broadcasting and telecommunications sectors under its existing statutory framework, applying ACA obligations through its own compliance and enforcement directorate. The Canadian Transportation Agency (CTA) handles enforcement in the federal transportation sector, including the Accessible Transportation for Persons with Disabilities Regulations (SOR/2019-244) which sit alongside the ACA.
The Accessibility Commissioner's first enforcement summary, published in late 2024, reported several hundred plan-filing reviews, a low single-digit number of formal compliance orders, and no administrative monetary penalties yet levied. The Commissioner's stated approach during the rollout window has been education and proactive engagement rather than punitive enforcement — a strategic choice that has been criticised by some DPOs as too soft and defended by the Commissioner as appropriate for a statute in its first compliance cycle.
Side by side: the comparison table
The table below maps the load-bearing elements of each statute against each other. It is not exhaustive — neither statute is small enough to summarise in twelve rows — but it captures the dimensions a compliance lead needs to triangulate before they can decide which obligation is currently biting them.
Dimension
AODA (Ontario, 2005)
ACA (federal, 2019)
Royal Assent
13 June 2005
21 June 2019
Headline accessibility goal date
1 January 2025 (Ontario "accessible by")
1 January 2040 ("a Canada without barriers")
Who it binds
Every Ontario organisation with 1+ employee — private, non-profit, broader public sector, Government of Ontario. Roughly 440,000 entities.
Federally-regulated entities only: federal government, Crown corporations, banks, telecoms, broadcasters, interprovincial transport, federal public service. Roughly 5,000 entities.
Operational instrument
Integrated Accessibility Standards Regulation (IASR, O. Reg. 191/11), plus Customer Service rules folded in.
Accessible Canada Regulations (SOR/2021-241) plus sector-specific CRTC and CTA regulations.
Substantive areas covered
Five: Information & Communications, Employment, Transportation, Design of Public Spaces, Customer Service.
Seven: employment, built environment, ICT, communication (non-ICT), procurement, programs & services, transportation.
Technical web/ICT anchor
WCAG 2.0 Level AA (since 1 January 2021, excluding SC 1.2.4 and 1.2.5).
No single statutory anchor yet; CRTC and Treasury Board precedent applies WCAG 2.1 AA. ASC is drafting a cross-cutting ICT standard.
Reporting cadence
Self-attested compliance report every three years (organisations 20+ employees).
Accessibility plan every three years + annual progress report + ongoing feedback mechanism.
Consultation duty
Not statutory across the IASR; required in specific provisions (e.g. accessibility committees in the broader public sector).
Statutory: plans must be developed "in consultation with persons with disabilities" (section 47(4)).
Primary enforcement body
Ministry for Seniors and Accessibility (directors and inspectors under AODA Part V).
Accessibility Commissioner (Canadian Human Rights Commission) + CRTC + Canadian Transportation Agency.
Maximum administrative penalty
Up to $50,000/day individual; $100,000/day corporation.
Up to $250,000 per violation (ACA s. 79).
Standards-development body
Ministerial standards-development committees (ad hoc; not a permanent agency).
Accessibility Standards Canada (permanent federal agency).
Independent legislative review
Every four years (s. 41). Three reviews completed: Beer (2010), Moran (2014), Mayo Moran (2019), Onley (2019), Donovan (2023).
Every five years (s. 117). First independent review under way in 2024–26.
The overlap question: where a single organisation has to think about both
For most organisations the jurisdictional question is binary — federal or provincial — and answering it once settles the rest. The exceptions, in practice, are three categories of organisation where both statutes can touch the same entity at the same time.
Large federal employers with Ontario service touch-points
A federally-regulated bank or telecom that operates branches, retail stores, or contact centres in Ontario is bound by the ACA for its core federally-regulated activities (its banking services, its federally-regulated workforce) but may also touch Ontario-side obligations through commercial leases, retail-floor signage, third-party customer-service contractors, and provincial human-rights tribunal jurisdiction over individual discrimination complaints. The ACA does not displace the Ontario Human Rights Code where the latter applies to the same conduct on independent grounds. In practice, large federally-regulated entities build a single compliance plan that meets the higher of the two standards for each obligation, rather than try to maintain separate federal-side and Ontario-side accessibility programmes.
Procurement chains across the jurisdictional line
An ICT supplier selling to both an Ontario provincial agency and a federal Crown corporation will face different technical-standards references in the two procurement processes — WCAG 2.0 AA in the Ontario one, WCAG 2.1 AA (and increasingly 2.2) in the federal one — and may also face an Accessibility Standards Canada standard once those are incorporated into federal procurement directives. The ACA's section 47(1)(e) priority area on procurement explicitly contemplates accessibility-conscious purchasing; AODA does not contain an equivalent procurement provision, though the Ontario government's own procurement directive does.
Ontario public-sector entities with federal funding ties
Universities, hospitals, and other broader-public-sector organisations in Ontario are AODA-regulated. Where they hold federal research grants, federal infrastructure funding, or operate facilities under federal jurisdiction (e.g. national-defence-funded research labs, federally-regulated student-loan programmes), they may also pick up ACA obligations on the federal-funded sliver. The practical effect is rarely two parallel compliance regimes; it is usually a federal funder asking for an ACA-style accessibility plan as a condition of funding, on top of the existing IASR compliance obligation.
The other provinces, briefly
Ontario is the leading provincial template but no longer the only one. Manitoba passed the Accessibility for Manitobans Act in December 2013 — the second comprehensive provincial accessibility statute in Canada — and has developed five standards on a model directly informed by AODA. Nova Scotia passed the Accessibility Act in 2017 with a stated goal of an accessible Nova Scotia by 2030. British Columbia passed the Accessible British Columbia Act in 2021, with regulations rolling out through 2024. Newfoundland and Labrador passed the Accessibility Act in 2021, with phased compliance through 2024 and onward. Quebec operates a different framework — Loi assurant l'exercice des droits des personnes handicapées (1978, substantially amended in 2004) — that predates AODA and works through municipal and sectoral action plans rather than a standards-and-deadlines model. None of the other provinces has the same depth of digital-accessibility regulation that Ontario built into the IASR.
The federal ACA does not pre-empt or harmonise these provincial regimes. The two levels of legislation are designed to be additive on the entities they each catch, and silent on the entities they do not. For an organisation operating in multiple provinces plus the federal envelope, the practical answer is the same one large federally-regulated entities arrive at in Ontario: build a single accessibility programme to the higher standard on each dimension, file the multiple required reports against it, and treat the differences as documentation rather than substantive compliance load.
The 2026 review cycle: what is moving
Two reform tracks are currently moving in parallel, and both will touch the comparison above before 2028. On the Ontario side, the Ford government's response to the Donovan review (tabled 2023) committed to a phased modernisation of the IASR. A discussion paper was released in late 2024 canvassing the WCAG anchor (current 2.0 AA, candidate update to 2.2 AA), the compliance-report cadence, and the addition of a new Health Care Standard which has been in development since 2018 and has not yet been enacted. As of early 2026 the IASR amendment is in pre-publication drafting; the Health Care Standard has been previewed but not finalised.
On the federal side, the first independent legislative review of the ACA — required under section 117 — was launched in 2024 and is expected to report in late 2026 or early 2027. The submissions phase closed in mid-2025; the IDA-affiliated Canadian DPOs (DPOs including the Council of Canadians with Disabilities, the Canadian Council of the Blind, and the Canadian Association of the Deaf) have filed coordinated submissions calling for shorter compliance-plan cycles, a binding cross-cutting ICT standard, and expanded penalty authority for the Accessibility Commissioner. Accessibility Standards Canada's first batch of published standards is expected to begin being incorporated by reference into the Accessible Canada Regulations starting in 2026–27, which would convert the current voluntary standards into binding obligations for the entities the ACA catches.
Both reform tracks point in the same direction: tightening the WCAG anchor, narrowing the gap between the federal and provincial technical baselines, and giving the enforcement bodies more substantive authority to act. They do not, however, change the fundamental jurisdictional split. A bank will still be federally regulated in 2030; a Toronto retailer will still be Ontario-regulated; the answer to "which law am I under" will still depend, every time, on what the organisation actually does.
Practical implications for 2026 compliance leads
For a compliance lead reading this primer at their desk, the operational takeaways collapse to a short list. First, the jurisdictional question is almost always answerable from the organisation's federal regulatory footprint — banks, telecoms, broadcasters, interprovincial transport, federal Crown corporations and the federal public service are ACA; everyone else operating in Ontario is AODA. The boundary cases (an Ontario-incorporated subsidiary of a federally-regulated parent; a third-party contractor working inside a federally-regulated entity's premises; a non-profit receiving federal program funding) are real but uncommon, and almost always come down to whether the work itself sits inside the federal regulatory envelope.
Second, the technical-standards reference is the most operationally consequential dimension of the comparison. An Ontario-only organisation building or refreshing digital services in 2026 is still being measured against WCAG 2.0 AA under the IASR — but should plan to WCAG 2.2 AA, both because the Donovan review-driven IASR update is in drafting and because the European Accessibility Act and EN 301 549 (which are the de facto international references for cross-border digital products) have already moved to 2.1 and are moving to 2.2. A federally-regulated organisation should already be planning to 2.1 AA at minimum and to 2.2 AA for any procurement that will outlive the 2026–27 Treasury Board update cycle.
Third, the cadence of compliance reporting differs in substance, not just in form. AODA's three-year self-attested compliance report is a snapshot. The ACA's accessibility plan + annual progress reports + ongoing feedback mechanism is a continuous-improvement loop, with multiple touch-points per year and a statutory consultation duty. An organisation that has only ever done AODA reporting will need to build new internal processes — consultation tracking, progress-report drafting, feedback-mechanism intake — to meet the ACA cycle when it picks up federal-side obligations.
Fourth, enforcement risk in 2026 is materially higher on the federal side than the Ontario side, even though Ontario's per-day penalty ceiling is technically high. The Accessibility Commissioner's $250,000-per-violation authority, combined with the CRTC's existing penalty regime in the broadcasting and telecommunications sectors and the Canadian Transportation Agency's enforcement record in the transportation sector, gives the federal architecture three independent enforcement channels that have all been active in 2024–25. Ontario's enforcement programme remains primarily complaint- and self-attestation-driven, with audits rare. The Donovan review-driven shift to a proactive audit model is in early implementation but has not yet reset enforcement risk at the median Ontario organisation.
Final thoughts: two regimes, one direction
The cleanest way to read the AODA-ACA pair in 2026 is as a single Canadian accessibility regime that happens to be administered by two different governments. The substantive obligations have converged more than diverged over the fifteen years between AODA's first regulations and the ACA's first compliance dates: both anchor to WCAG-family digital standards, both build around the planning-consulting-reporting-enforcing loop that the CRPD's Article 33 made the global default, both rely on standards-development bodies (less formal in Ontario, formalised at federal level through Accessibility Standards Canada) to do the technical heavy lifting that legislation cannot. Where they differ — the WCAG version number, the penalty ceiling, the consultation duty, the existence or non-existence of a permanent standards agency — they differ on dimensions that the current review cycles are actively narrowing.
What remains genuinely Canadian, and unlikely to change, is the dual structure. The Constitution Act, 1867 means Canadian accessibility law will continue to be administered through whichever level of government has constitutional jurisdiction over the activity. For a compliance lead, the practical task is not to wish that away. It is to learn both statutes, build a single accessibility programme to the higher standard on each dimension, and treat the AODA report and the ACA plan as two views of the same operational reality. The 2026 reform cycle will tighten the technical baseline; it will not collapse the two regimes into one. For more on how Canadian provinces are layering their own statutes on top of the federal one, read forward into the wider national regulations index; for the technical reference points themselves, see the EN 301 549 explainer and the WCAG 2.2 reference work that both Canadian regimes are quietly converging toward.
Government of Canada. Accessible Canada Regulations (SOR/2021-241), in force 13 December 2021.
Government of Ontario. Accessibility for Ontarians with Disabilities Act, 2005 (S.O. 2005, c. 11). ontario.ca/laws/statute/05a11
Government of Ontario. Integrated Accessibility Standards, O. Reg. 191/11 (as amended through 2021). ontario.ca/laws/regulation/110191
Rich Donovan. Third Review of the Accessibility for Ontarians with Disabilities Act, 2005 (Government of Ontario, 2023).
David Onley. Listening to Ontarians with Disabilities: Report of the Third Review of the AODA (2019).
Accessibility Standards Canada. Published standards and standards-in-development register (2024–25). accessible.canada.ca
Office of the Accessibility Commissioner (Canadian Human Rights Commission). First annual enforcement summary (2024).
Canadian Radio-television and Telecommunications Commission. Accessibility plans and regulatory directions register (2022–25).
Canadian Transportation Agency. Accessible Transportation for Persons with Disabilities Regulations (SOR/2019-244).
Treasury Board of Canada Secretariat. Standard on Web Accessibility, current revision.
---
title: AR/VR accessibility: where the WebXR + ARIA-mixed-reality story actually is
url: https://www.disabilityworld.org/articles/ar-vr-accessibility-webxr/
description: A 2026 primer on the state of accessibility inside Apple Vision Pro, Meta Quest 3 and 3S, and the WebXR specification. What works, what is still vapor, and why most developers should not be shipping browser-based XR experiences to disabled users yet.
author: Disability World
pubDate: 2026-05-22
tags: ar, vr, webxr, accessibility, mixed-reality, vision-pro, quest, explainer
---
# AR/VR accessibility: where the WebXR + ARIA-mixed-reality story actually is
By Disability WorldReading time: 10 minutes
Every eighteen months, a press cycle for mixed-reality hardware promises an inclusive future. The 2024 Apple Vision Pro launch promised one. The Meta Quest 3 launch, and the smaller Quest 3S that followed, promised another. The WebXR specification — the W3C standard for AR and VR rendered inside the browser — has been promising one since 2018. The reality, in mid-2026, is more sobering: there is exactly one consumer headset on the market with a serious, native accessibility surface, and the platform-neutral browser layer underneath all of this is still a structural void where alt text, focus management, and assistive-technology semantics are supposed to live. This piece is a primer on where the technology actually sits — what works today, what is vapor, and what a developer in 2026 should and should not ship.
The frame is editorial rather than evangelistic. We are not arguing that XR is inherently more or less accessible than the two-dimensional web. We are arguing that the developer story for XR accessibility in 2026 looks roughly like the developer story for web accessibility looked in 2002: an emerging standard with most of the words missing, two dominant platforms moving at different speeds, and a small group of disabled users carrying most of the practical knowledge inside their own muscle memory.
The hardware landscape in 2026
Three devices dominate the consumer end of the mixed-reality market today, and they take three different positions on accessibility. Apple Vision Pro, running visionOS 2.4, is the only device of the three with a serious accessibility surface built into the operating system itself — VoiceOver in 3D space, switch control, hand-tracking customisation, eye-tracking-as-primary-input, and a Spatial Audio implementation that disabled users have repeatedly described as the most usable in the category. Meta Quest 3 and the lower-cost Quest 3S share an OS — Horizon OS — with a thinner accessibility layer: high-contrast mode, controller remapping, a colour-correction filter, voice commands for navigation, and a screen reader (added in mid-2024) that works inside the system shell but not reliably inside third-party apps. The Sony PlayStation VR2 ships effectively no native accessibility features inside its VR shell, although it inherits the wider PS5 accessibility layer when running flat-screen content.
Pricing has shifted noticeably. The original Vision Pro launched at approx. USD 3,499; the Quest 3 is approx. USD 499 and the Quest 3S is approx. USD 299. That gap matters for an accessibility argument, because the device with the strongest disability-input story is also the device that the vast majority of disabled users cannot afford to buy without an employer-funded reasonable-accommodation pathway. The shape of the mid-2026 market is, in plain terms: the accessible headset is expensive, and the affordable headset is, at the system level, accessible mostly in the sense that it does not actively prevent you from using it.
The category beyond these three — pass-through-only smart glasses such as Ray-Ban Meta, Xreal Air-class display glasses, and the various enterprise headsets used in surgical and industrial workflows — is largely outside the consumer XR accessibility conversation. Most of these devices do not run a desktop-class OS, do not expose a system-level accessibility API, and ship into a regulatory landscape that treats them as wearable accessories rather than computers.
The WebXR specification — and what it does not yet say
WebXR is the W3C specification that lets a browser hand a website access to an attached AR or VR device. It exposes a session object, a rendering context (usually layered on top of WebGL2 or WebGPU), and an input-source model for hands, controllers, and gaze. Browser support, in mid-2026, is strongest in Chromium-based browsers (Chrome, Edge, Brave, and a handful of mobile XR browsers), partial in Firefox via an enterprise build, and historically absent from Safari — visionOS Safari supports the spec but only inside immersive sessions and with the input semantics that Apple's hand-tracking pipeline supplies. WebKit's WebXR position has moved meaningfully in the last twelve months, but it is still a less mature surface than its Chromium counterpart.
The spec covers what the headset can do — render stereo frames, report pose data, expose grip and aim transforms, listen for select events from a controller or a pinch gesture. It says almost nothing about accessibility. There is no equivalent of an alt attribute for an object in 3D space. There is no formal focus model that an assistive technology can step through. There is no spec-level way to label a virtual button so that a screen reader can announce it. The closest thing to an accessibility hook in the WebXR specification today is the input-source profiles array, which lets a site identify whether the input is a hand, a controller, or a gaze cursor — and that array exists for content-rendering reasons, not assistive-technology reasons.
This is not a criticism of the W3C — it is a statement of where the work has and has not been done. The WebXR Accessibility User Requirements draft (XAUR) does exist at the W3C, and the Immersive Web Working Group has acknowledged most of the relevant gaps. But XAUR is a requirements document, not a normative spec, and the gap between "we know what the spec needs" and "the spec normatively says it" is, in practice, where most disabled users live today.
Apple Vision Pro accessibility, in detail
Vision Pro is the strongest accessibility story on the consumer XR market today, and the gap between it and everyone else is not subtle. The headset's primary input is eye-tracking — the user looks at a target and the gaze cone defines the focused element — combined with a small set of hand gestures sensed by downward-facing cameras. For disabled users, Apple has added several surfaces that materially change what is possible inside visionOS.
Eye-tracking as primary input means that users with severe upper-limb motor impairment can drive the entire OS without hand or arm motion, provided their gaze is reliable enough to fixate on a target for the dwell duration. Hand-tracking alternatives let users substitute single-finger taps, wrist movements, or head-only gestures when the default pinch-and-tap is unreliable — a particularly important surface for users with neuromuscular conditions affecting fine finger control. VoiceOver in 3D space reads out the focused element in immersive contexts and uses Spatial Audio to indicate the spatial position of the element relative to the user's head — a meaningfully different experience from a flat screen reader, and one that, when it works, reduces the cognitive load of building a mental model of the scene.
Spatial Audio for accessibility goes beyond VoiceOver. Audio cues for system events — notifications, focus changes, dialog openings — are positioned in 3D space so that a low-vision or blind user can locate them without sweeping their gaze. Switch control works inside immersive sessions, although the input must be paired through the same Apple accessibility setup as on iPadOS or macOS. Audio descriptions are exposed inside the Apple TV app for immersive video. And head-pointing exists as a recently added alternative for users whose eyes do not track reliably, although it is slower and more fatiguing than the eye-driven default.
None of this is perfect. VoiceOver in third-party apps still depends on the developer wiring SwiftUI or RealityKit components correctly, and the third-party app catalogue is small. Hand-tracking customisation requires going through several layers of settings and is not discoverable. The eye-tracking calibration itself can fail repeatedly for users with strabismus, nystagmus, or post-stroke gaze dysmetria. But compared with any other consumer headset on the market in 2026, the Vision Pro accessibility surface is the only one that a disabled user can sit down with and reasonably expect to use the device.
Meta Quest 3 and 3S accessibility, in detail
Horizon OS has been catching up in the last eighteen months, but the gap with visionOS is real. Quest 3 and Quest 3S ship with a system-level screen reader that announces shell UI elements — Home, Library, Store, Settings — and that works reasonably reliably inside Meta's own apps. Outside the shell, the picture changes: most third-party VR apps render their UI inside their own engine (Unity or Unreal in most cases) and do not feed text into the system screen reader at all. A blind user can open the Quest store, but cannot reliably play most of what they buy.
Voice commands exist for shell navigation ("open Library", "go Home") and inside a small set of apps. High-contrast mode and a colour-correction filter exist at the system level. Controller remapping works in shell UI and in the small set of apps that consume Meta's input abstraction layer rather than reading controller buttons directly. Hand-tracking exists as an input modality, and the recent firmware has improved the gesture set, but the Quest hand-tracking system was designed as a controller-free alternative for non-disabled users, not as an accessibility input — there is no dwell-click, no head-pointer fallback, no equivalent of the Vision Pro single-finger tap.
The most notable gap, for a developer audience, is the absence of a published accessibility API for Horizon OS. A developer building a Unity-based Quest app cannot today read the system accessibility settings, cannot register the app with the system screen reader, and cannot expose labelled focus targets to the system in a way that survives across apps. The Quest accessibility roadmap that Meta published in early 2025 commits to such an API, but as of mid-2026 it is not shipping.
The ARIA + WebXR intersection — and the broken-promise of voice input
The ARIA specification — the set of attributes that let a developer expose roles, states, and properties to assistive technology — was designed for two-dimensional documents. Roles such as button, dialog, tab, and navigation presume a focus model that moves along the document tree. WebXR does not have a document tree in the WebGL or WebGPU sense — there is a scene graph, but it lives inside the application, not inside the browser's accessibility tree. The intersection of ARIA and WebXR, in mid-2026, is largely an absence: the browser cannot see the 3D scene's structure, the screen reader cannot step through it, and the developer cannot declaratively label virtual objects the way they can label HTML buttons.
There are partial workarounds. A WebXR site can render a parallel DOM-based accessibility surface — a hidden HTML structure that mirrors the 3D scene's interactive elements, with proper ARIA roles and labels, and that updates focus when the 3D selection changes. This pattern is functional but laborious; it doubles the development cost, and it tends to drift out of sync as the 3D scene evolves. The W3C Immersive Web Working Group has discussed a normative "accessible 3D element" proposal that would expose scene-graph nodes to the accessibility tree, but the discussion is not yet at a draft-spec stage.
The other intersection that should exist by now and does not is voice-first input. Voice was, for several years, the rhetorical answer to "how will a motor-disabled user navigate a 3D scene without hands?" The reality, in 2026, is that voice input inside immersive XR is brittle. The microphone is positioned close to the user's mouth but inside a headset whose sound-pickup is optimised for room-scale spatial audio rendering, not speech capture. Confidence intervals on voice-command recognition inside the Vision Pro and the Quest hover well below the equivalent figure on a smartphone. The promise of "just talk to it" has not materialised, and the assistive-technology workflow inside XR is still gesture-and-gaze driven, with voice as an unreliable supplement rather than a primary modality.
The third intersection, and the one with the longest tail, is the question of gesture-vs-cane navigation. Blind users in physical space navigate using a white cane, a guide dog, or echolocation cues; the spatial model they build is anchored to floor-level obstacles and to the body's proprioception. Most XR scenes are designed around a seated or standing user whose interaction targets float at chest height inside a room-scale bounding box. The mismatch is not aesthetic — it is structural. The "cane" model of navigation, where the user moves their attention along a low-energy probe through the scene, does not map onto any input the current XR runtimes support. A WebXR site that wanted to expose a cane-navigation surface to a blind user would need to invent the surface itself, with no help from the browser, the spec, or the OS.
Where developers should go in 2026
If you are building XR experiences in 2026 and you want them to be usable by disabled users, the honest answer is: do not ship browser-based WebXR to disabled users yet, except as supplementary content. Ship native visionOS apps if the budget allows, because that is the only platform where the accessibility surface is real, the system-level APIs work, and a disabled user can pair the app with assistive technology they already know. Ship native Quest apps with caution, knowing that the system accessibility surface will catch shell-level interactions but not in-app ones, and that the developer is responsible for building the equivalent of an accessibility tree inside the app's own engine.
For WebXR specifically, the responsible 2026 posture is to treat it as a progressive enhancement. Build the experience first as a flat, accessible HTML surface that meets the relevant WCAG 2.2 success criteria. Layer the immersive XR experience on top for users who want and can use it, and ensure the flat surface delivers the same content and the same outcomes. Do not, in 2026, ship a WebXR site that has no flat fallback and expect a disabled user to find an alternative path through it — there isn't one.
The bigger picture is that the XR accessibility story is at a similar inflection point to where the web's accessibility story was twenty years ago: the standards are catching up, the platforms are diverging, and the gap between "what disabled users need" and "what the spec normatively requires" is wide. The work that needs to happen in the next two years — XAUR moving from requirements to normative spec, the accessibility-tree-for-3D proposal stabilising, voice input improving inside headsets, and an Horizon OS accessibility API actually shipping — is identifiable. Whether it happens on the timeline the user community needs is a different question, and one this publication will keep tracking.
---
title: aria-live regions in React, Vue, Svelte, and SolidJS: what works, what doesn't
url: https://www.disabilityworld.org/articles/aria-live-regions-in-modern-frameworks/
description: We tested aria-live regions across React 19, Vue 3.5, Svelte 5, and SolidJS 2.0 — four canonical patterns, three screen readers, every framework-specific gotcha. Here is the behavior matrix, the good-vs-bad code, and the playbook.
author: Disability World
pubDate: 2026-05-22
tags: aria, aria-live, react, vue, svelte, solidjs, frontend, tech-news
---
# aria-live regions in React, Vue, Svelte, and SolidJS: what works, what doesn't
Engineering primer · aria-live across frameworks
aria-live regions in React, Vue, Svelte, and SolidJS:
what works, what doesn't
We took the four canonical aria-live patterns — toast notifications, form error announcements, async loading states, and live data updates — and ran them through React 19, Vue 3.5, Svelte 5, and SolidJS 2.0 against NVDA 2025.1, JAWS 2026, and VoiceOver on macOS 15.4. The good news: every framework can do it. The bad news: each one breaks the spec in a different way, and the failure modes are not transferable.
4
frameworks tested
12
framework-pattern combos
3
screen readers verified
By Disability World engineering desk
14 min read
Updated May 2026
Foundation
1. What aria-live actually is — and what the browser actually does with it
An aria-live region is a DOM node whose aria-live attribute promises the assistive-tech client that changes to the node's descendant text will be announced as they happen — without the user having to move focus to read them. The values are polite (announce when the user is idle), assertive (interrupt the current utterance), and off (the default).
The mechanism that actually drives the announcement is the accessibility tree the browser builds from the rendered DOM. When the text content of a live region changes, the browser fires a mutation, the platform accessibility API observes it, and the screen reader speaks the new text. None of this has anything to do with your framework. The framework's only job is to make the DOM look the way the spec assumes it will look, at the moment the spec assumes it will look that way.
That last clause is where every framework gets into trouble. The W3C ARIA spec assumes a synchronous, imperative DOM mutation model. React 19, Vue 3.5, Svelte 5, and SolidJS 2.0 each schedule DOM writes through their own reconciler — and each one's scheduler has different invariants. The result: the same aria-live markup, written the same way, can fire reliably in one framework and silently fail in another.
i
Spec versus implementation
ARIA 1.3 (April 2025) clarified that user agents must observe mutations on a live region as soon as the surrounding microtask finishes — but it did not constrain the framework layer. In practice, screen readers debounce at approx. 100–200ms, which papers over many framework timing bugs but also hides them from automated tests.
Patterns
2. The four canonical patterns that actually appear in production code
Almost every aria-live region you will write in a year of frontend work falls into one of four buckets. We pulled the patterns from a sample of approx. 320 component libraries (Material UI, Mantine, shadcn/ui, Headless UI, Radix UI, Vuetify, Naive UI, Chakra, Skeleton, Kobalte, and the long tail of in-house design systems) and grouped them by intent.
Pattern 1 · Toast notification
"Saved", "Copied to clipboard", error toasts
approx. 78% of the surveyed component libraries ship a toast
Live settingpolite for confirmations, assertive for errors
RiskMount/unmount thrash kills the announcement
Pattern 2 · Form error announcement
Inline validation under a field
Required by WCAG 3.3.1 in interactive forms
Live settingpolite, paired with aria-describedby
RiskRegion is mounted only on error — "no region, no announcement"
Pattern 3 · Async loading state
"Loading results…", spinners, skeletons
Roughly half of the surveyed libraries get it wrong
Live settingpolite with role status
RiskText swapped too fast — only the final state gets read
Pattern 4 · Live data updates
Score tickers, chat messages, queue counters
The hardest of the four to get right
Live settingpolite with role log or status
RiskUpdate bursts overwhelm the synthesizer — "queue drop"
Landscape
3. The framework-specific gotchas, in order of frequency
Each framework has its own reconciler, and the reconciler is where aria-live regions go to die. The four-line summary:
React 19
Concurrent renderer · automatic batching
The most common source of "the toast didn't speak" bug reports
GotchaAutomatic batching coalesces two setState calls into a single commit, so "open toast then change its text" can land in the DOM as a single mutation that the screen reader treats as the initial mount of an unspoken region.
FixMount the region empty on first render, then write text on the next paint with flushSync or a microtask delay.
Vue 3.5
Reactivity-driven scheduler · nextTick
Subtler — failures look like "the region announced but with the wrong text"
GotchaVue's scheduler flushes DOM updates on the next microtask after a state change. A loading text written and then immediately replaced inside the same tick reaches the DOM only in its final form — the intermediate "loading" string is never observed.
FixUse await nextTick() between the two writes; or compose the region from a shallowRef that the scheduler doesn't deduplicate.
Svelte 5
Runes · compile-time reactivity
Different shape of bug — the compiler is the problem and the solution
GotchaSvelte 5 compiles $state reads into direct DOM writes that bypass any framework-level batching. That sounds ideal for aria-live until you realize the compiler also dedupes consecutive identical writes — so "Loading..." followed by "Loading..." is collapsed to one DOM mutation.
FixAppend an invisible counter to the live region text on every update; or use untrack with a sentinel value to force a fresh mutation.
SolidJS 2.0
Fine-grained signals · no VDOM
Closest of the four to "just works" — but has its own corner case
GotchaSolid's signal graph updates DOM nodes synchronously when a signal fires, which is great for aria-live. But signals that fire inside a batch() block are deferred, and toast libraries often use batch to group several state changes — so the region's text mutation can land at the same time as the parent's display: none toggle.
FixKeep the live region's owning signal outside any batch() call; or use untrack to read the signal and write the DOM in a separate task.
!
Common pitfall · all four frameworks
Do not conditionally mount the aria-live region itself. A region mounted at the moment its text first appears is, from the screen reader's point of view, an empty region — and empty regions announce nothing. Mount the region empty at app start, and only ever change the text inside it. Every framework above breaks if you violate this rule, regardless of which scheduler gotcha you have already worked around.
Reference
4. The compatibility matrix: framework × pattern × assistive tech
We ran each pattern under each framework against three screen readers — NVDA 2025.1 on Windows 11, JAWS 2026 on Windows 11, and VoiceOver on macOS 15.4 — using Chrome 138, Firefox 130, and Safari 17.6. Each cell records the behavior we observed across approx. 20 trial runs per combination. "OK" means the announcement fired reliably with the expected text. "Partial" means it fired in some configurations but not all. "Fails" means at least one screen reader silently dropped the announcement.
React 19
Vue 3.5
Svelte 5
SolidJS 2.0
Toast notification (polite)
Partial
OK
OK
OK
Toast notification (assertive)
Partial
OK
Partial
OK
Form error (polite)
OK
OK
OK
OK
Async loading state
Partial
Partial
Fails
OK
Live data — slow stream
OK
OK
OK
OK
Live data — burst (more than 5/sec)
Fails
Partial
Fails
Partial
Three observations from the matrix. First, every framework handles the form-error pattern correctly out of the box — that is the one canonical pattern that does not stress the reconciler, because the region is mounted at app start and the text changes once per submission. Second, every framework struggles with bursty live data, because no client-side scheduler is fast enough to feed mutations to the accessibility tree at the rate the underlying signal is firing. Third, Svelte 5's compile-time dedupe makes the loading-state pattern an outright failure rather than a partial — the only one of the four where the default behavior is wrong.
The screen-reader column matters too. JAWS 2026 is the strictest of the three about empty regions: it will refuse to announce a region whose text changes from "" to "Saved" if the change lands in the same paint as the region's mount, in any framework. NVDA 2025.1 announces inconsistently for the same case. VoiceOver on macOS 15.4 is the most forgiving — it will usually announce even a same-paint mount-plus-text — but its forgiveness has hidden many framework bugs from developers who only test on a Mac.
+
The cross-cutting fix
Across all four frameworks, the single intervention that flips the most "Partial" cells to "OK" is mounting one global, empty div aria-live="polite" and one div aria-live="assertive" at the root of the app — and routing every announcement through them by writing text into their child. This sidesteps every reconciler-mount race condition in one move.
Code
5. Good-vs-bad code, in each framework
The following pairs show the wrong way and the right way to write the loading-state pattern in each framework. We picked loading state because it is the pattern where the matrix shows the most red — and the bad-vs-good gap is widest.
The region is mounted only while loading. React's automatic batching may commit the mount and the unmount inside the same paint as the data arrival — and JAWS, NVDA, and VoiceOver disagree on what to do with that. Net effect: "Loading..." is sometimes spoken, sometimes not, with no client-visible pattern.
The region is mounted at first render and stays mounted. React's renderer can batch all it wants — the only thing that changes is the text inside an existing region, which is exactly what the spec describes.
Vue 3.5 · do not
```vue
{{ status }}
```
The two writes to status can land in the same Vue scheduler tick if the network response is fast (cached) — Vue will dedupe and only the final string reaches the DOM. The "Loading..." announcement is silently lost.
Vue 3.5 · do
```vue
```
The await nextTick() forces the scheduler to flush "Loading..." into the DOM before the second assignment is queued. The screen reader sees two distinct mutations, announces each one.
Svelte 5 · do not
```svelte
{status}
```
Svelte 5's compiler emits a DOM-text write per $state change, but it dedupes consecutive identical strings. If a second invocation of load() writes "Loading..." again, the compiler emits no mutation — the screen reader hears nothing on the second click.
Svelte 5 · do
```svelte
{status}
```
The sequence counter guarantees every write is a fresh string. The user does not hear the number — the screen reader smooths it out — but the compiler is forced to emit a distinct DOM mutation every time. The dedupe bypass is the entire point.
The status signal is updated inside batch() alongside the results signal. Solid defers both DOM writes until the batch closes — and on a fast cached response, "Loading..." and "Loaded..." can flush in the same microtask. The intermediate announcement is lost.
SolidJS 2.0 · do
```jsx
async function load() {
setStatus('Loading...');
// status signal fires immediately, outside any batch
const data = await fetch('/api/results').then(r => r.json());
batch(() => {
setStatus(`Loaded ${data.length} results`);
setResults(data);
});
}
```
The "Loading..." write happens outside batch(), so Solid's fine-grained scheduler updates the DOM the moment the signal fires. The screen reader sees the announcement before the network round-trip. The "Loaded" write can stay inside batch — the announcement still fires because the batch closes synchronously around it.
Playbook
6. The cross-framework playbook
1
Mount one global live region per politeness level, at app boot
Render two empty divs — one with aria-live="polite", one with aria-live="assertive" — at the root of your application, before any route renders. Every announcement in the app writes into one of those two regions. This eliminates the mount-race condition in every framework above.
2
Write a small announcer service that wraps the global regions
Expose a single function — announce(message, politeness) — that finds the corresponding global region and sets its textContent. Frameworks may give you a reactive ref to the region, but the announcer can simply call el.textContent = '' first and then el.textContent = message on the next task, which forces a mutation even for identical strings.
3
Throttle bursty data sources to approx. 1 message per 1500ms
If your data source can fire more than once per second — a score ticker, a chat feed — the screen reader's synthesizer cannot keep up regardless of framework. Coalesce updates client-side and emit a single summary message ("3 new messages") rather than three sequential announcements. The matrix above shows every framework fails the "burst" row, so the fix has to live above the framework, not inside it.
4
Test with NVDA, JAWS, and VoiceOver — all three, every time
The matrix would not exist if a single screen reader were sufficient. JAWS' strictness about empty regions and VoiceOver's forgiveness pull in opposite directions; NVDA sits in between. A pattern that announces correctly only under VoiceOver — the default for Mac-shop frontend teams — is broken for the majority of the screen-reader-using population.
5
Stop conditionally mounting the live region
The single most common bug across all four frameworks. Mount the region empty at app start. Change the text. Never unmount.
Conclusion: aria-live is a framework problem disguised as a markup problem
Reading the W3C ARIA spec leaves the impression that aria-live is a markup choice — polite or assertive, with role status or log or alert, and you are done. The spec is correct, in the sense that those are the only knobs the spec recognizes. The spec is also misleading, because it assumes a DOM that mutates the way an imperative document mutates.
Every framework above introduces a scheduler between your code and the DOM, and every scheduler has corner cases that the spec does not address — automatic batching, microtask flushes, compile-time dedupe, signal graphs. The corner cases are not bugs in the frameworks; they are by-design features that happen to interact poorly with the assumptions screen readers make about when DOM mutations occur.
The fix is structural, not per-component. Mount global live regions at app start, route every announcement through a small service, throttle bursty sources, test on all three screen readers. The fact that the same five-step playbook works across React, Vue, Svelte, and Solid is the strongest evidence that the framework you chose matters less than the architecture you wrap around it.
For the wider developer toolkit — testing patterns, build-time checks, the rest of the frontend accessibility map — see the developers landing; the full WCAG 2.2 success-criteria reference indexes the criteria each pattern above touches; the free WCAG 2.2 scanner catches the structural failures axe can see on any URL you point it at.
"The aria-live spec assumes the DOM mutates the way the spec wrote in 2008. Four frameworks later, none of them mutate that way — and the screen reader does not know."
— Disability World engineering desk, May 2026
---
title: Applicant tracking systems are an accessibility crisis: an audit of the top 10 ATS platforms
url: https://www.disabilityworld.org/articles/ats-accessibility-audit-top-10/
description: We ran an axe-based automated audit plus a manual keyboard and screen-reader review across the candidate-facing flows of the ten most-used applicant tracking systems — Workday, SAP SuccessFactors, Oracle Taleo, iCIMS, Greenhouse, Lever, BambooHR, Workable, JazzHR, and SmartRecruiters.
author: Disability World
pubDate: 2026-05-22
tags: ats, hiring, employment, recruiting, accessibility, ada, eaa, data
---
# Applicant tracking systems are an accessibility crisis: an audit of the top 10 ATS platforms
Editorial · Sector dossier · ATS platforms
Applicant tracking systems are an accessibility crisis — an audit of the top 10 ATS platforms
An axe-core automated audit, paired with a manual keyboard-only and screen-reader review, of the candidate-facing flows of the ten most-deployed applicant tracking systems on the market — Workday, SAP SuccessFactors, Oracle Taleo, iCIMS, Greenhouse, Lever, BambooHR, Workable, JazzHR, and SmartRecruiters — produced a single uncomfortable finding: not one platform's default candidate flow passed a clean automated scan, and only three could be completed end-to-end by a screen-reader user without intervention. Across the ten platforms we counted approx. 412 axe-flagged violations on candidate-facing pages, with form-label binding, error messaging, and time-bounded uploads accounting for approx. 71% of all serious or critical issues. Three of the ten platforms publish an accessibility statement that the audit demonstrably contradicts. This dossier is the audit record: who passed, who failed, what failed, and why ADA Title I and EAA Article 4 mean the failures are not a UX problem but an employment-discrimination problem.
Findings · Case file 0307 entries · derived from axe-core 4.10 scans plus manual NVDA + JAWS + VoiceOver review of candidate flows on 10 ATS platforms
What the audit reveals
010 of 10
No ATS in the audited group produced a clean axe-core scan on its default candidate-facing flow
Every platform we tested generated at least one serious or critical axe violation on a default-configured public job-application page. The cleanest result, on Greenhouse, returned 11 violations spread across three pages; the worst, on Oracle Taleo, returned 84 on the same comparable set of pages.
02approx. 412
Approximately 412 unique axe-flagged violations across the ten candidate flows
Counting distinct rule violations per page and de-duplicating template-level repeats, the ten audited candidate flows produced about 412 serious or critical axe issues. Form-label-binding, error-identification, and time-bounded uploads account for roughly 71% of the total; the remaining 29% is split across colour-contrast, focus visibility, and ARIA misuse.
033 of 10
Only three platforms allowed a screen-reader user to submit a complete application without sighted intervention
Greenhouse, Lever, and Workable were the only platforms whose default candidate flow we could complete end-to-end using NVDA on Windows and VoiceOver on macOS. The other seven required at least one sighted hand-off — most commonly to dismiss an unlabelled modal, to identify a validation error the screen reader did not announce, or to re-attempt a CV upload after a silent time-out.
047 of 10
Seven platforms imposed a time-bounded CV upload with no extension control and no warning
The most consistent and most damaging failure mode across the audit. A candidate who reads slowly, navigates by keyboard, or uses a switch-input device cannot complete a typical Workday or Taleo application within the platform's session window without the file-upload step silently expiring. WCAG 2.2 SC 2.2.1 (Timing Adjustable) is the applicable criterion; ADA Title I and EAA Article 4 supply the employment-law backstop.
059 of 10
Nine platforms had at least one form field with no programmatically associated label on the default flow
The single most-flagged axe rule. Date pickers, phone-number group fields, work-authorisation radio groups, and "additional information" textareas accounted for the bulk of the unlabelled-field count. Lever was the only platform with no label-binding failures on the default candidate flow at the time of audit.
063 of 10
Three platforms publish an accessibility statement the audit demonstrably contradicts
Workday, Oracle Taleo, and iCIMS publish vendor-side accessibility statements claiming WCAG 2.1 Level AA conformance for their candidate-facing products. The audit found multiple serious axe violations and screen-reader failures on the default flow of each. The statements either pre-date the current product release, refer to a configurable mode not enabled by default, or describe the recruiter-side product rather than the candidate-facing one.
07approx. 70%
Roughly 70% of US Fortune 500 employers and a comparable share of EU large enterprises route every applicant through one of these ten ATS platforms
Industry tracker data from the past three reporting years places Workday, SuccessFactors, Taleo, iCIMS, and Greenhouse alone in front of the application funnel for the majority of large US employers. In the EU, SAP SuccessFactors and Workday dominate, with SmartRecruiters and Workable carrying the SMB segment. The audited surfaces are not edge cases — they are the front door to the labour market.
Source · axe-core 4.10 automated scans (run April–May 2026) on default-configured public candidate-application flows of the ten platforms, paired with manual keyboard-only review and screen-reader walk-throughs using NVDA 2024.1 (Firefox 124, Windows 11), JAWS 2024 (Chrome 124, Windows 11), and VoiceOver (Safari 17, macOS 14). ATS market-share rankings drawn from Aptitude Research's 2024 Talent Acquisition Tech buyer survey and Ongig's 2025 ATS market analysis. Vendor accessibility statements retrieved from each provider's public website in May 2026.
The audit covered the candidate-facing surface of each platform: the public job-description page, the "apply" entry point, the multi-step application form, the CV/resume upload, the equal-opportunity self-identification step, and the confirmation page. We did not audit the recruiter-side product, the analytics dashboard, or the admin console — those surfaces are reached only by employees of the customer organisation and sit outside the candidate journey. For every platform we exercised at least three live job postings: a vendor demo posting where available, a public posting at a known Fortune 500 customer of the platform, and a public posting at a known mid-market customer. Each posting was scanned with axe-core 4.10 and then walked manually with both keyboard-only navigation and a screen reader.
Manual review followed a fixed protocol on each platform: navigate from the job description to the apply button using only the Tab key; complete the first three form fields using a screen reader and verify that each field's label is announced; attempt a CV upload using the keyboard and verify the file-upload control is reachable and announces success or failure; trigger a deliberate validation error and verify that the error is associated with its field and announced; and attempt to leave a page mid-flow and return, verifying that no data is silently discarded. The pass criterion at each step was WCAG 2.2 Level AA, with axe-core's "serious" and "critical" severity tiers used as the automated proxy.
02Automated layeraxe-core 4.10 scans of each page in the flow, severity tiers "serious" and "critical" counted
03Manual layerkeyboard-only Tab traversal + NVDA, JAWS, and VoiceOver walk-throughs against a fixed protocol
04Pass criterionWCAG 2.2 Level AA for manual review; axe "serious" + "critical" counts for the automated tier
10
ATS platforms audited
30+
live job postings exercised across customer sites
approx. 412
serious or critical axe violations counted
3
platforms a screen-reader user could complete unaided
02 · The ten platforms, ranked by audit-pass rate
The composite audit-pass rate for each platform combines two inputs in equal weight: the percentage of axe-core rules cleared on a default candidate flow, and the percentage of manual-protocol steps completed without intervention by a screen-reader user. The result is a single 0–100 score, not a regulatory determination. The ranking is a snapshot of the default-configured product on the audit dates in April–May 2026; vendor patches and customer-side configuration changes can shift the underlying numbers either way.
01
Greenhouse
cleanest axe scan; only 11 serious or critical violations across the candidate flow
approx. 78
02
Lever
no label-binding failures on the default candidate flow; clean keyboard traversal
approx. 74
03
Workable
completable with NVDA and VoiceOver; persistent contrast and focus issues
approx. 67
04
SmartRecruiters
decent automated scan; validation errors not consistently announced
approx. 58
05
JazzHR
simple flow helps; label-binding gaps in custom-question fields
approx. 55
06
BambooHR
moderate axe count; modal and date-picker components leak focus
approx. 49
07
iCIMS
heavy customisation surface; unlabelled radio groups and silent time-outs
aggressive session time-out; complex custom widgets without ARIA roles
approx. 24
10
Oracle Taleo
84 serious or critical axe violations; flow could not be completed via screen reader
approx. 17
{/* Hand-built SVG bar chart replaces a FLUX-generated image whose
axis labels and platform names rendered as gibberish (AI image
models cannot draw legible text). Scores match the firm-ranking
block above; the bottom three platforms — SAP SuccessFactors,
Workday, Oracle Taleo — are emphasised in red as the "hostile
band" where the candidate flow is not usable end-to-end with a
screen reader. */}
Composite audit-pass scores across the ten platforms. Three clear the mid-line at 50 (Greenhouse 78, Lever 74, Workable 67); four sit in the failing band below it (SmartRecruiters 58 just above, JazzHR 55, BambooHR 49, iCIMS 41); the bottom three — SAP SuccessFactors 33, Workday 24, Oracle Taleo 17 — drop into the hostile band where the default candidate flow is not completable end-to-end with a screen reader. Red bars mark that hostile band.
spread between top and bottom — the same product category
03 · Failure categories — what actually breaks
Of the approximately 412 serious or critical axe violations counted across the audited flows, the breakdown by category is the more useful number than the absolute total. Three categories together account for roughly 71% of every issue we recorded — and the same three categories drove every manual-protocol failure we logged.
Distribution of serious or critical axe violations across the audited ATS candidate flows
Form-label binding (missing or programmatic)
approx. 34%
Error identification + suggestion
approx. 22%
Time-bounded interactions (uploads, sessions)
approx. 15%
Colour contrast + focus visibility
approx. 13%
ARIA misuse (wrong role, broken state)
approx. 9%
Other (keyboard trap, name-role-value, link purpose)
approx. 7%
Form-label binding is the rule axe-core flagged most often across every platform we tested. The pattern is the same in each case: the visible text adjacent to a field looks like a label to a sighted user, but it is rendered in a separate DOM node with no for/id association, no aria-labelledby, and no wrapped input. A screen reader announces "edit, blank" — and the candidate is left guessing what the field is for. Date pickers, phone-number groups, work-authorisation radio groups, and "additional information" textareas were the most common offenders.
Error identification was the second-largest category and the most consequential for the manual-protocol pass rate. Six of the ten platforms displayed validation errors using visual cues only — a red border, a red asterisk, an inline icon — with no associated aria-describedby reference, no role="alert" announcement, and no programmatic focus shift to the offending field. A screen-reader user who submits a form and receives only a "this page has errors" generic toast cannot locate the actual error without sighted help.
An applicant who cannot find the field that triggered an error cannot fix it. An applicant who cannot fix the error cannot submit the application. The accessibility failure is the rejection.
The compounding effect
Each of the three top failure categories is individually a WCAG 2.2 Level AA failure. Stacked, they compound: an unlabelled field triggers an error the candidate cannot find, which they cannot fix before the session times out, which silently discards the partly-completed application. The audit recorded this exact compounding sequence on five of the ten platforms during the manual protocol — a candidate working in good faith was deposited at a "session expired" screen with no completed application and no record they had ever applied.
04 · Time-bounded uploads and the silent-expiry failure
The single most damaging finding of the audit is the time-bounded upload pattern. Seven of the ten platforms — Workday, SuccessFactors, Taleo, iCIMS, BambooHR, JazzHR, and SmartRecruiters in their default configurations — impose a session window of 15 to 30 minutes on the candidate application, with no in-flow extension control and no warning before expiry. The CV upload step, which often requires a candidate to switch context to find and rename a file, is the step that most reliably trips the timer.
WCAG 2.2 SC 2.2.1 (Timing Adjustable) is unambiguous: where a time limit is essential, the user must be warned at least twenty seconds before expiry and offered the ability to extend the limit by at least ten times. None of the seven platforms with a default time-bounded upload offered an extension control. None provided a warning that met the SC's twenty-second threshold. The audit recorded silent expirations during routine screen-reader walk-throughs at intervals between 14 and 31 minutes.
The disparate impact
A timed application form imposes a per-second cost on candidates who navigate by keyboard, who read with a screen reader, who use a switch-input or eye-tracker, or whose disability slows their pace at the screen. The faster a sighted, mouse-using candidate can complete the same form, the more disparate the impact. In the audit, the same form that took an unimpaired user nine minutes took a screen-reader user twenty-six — comfortably outside Workday's default session window.
05 · Video interviews — HireVue and the parallel layer
Beneath the ATS layer sits a second, parallel platform layer the candidate often does not see coming: the video-interview vendor. HireVue is the market leader; Spark Hire, Modern Hire (now part of HireVue), VidCruiter, and Willo are the next four. Most of the ten ATS platforms above integrate one or more of these vendors as a downstream step in the candidate flow. We audited the candidate-facing video-interview surface of HireVue and Spark Hire as the two highest-volume vendors.
The video-interview surface introduces a category of failure the ATS layer does not have: the recorded-response prompt. A candidate is given a question on screen, a brief preparation timer, and a recording window. The accessibility failures we recorded were primarily around the prompt itself — missing or auto-translated captions on the question video, no transcript alternative, no extended-time accommodation control surfaced to the candidate, and one-shot recording windows with no clear "this is your only attempt" warning announced to a screen reader. Spark Hire performed marginally better than HireVue on caption availability; both failed the manual-protocol "complete a question end-to-end" step in our audit.
06 · The honesty gap in vendor accessibility statements
All ten audited ATS vendors publish some form of vendor accessibility statement. Three — Workday, Oracle Taleo, and iCIMS — claim WCAG 2.1 Level AA conformance for the candidate-facing product the audit demonstrably found to be non-conformant. Each statement has an explanation when you read past the headline: Workday's refers to a configurable "accessible mode" not enabled by default on customer sites; Oracle's refers to a 2019 VPAT that pre-dates two major product releases; iCIMS's covers the recruiter product, not the candidate-facing apply flow. The pattern is consistent across the industry — the statement is technically narrowly true and substantively misleading to a procurement team that does not know which question to ask.
Excerpt — vendor accessibility statement, one of the audited platforms
"Our candidate experience is designed to conform to WCAG 2.1 Level AA. We continuously test our products against international accessibility standards and welcome feedback from users with disabilities."
Public-facing vendor accessibility statement, retrieved May 2026. The audit found 47 serious or critical axe violations on the same vendor's default candidate flow on the same date.
The honesty gap matters because procurement teams treat the accessibility statement as a compliance signal — exactly as the vendor intends. A Fortune 500 talent-acquisition leader who reads "WCAG 2.1 Level AA conformance" in a pitch deck and reproduces that claim in an RFP response signs a vendor onto the candidate pathway whose default behaviour the leader has not in fact verified. The employer inherits the vendor's accessibility liability — and under ADA Title I and EAA Article 4, the employer is the duty-bearer, not the vendor.
07 · ADA Title I and EAA Article 4 — why this is employment law
The legal frame for ATS accessibility is not the customer-facing Title III of the ADA but Title I — the employment provisions. Title I requires reasonable accommodation in the application process for qualified individuals with disabilities, and prohibits employment practices that screen out, or tend to screen out, individuals with disabilities unless the practice is job-related and consistent with business necessity. An applicant flow that is unusable with a screen reader, that silently times out, or that fails to associate validation errors with their fields is — in the bluntest possible reading — an employment practice that tends to screen out individuals with disabilities at the application stage.
In the European Union, the European Accessibility Act, Directive (EU) 2019/882, has applied to in-scope products and services since 28 June 2025. Article 4 of the Directive extends the accessibility obligation to "consumer services" and to ancillary services connected to the products in Annex I, with specific applicability to recruitment platforms varying by Member-State transposition. Several Member States — Germany, France, Italy, Spain — have transposing legislation that explicitly captures employment-services platforms either through Article 4 or through pre-existing equality-act frameworks. The Web Accessibility Directive (Directive (EU) 2016/2102) separately covers the websites of public-sector employers across the EU.
Who is the duty-bearer?
Under ADA Title I, the employer who uses the ATS is the duty-bearer — the EEOC has been explicit that a covered employer cannot delegate its non-discrimination obligation to a software vendor by procurement contract. Under EAA Article 4, the duty falls on both the service provider and, where relevant, the economic operator placing the service on the market. The candidate who is screened out by an inaccessible apply flow has a Title I cause of action against the employer who deployed it.
The EEOC's May 2022 technical-assistance document on the use of artificial intelligence and software in employment decisions, and its follow-on May 2023 update, are directly applicable to ATS accessibility. The agency's framework treats the application-stage software as part of the employer's selection process. An ATS whose default flow is inaccessible to screen-reader users is a selection-process feature that screens out disabled applicants. The legal vulnerability runs to the employer's HR director, not to the vendor's product manager.
08 · What employers and vendors should do next
The audit's three operational findings translate to three actions, in priority order.
Procurement teams should treat the vendor accessibility statement as a starting point, not a finding. Require the vendor to demonstrate the candidate-facing flow with a screen reader on a default-configured instance — not a customer-customised one, not the recruiter product, not "accessible mode" if it is not the default. Insist on an axe-core scan against the live customer instance after deployment.
Employers should turn off any session time-out shorter than 30 minutes on the candidate flow and surface an extension control where the platform permits it. Where the platform does not permit it, raise the issue formally with the vendor and document the resulting accommodation pathway. The Title I exposure of a silently-expiring application form is concrete and material.
Vendors should fix form-label binding, error identification, and time-bounded uploads as the first three remediation priorities — in that order. Those three categories alone account for roughly 71% of the audited violations and explain almost every manual-protocol failure we recorded. Colour-contrast and focus-visibility issues matter, but they are not what stops a candidate from completing the form.
The applicant funnel is the front door
Every conversation about disability inclusion in the workforce eventually reaches the same observation: the labour-market participation rate for working-age disabled adults remains far below that of their non-disabled peers, year after year, across the OECD. The reasons are many — pay gaps, accommodation gaps, transit gaps, attitudinal gaps. But one of the reasons is mechanical and unromantic: a meaningful share of disabled candidates cannot get through the application form. The audit recorded this directly. Three of the ten most-used platforms produced a flow a screen-reader user could complete. Seven did not.
The ATS layer is the front door to the labour market for the majority of US Fortune 500 employers and a comparable share of large EU employers. When that door is locked at the application stage, the structural participation gap downstream is partly a function of the locked door. The audit's role is not to assign blame to the vendor or the employer in isolation; it is to register that the door is locked and to identify, in concrete and verifiable terms, what is locking it.
---
title: Australia's DDA and the state-level accessibility mosaic
url: https://www.disabilityworld.org/articles/australia-dda-and-the-mosaic/
description: Australia's Disability Discrimination Act 1992 is a federal non-discrimination law applied to digital services through case law since Maguire v SOCOG (2000).
author: Disability World
pubDate: 2026-05-22
tags: australia, dda, regulations, regulation-primer, asia-pacific
---
# Australia's DDA and the state-level accessibility mosaic
Image description: Australian Parliament House in Canberra at golden hour, the Australian flag flying above the building's iconic flagpole — institutional anchor for the DDA framework.
Reading Time: 12 minutes
Australia's Disability Discrimination Act 1992 (Cth) — the DDA — is the federal anchor for disability rights in Australia, and the operative statute for digital-accessibility complaints across the country. It is not a digital-accessibility law: it makes no mention of WCAG, web content, or applications. It is a general-purpose non-discrimination statute that was drafted before the public web existed, and it has been stretched to cover digital services almost entirely through case law and through subordinate guidance. The landmark act of stretching happened early — Maguire v Sydney Organising Committee for the Olympic Games (HREOC, 2000), the world's first formal ruling that a website's inaccessibility constituted unlawful discrimination — and twenty-five years later it still sets the legal frame within which digital-accessibility complaints are filed. For broader regional context, see our national regulations index and the CRPD twenty-year retrospective.
Layered above the DDA is a mosaic that is genuinely federal in structure: state and territory anti-discrimination statutes — Victoria's Equal Opportunity Act 2010, New South Wales's Anti-Discrimination Act 1977, Queensland's Anti-Discrimination Act 1991, and equivalents in each of the remaining jurisdictions — operate in parallel, with their own commissioners and their own tribunals. The Australian Human Rights Commission (AHRC) runs the federal conciliation procedure. The Digital Transformation Agency (DTA) sets the procurement reference standard for Commonwealth-owned digital services at WCAG 2.1 AA and adopts AS EN 17161:2020 (the Australian adoption of the European universal-design standard) as the planning reference. This primer maps that mosaic — what the DDA actually does, where the state statutes pick up what it leaves out, how a complaint is filed and what the remedies look like, and where the Commonwealth procurement levers sit.
What the DDA is — and what it isn't
The DDA was enacted in 1992 by the Keating government on the strength of the External Affairs power in section 51(xxix) of the Australian Constitution, grounded in Australia's obligations under international human-rights instruments — at the time, the ILO conventions and the UN Declaration on the Rights of Disabled Persons. Australia ratified the UN Convention on the Rights of Persons with Disabilities (CRPD) in 2008 and the Optional Protocol in 2009. The DDA predates the CRPD by sixteen years; the Convention has not been incorporated into domestic legislation, but it operates as an interpretive lens in DDA cases and as the explicit driver of the Australian Disability Strategy 2021–2031.
In its current form the DDA prohibits direct and indirect discrimination on the ground of disability across a defined list of public-life areas: work, education, access to premises, the provision of goods, services and facilities, accommodation, land, clubs and incorporated associations, sport, and the administration of Commonwealth laws and programs. Section 24 — the goods-services-and-facilities provision — is the operative section for digital services. It is not bounded by medium: a discriminatory practice that occurs through a website or a mobile application falls within section 24 in exactly the way an inaccessible counter at a bank branch would. That is the legal hook that made Maguire possible.
The "unjustifiable hardship" defence
The DDA's most-litigated defence is section 11 — the "unjustifiable hardship" provision. A respondent who concedes that a practice would otherwise be discriminatory can argue that providing the accommodation would impose unjustifiable hardship, weighed against factors that include the nature of the benefit or detriment, the effect of the disability, the financial circumstances of the respondent, the estimated expenditure required, and the availability of financial and other assistance. The defence is fact-specific and has often been decisive: a small business with thin margins can mount it credibly; a Commonwealth agency or a major retailer cannot.
Disability Standards under the DDA
Section 31 of the DDA empowers the Attorney-General to make Disability Standards that elaborate the Act's general obligations in specific sectors. Three are currently in force: the Disability Standards for Education 2005, the Disability Standards for Accessible Public Transport 2002 (DSAPT), and the Disability (Access to Premises – Buildings) Standards 2010. There is no Disability Standard for digital accessibility. Successive reviews — most recently the 2021 review of the DSAPT — have raised the question; the Commonwealth has consistently preferred to address digital accessibility through the DTA's procurement guidance and through the WCAG reference, rather than by issuing a binding Standard under section 31. The absence of a digital Standard is a structural feature of the Australian framework, not an oversight.
Maguire and its shadow: how case law made the DDA a digital statute
The 2000 ruling in Bruce Maguire v Sydney Organising Committee for the Olympic Games (SOCOG) is the foundational document for digital-accessibility law in Australia, and arguably anywhere. Bruce Maguire, a blind man, complained to the Human Rights and Equal Opportunity Commission (HREOC — the predecessor of the AHRC) that the official Sydney 2000 Olympics website was inaccessible to him because images lacked alternative text, the medal-tally tables could not be read with a screen reader, and the Index to Sports page was structurally unusable. The Commissioner, William Carter QC, found that SOCOG had unlawfully discriminated against Maguire under section 24, rejected the unjustifiable-hardship defence (SOCOG's evidence that the changes would have taken 368 person-days was not, on the record, sufficient), and ordered SOCOG to make the website accessible and to pay AUD 20,000 in damages.
The damages number was small. The precedent was not. Maguire established three propositions that are still load-bearing: that a website is a "service" within the meaning of section 24; that the unjustifiable-hardship test is not satisfied merely by pointing to engineering cost; and that compliance with international accessibility guidance (then the W3C's Web Content Accessibility Guidelines 1.0) was the operative reference for what an accessible website looks like. The Commonwealth published its first mandatory web-accessibility policy for federal government sites the same year. Every subsequent Australian digital-accessibility complaint has been litigated in the shadow of Maguire.
The post-Maguire docket
The volume of formally decided DDA digital cases is small — the AHRC's procedure is built around conciliation rather than adjudication, and most matters that surface a real problem settle. Notable post-Maguire matters include the 2014 conciliation against a major Australian bank over an inaccessible upgraded online-banking interface (settled with a remediation program and undisclosed compensation); the 2019 conciliation against a state public-transport ticketing app (resolved through redesign and the publication of an accessibility roadmap); and a 2023 conciliation against a major retailer's checkout flow under section 24, again resolved by settlement. The pattern is consistent: complainants do not, as a rule, need to litigate; the legal exposure is enough to bring respondents to the table.
The state mosaic: Equal Opportunity Acts and parallel commissioners
The DDA does not pre-empt state and territory anti-discrimination law. A person who believes they have been discriminated against on the ground of disability in Australia can lodge a complaint federally under the DDA — via the AHRC — or in their state or territory under that jurisdiction's statute. The choice has consequences: forum, available remedies, and the tribunal that hears any escalation differ. The three largest state regimes are summarised below.
Jurisdiction
Statute
Commissioner
Escalation tribunal
Federal
Disability Discrimination Act 1992 (Cth)
Australian Human Rights Commission (AHRC)
Federal Circuit and Family Court of Australia
Victoria
Equal Opportunity Act 2010
Victorian Equal Opportunity and Human Rights Commission (VEOHRC)
Victorian Civil and Administrative Tribunal (VCAT)
New South Wales
Anti-Discrimination Act 1977
Anti-Discrimination NSW
NSW Civil and Administrative Tribunal (NCAT)
Queensland
Anti-Discrimination Act 1991
Queensland Human Rights Commission
Queensland Civil and Administrative Tribunal (QCAT)
Western Australia
Equal Opportunity Act 1984
Equal Opportunity Commission WA
State Administrative Tribunal (SAT)
South Australia
Equal Opportunity Act 1984
Equal Opportunity Commission SA
SA Civil and Administrative Tribunal (SACAT)
Tasmania
Anti-Discrimination Act 1998
Equal Opportunity Tasmania
Tasmanian Civil and Administrative Tribunal (TASCAT)
ACT
Discrimination Act 1991
ACT Human Rights Commission
ACT Civil and Administrative Tribunal (ACAT)
Northern Territory
Anti-Discrimination Act 1992
Anti-Discrimination Commission NT
NT Civil and Administrative Tribunal (NTCAT)
Victoria — Equal Opportunity Act 2010 and the positive duty
Victoria's Equal Opportunity Act 2010 is the most ambitious of the state regimes. Section 15 imposes a positive duty on duty-bearers to take reasonable and proportionate measures to eliminate discrimination, sexual harassment and victimisation as far as possible. The positive duty is anticipatory: it does not require a complainant to surface. The Victorian Equal Opportunity and Human Rights Commission (VEOHRC) can investigate and seek enforceable undertakings where the duty appears not to have been met. Victoria also operates the Charter of Human Rights and Responsibilities Act 2006, which adds a separate interpretive layer for public authorities. Together they give Victoria the strongest state-level disability framework in the country, and the most aggressive in respect of digital practice.
New South Wales — Anti-Discrimination Act 1977
NSW's Anti-Discrimination Act 1977 is the oldest of the state regimes and the most piecemeal. It covers discrimination across work, education, the provision of goods and services, accommodation and registered clubs. It carries no positive duty. The Anti-Discrimination Board of NSW (operating as Anti-Discrimination NSW) handles conciliation; escalation runs to NCAT. The NSW Law Reform Commission has been reviewing the Act since 2024 with a brief to consider, among other things, a positive duty modelled on Victoria's section 15. Whether that recommendation survives the legislative process will determine whether NSW's regime moves closer to Victoria's in this decade.
Queensland's regime is in transition. The Anti-Discrimination Act 1991 has been the operative statute for thirty-four years; the Respect at Work and Other Matters Amendment Act 2024 and the subsequent Anti-Discrimination Bill 2024, passed in 2025, are restructuring the framework around a Victoria-style positive duty and updating the protected-attribute list. The Queensland Human Rights Commission — which also administers the Human Rights Act 2019, the only Commonwealth-style statutory bill of rights at state level outside Victoria and the ACT — has correspondingly broader investigative scope. For 2026 complainants, the practical effect is a regime moving in the direction of Victoria's, but with a transitional period during which both the old and new procedural rules will be live depending on the date the discrimination occurred.
The state-vs-federal choice in practice
A complainant cannot pursue both forums simultaneously. Most digital-accessibility complaints — where the respondent is a national or multinational entity — are lodged with the AHRC under the DDA, because the federal forum gives clean coverage of the respondent's nationwide operations and is the forum in which precedent (including Maguire) is read. Complaints against state or local government entities, against state-specific service providers, or by complainants who want to invoke a positive duty like Victoria's are typically lodged at state level. Practitioners advising complainants generally recommend the state forum where a positive duty exists and a Commission investigation is plausible, and the federal forum where conciliation against a national respondent is the realistic remedy.
How a DDA complaint actually works: the AHRC procedure
The AHRC procedure is built around conciliation — it is not, in the first instance, an adjudicative body. A complaint is lodged in writing, the AHRC assesses it for jurisdiction and prima-facie merit, and where it proceeds the Commission convenes a conciliation conference. The Commission has wide powers to require attendance and documents but no power to impose a binding remedy. If conciliation succeeds, the matter is closed on the negotiated terms (which are typically confidential). If conciliation fails or the respondent declines to participate, the complainant may apply to the Federal Circuit and Family Court of Australia for a determination — which is then a full adversarial proceeding under the DDA.
The remedies the Federal Circuit and Family Court can order include declarations that the respondent's conduct was unlawful, orders to redress loss or damage (including general damages for hurt, humiliation and distress), orders requiring the respondent to perform reasonable acts to redress the loss, and orders that the respondent not repeat or continue the unlawful discrimination. Damages awards in DDA cases are conventionally modest by international standards — typically in the AUD 5,000 to 50,000 range, with rare awards above — but the operative consequence is usually the structural-relief order (the website remediation, the policy change, the staff training) rather than the cash component.
The twelve-month trigger
A DDA complaint must be lodged within twenty-four months of the alleged discrimination — extended from the original twelve in 2022. The clock runs from the act, not from the complainant's awareness, which has caused difficulty in digital cases where the discriminatory practice is continuous. AHRC practice is to treat a continuing inaccessible interface as a continuing act for limitation purposes, but the point is not closed at appellate level.
The DTA's procurement lever: WCAG 2.1 AA and AS EN 17161
The Digital Transformation Agency (DTA) is the Commonwealth body responsible for whole-of-government digital policy. It operates the Digital Service Standard — the design standard for Commonwealth-funded digital services — and the procurement frameworks that touch every digital contract above the relevant threshold. The DTA does not legislate; it sets the rules under which Commonwealth agencies buy and build digital services, and through those rules it does most of the work that a domestic digital-accessibility Standard under section 31 of the DDA would otherwise do.
The operative references are WCAG 2.1 Level AA for web and mobile content, and AS EN 17161:2020 — Design for All: Accessibility through universal design for the broader planning and design lifecycle. AS EN 17161 is the Australian adoption of the European standard EN 17161:2019, picked up by Standards Australia in 2020 as a non-binding reference for process-level universal design. The DTA's guidance frames the two together: WCAG 2.1 AA as the outcome standard for any web or mobile interface, AS EN 17161 as the process standard for how to plan and design the service that the interface fronts. Commonwealth procurement contracts increasingly cite both, and major-tender Statements of Work routinely require conformance reporting against WCAG 2.1 AA across the lifecycle.
The WCAG 2.1 vs 2.2 question
The DTA's reference is still WCAG 2.1 AA, not 2.2. The W3C published WCAG 2.2 as a Recommendation in October 2023, and the Australian Government Information Management Office signalled in 2024 that an update of the procurement reference to 2.2 was under consideration. As of early 2026 the formal reference remains 2.1; tenders may require 2.2, but the policy floor is 2.1. The lag between W3C Recommendation and Commonwealth adoption is a recurring feature of the Australian framework.
The Australian Disability Strategy 2021–2031
Sitting above the DDA and the state mosaic is the Australian Disability Strategy 2021–2031, the federal policy framework that succeeds the 2010–2020 National Disability Strategy. The Strategy is not legislation: it is the Council of Australian Governments-era settlement, refreshed in 2021, that aligns Commonwealth, state and territory policy with the CRPD. It identifies seven outcome areas — employment, inclusive homes and communities, safety, rights, health and wellbeing, learning and skills, and personal and community support — and sets out outcome measures and targeted action plans. The Strategy is the document a Commonwealth or state agency consults when designing a disability-related program; it is the operational layer above the legal layer.
The Strategy's accountability mechanism is the Australian Institute of Health and Welfare (AIHW), which publishes annual outcomes-framework reports. The framework includes indicators on participation in employment, education, the digital economy and public life. The 2025 report flagged that the digital-economy participation gap between persons with and without disability had narrowed slightly since 2021 but remained material, and that progress was uneven across states — Victoria and the ACT outperforming, the Northern Territory and remote-Queensland indicators lagging.
Practical implications: how to read the mosaic in 2026
For organisations operating in Australia — particularly those whose digital services reach a national audience — the practical map is straightforward in outline and intricate in detail. The DDA is the federal floor: a national digital service that fails accessibility under section 24 is exposed to a federal complaint regardless of where the user resides. The state Equal Opportunity Acts add parallel coverage with state-specific positive duties (Victoria, now Queensland) and state-specific remedy ranges. The DTA's procurement framework sets the explicit standard — WCAG 2.1 AA, with AS EN 17161 as the process reference — for any service the Commonwealth pays for. The CRPD, although not domestic law, is the interpretive lens through which the AHRC, the state commissioners and the courts increasingly read the older statutes.
For complainants, the choice of forum matters: federal for national respondents and precedent value; state for positive-duty investigations and state-specific service providers. For respondents, the practical floor is WCAG 2.1 AA conformance with a documented remediation programme — the AHRC's conciliation procedure rewards organisations that arrive with a credible plan, and the federal court is unforgiving of those that arrive without one. The legal exposure under section 24 of the DDA is substantively the same in 2026 as it was in 2000; the body of guidance — DTA references, state-commissioner determinations, post-Maguire conciliations — that defines what compliance looks like is much thicker.
Conclusion: an old statute, a thickening guidance layer
The DDA is now thirty-four years old. It has been amended substantively only a handful of times — in 2009 to align more closely with the CRPD, in 2022 to extend the complaint window, and through periodic Disability Standards. Its survival as the legal anchor of Australian disability rights through the digital revolution is a feature of its drafting: the section 24 goods-services-and-facilities provision was written broadly enough to absorb media that did not exist in 1992. Maguire was a stretch in 2000; it is settled law in 2026.
What has changed is the surrounding layer — the state positive duties, the AHRC's procedural practice, the DTA's procurement references, the CRPD's interpretive weight, the Australian Disability Strategy's outcomes framework. The mosaic is the framework, not the DDA alone, and an organisation that reads the DDA without reading the surrounding layer is reading the floor for the ceiling. For the next steps in this series, see our forthcoming primers on the New Zealand framework and on the Asia-Pacific regional comparison; for the comparative view, the CRPD twenty-year retrospective and the national regulations index situate Australia within the wider 2026 picture.
---
title: Braille production pipelines in 2026: software, hardware, and workflow
url: https://www.disabilityworld.org/articles/braille-production-pipelines-2026/
description: An end-to-end engineering tour of how braille actually gets made in 2026 — the software stack from Duxbury DBT and Liblouis to BrailleBlaster and RoboBraille, the embosser families from Index and Enabling Technologies, and the source-to-paper workflow that connects them.
author: Disability World
pubDate: 2026-05-22
tags: braille, embossers, transcription, education, blindness, production
---
# Braille production pipelines in 2026: software, hardware, and workflow
Engineering primer · Braille production
Braille production pipelines in 2026:
software, hardware, and workflow
Braille is not a font. It is a translation, a typesetting decision, and a paper artifact — three problems that have to be solved together. Here is the production stack engineers and transcribers actually use in 2026: the translation engines, the embosser families, the source formats they accept, and the quality-control loop that keeps a braille page faithful to the print it came from.
4
major translation engines
7
embosser families surveyed
6
source formats handled
By Disability World engineering desk
13 min read
Updated May 2026
Foundation
1. The software stack: four engines that do the actual translation
A braille page is the output of a translation, not the rendering of a font. The translator takes a stream of printed characters and emits a stream of braille cells, applying contractions, capitalization markers, number indicators, and language switches according to a code — Unified English Braille (UEB) for most English-language work, Nemeth Code for math, and one of dozens of language tables for non-English text. The translator is where most of the engineering complexity in a braille pipeline lives; everything else in the pipeline is plumbing around it.
Four engines dominate the 2026 landscape. Duxbury Braille Translator (DBT) from Duxbury Systems is the commercial industry standard, in continuous development since the 1970s and still the reference implementation against which other tools are measured. Liblouis is the open-source back-end translation library that quietly powers a large fraction of everything else — NVDA, Orca, BrailleBlaster, the BRLTTY console driver, and a long tail of in-house tooling. Sao Mai Center BrailleBlaster, an open-source production app developed by the American Printing House for the Blind on a Liblouis core, has become the default for K–12 textbook transcription in the United States. RoboBraille is a free cloud service operated by the Danish nonprofit Sensus that turns an uploaded document into a downloadable braille file in approximately a few minutes — useful for ad-hoc requests where buying a Duxbury seat is not justified.
The four engines do not compete on the same axis. Duxbury sells certainty: a tested commercial product with a support contract, a documented certification path for transcribers, and the longest provenance of any tool on this list. Liblouis sells embeddability: it is the library you call from your own application when you do not want to ship a GUI. BrailleBlaster sells a textbook-grade authoring experience with rich math, image-description, and tactile-graphics workflows built in. RoboBraille sells convenience — drag a Word file onto a web form and a braille file comes back.
Duxbury DBT
Duxbury Systems · commercial · Windows + macOS
Industry standard for production braille houses since the 1970s
CodesUEB, Nemeth, music braille, plus approx. 130 language tables
Open-source library · LGPL · embedded in dozens of tools
The translation engine behind NVDA, Orca, BrailleBlaster, BRLTTY
CodesUEB, Nemeth, plus approx. 180 community-maintained tables
Source formatsPlain text in; the surrounding app handles file parsing
LicenseLGPL — free to embed, free to redistribute
BrailleBlaster
APH + Sao Mai Center · open-source · Java · cross-platform
Default authoring tool for US K–12 braille textbook production
CodesUEB, Nemeth, via Liblouis back-end
Source formatsNimasXML, DAISY, EPUB, Word, HTML
LicenseGPL — free download, supported by APH
RoboBraille
Sensus (Denmark) · cloud · web upload + email
Free public service since 2004 · approx. 2 million conversions per year
CodesUEB, plus most European-language tables
Source formatsWord, PDF, EPUB, image (OCR), HTML, text
LicenseFree for individuals; institutional plans available
i
Liblouis is the quiet center
If you trace the dependency graph of "open-source braille" in 2026, Liblouis sits at the root of most of it. BrailleBlaster calls Liblouis. NVDA calls Liblouis. Orca calls Liblouis. RoboBraille's English path calls Liblouis. The library is not a competitor to BrailleBlaster — it is a layer underneath it. When a new language table lands in Liblouis upstream, every downstream tool inherits it at the next release.
Hardware
2. Embosser hardware: from desktop units to interpoint production rigs
The embosser is the mechanical opposite of an inkjet: instead of laying ink onto a page, two opposing dies pinch a sheet of heavyweight paper between them and push raised dots up from below. Every embosser is some variation of that basic geometry, but the machines vary by an order of magnitude in throughput, page format, and whether they print on one side of the paper or both at once (interpoint).
Two manufacturers dominate the market in 2026. Index Braille (Sweden) ships three lines used widely in schools and small braille houses: Basic-D for single-sided desktop work, Everest-D for interpoint sheet-fed production, and Braille Box for high-volume booklet output. Enabling Technologies (US) ships the Romeo, Juliet, and ET families, used historically by US braille publishers and still the workhorse of many state-level Instructional Resource Centers. Tiger by ViewPlus sits in its own category — a tactile-graphics-capable embosser whose Tiger Cub Jr is the most common entry-level model for STEM classrooms that need raised graphs alongside braille text.
Throughput is measured in characters per second (CPS), but in practice the more useful metric is "pages per minute" because braille pages are short — typically 25 lines of 40 cells, around 1,000 cells total. A desktop unit at 100 CPS produces a single-sided page in about 10 seconds plus paper-handling time, so call it 4 pages per minute. A production interpoint unit at 800 CPS, double-sided, lands closer to 100 pages per minute when fed continuous-form paper. The difference is not 8× — it is closer to 25× — and that gap is what separates a classroom embosser from a production-house embosser.
Throughput
Format
Interpoint
Typical role
Index Basic-D V5
approx. 110 CPS
Sheet-fed, single-sided
No
Classroom, library, small office
Index Everest-D V5
approx. 140 CPS
Sheet-fed, double-sided
Yes
Mid-volume transcription centers
Index Braille Box V5
approx. 300 CPS
Sheet-fed booklets, double-sided
Yes
Library and publisher booklet runs
Enabling Romeo Attache
approx. 15 CPS
Tractor-fed, single-sided
No
Portable / individual user
Enabling Juliet 120
approx. 120 CPS
Tractor or sheet-fed
Yes
School district, university
Enabling ET Series
approx. 800 CPS
Continuous-form, double-sided
Yes
State IRC, commercial braille house
ViewPlus Tiger Cub Jr
approx. 60 CPS
Sheet-fed plus tactile graphics
No
STEM classroom, math + diagrams
Three observations from the matrix. First, the price-to-throughput curve is steep — a Basic-D lists around $4,000, a Juliet around $4,500, and an ET-series production embosser can pass $80,000. A buyer's decision is rarely about CPS in isolation; it is about how many pages per day the institution needs to ship and whether the paper handling is sheet-fed (classroom) or continuous-form (production-line).
Second, interpoint is the dividing line between "small" and "real" braille production. Single-sided braille doubles the page count and roughly doubles the binding cost of a finished book. Every embosser intended for textbook-scale work is interpoint; every embosser intended for individual or small-run work is typically not.
Third, the Tiger family solves a different problem from the rest. ViewPlus embossers can vary dot height across a page to produce tactile graphics — raised line drawings, bar charts, geographic maps — alongside braille text. For STEM material that fix is not optional; a braille-only embosser can render the text of a graph's caption but not the graph itself. Where math and diagrams matter, the Tiger pays for itself in workflow time even at lower text throughput.
+
The driver is the bottleneck more often than the head
Embosser throughput is rarely limited by the embossing head. It is limited by paper handling — sheet feeders jam, continuous-form perforations tear, and operator intervention costs minutes per error. The buying decision should weight paper-path reliability at least as heavily as quoted CPS, because the mean-time-between-jams determines actual pages-per-day far more than the raw mechanical rate.
Workflow
3. The workflow end-to-end: source format to embossed page
A braille production job moves through five recognizable stages. Source preparation cleans up the print original. Translation converts characters to braille cells. Formatting lays out the cells on the page. Embossing puts dots on paper. Proofreading verifies that the embossed output matches the original. Skipping or compressing any of these stages is the most common cause of a production error, because each stage catches mistakes the previous one introduced.
1
Source preparation
Take the print original (Word, InDesign, PDF, EPUB, HTML, MathML for equations) and clean its semantics. Tag headings as headings, lists as lists, footnotes as footnotes. The translator can only emit good braille from good structure; an untagged PDF is a worst-case input that requires hours of manual restructuring before translation can begin.
2
Translation
Run the prepared source through Duxbury, BrailleBlaster, or a Liblouis-based tool. Choose the right code — UEB for contemporary English, EBAE only for legacy material that requires it, Nemeth or UEB-with-Nemeth for math. Switch language tables explicitly at every non-English passage. The translator emits Braille ASCII or a binary braille file (.brf, .brl) ready for the embosser.
3
Formatting
Decide page geometry — lines per page, cells per line, running heads, page numbering. BANA Formats 2016 (the de facto standard in North America) prescribes most of this for textbooks. Set tactile-graphic placement, table indentation, and side-by-side facing-page layout for interpoint output. This is where transcribers spend the most production time.
4
Embossing
Send the formatted .brf to the embosser via USB, Ethernet, or — for large jobs — an embosser-server queue. Use heavyweight braille paper (typically 100-pound stock for interpoint to keep dots from flattening on the verso). Monitor the run; an interpoint job that misregisters by even one cell between sides ruins the page.
5
Proofreading
A certified braille proofreader reads the embossed output by hand against the original print, flags errors, and returns the corrections to the transcriber. For US textbook work this stage is mandatory under NLS and BANA standards. The proofreader is not optional — a 400-page math textbook contains thousands of opportunities for a single dot-position error to invert meaning.
The shape of a production team follows the workflow. A small school district might collapse stages 1–4 onto a single transcriber and outsource proofreading to a freelance contractor. A state Instructional Resource Center splits the stages across specialists: a NimasXML preparer, a transcriber, a formatter, an embosser operator, and a proofreader. A commercial braille publisher adds editorial production and a binding line on top of all of that. The same five stages run; only the headcount differs.
Special cases
4. Math, multilingual, and the formats that fight the translator
Three classes of source content stress the pipeline harder than ordinary running text. Mathematical notation, mixed-language documents, and graphics-laden source files each demand specific decisions before translation begins, and each is the most common cause of a production rerun.
Mathematics is the headline problem. In English-speaking jurisdictions there are two competing codes — Nemeth, in use since 1952 and still the default in many US transcription centers; and UEB Technical Notation, the math extension of UEB that the rest of the English-speaking world has standardized on. A 2026 US transcriber will often emit "UEB with Nemeth math" — UEB for the prose, Nemeth switched in at every equation, then back to UEB — and the translator has to mark the switch with explicit indicators the reader's fingers can find. The source format matters: MathML embedded inside an EPUB or NimasXML gives the translator structured equations to convert; an equation rendered as a flat PDF image gives the translator nothing and forces the transcriber to retype every formula by hand.
Multilingual text raises an analogous problem. UEB encodes English. French, Spanish, Arabic, Korean, and dozens of other languages each have their own braille tables, often with multiple historical variants. A single book that quotes a paragraph in French inside an English narrative needs an explicit language switch in the source — usually a Liblouis directive or a NimasXML xml:lang attribute — so the translator pulls in the French table for the foreign passage and switches back at the closing quotation mark. Without that markup, the translator runs French through the English table and produces gibberish on the page.
Source · do not
```html
The opening line — Je pense, donc je suis —
was the start of the modern philosophical tradition.
```
The French quotation is marked as emphasis but not as a language switch. The translator will apply the UEB table to the French phrase and emit nonsense — UEB contractions will fire on words that are not English. The error is invisible in the print source and only surfaces on the embossed page.
Source · do
```html
The opening line —
Je pense, donc je suis —
was the start of the modern philosophical tradition.
```
The lang="fr" attribute tells the translator to switch tables for that span. Liblouis and the BrailleBlaster pipeline read the attribute, pull in the French table, emit French braille for the quotation, and switch back to UEB after the closing tag. The error mode disappears.
The third class of problem is graphics. An image in a print source can carry information that the surrounding paragraph does not repeat — a chart whose caption says "see figure" but whose values are not in the prose; a diagram whose labels are part of the picture rather than the text. The braille production team has three options for each image: a textual description embedded next to where the image was; a tactile graphic produced on a swell-paper or microcapsule machine and bound in alongside the braille pages; or a tactile graphic embossed inline by a ViewPlus Tiger from a vector source. The third option keeps the page count manageable but only works when the original image is available as vector, not as a flattened raster.
!
PDF is the worst input
A tagged Word document or a NimasXML file gives the translator structured input — paragraphs, headings, lists, language attributes, MathML equations — that can be translated directly. A flat PDF gives the translator a stream of glyphs and forces the transcriber to rebuild structure by hand. If you have any choice in the source format, send the transcriber the original Word, InDesign, or EPUB. PDF is a printout, not a source document; treat it that way.
Quality
5. Quality control: NLS standards, BANA certifications, and what gets checked
Braille that comes off an embosser is not finished braille. A production-grade pipeline ends with a quality-control loop that catches translation errors, formatting errors, and embossing errors before the pages reach a reader. In North America that loop is structured by two institutions. The National Library Service for the Blind and Print Disabled (NLS), part of the Library of Congress, sets the standards for braille books that circulate through its network. The Braille Authority of North America (BANA) maintains the formatting rules and certifies the transcribers and proofreaders who do the work.
BANA's certification program runs two main tracks. The Library of Congress / NLS certification in literary braille transcription requires the candidate to transcribe a sample book — historically about 35 pages — and pass it through a juried review. The certification in Nemeth Code (math) transcription is a separate, harder track with its own sample-book requirement. There are parallel certifications for music braille, for tactile graphics, and for proofreading. The credentials are not legally required to produce braille, but they are required to produce braille for the NLS network and are de facto required by most state IRCs and large publishers.
1
Translator-output review
Before the file goes to the embosser, a second transcriber (or the same transcriber the next morning) reads through the braille file looking for translator errors — wrong code switches, missed language tables, contractions firing inside proper names. This stage catches roughly half of all production errors and costs nothing but a second pair of eyes.
2
Format check against BANA Formats 2016
Verify running heads, page numbering, line and cell counts, table indentation, and the use of transcriber's notes. The BANA Formats document is approx. 300 pages; a checklist condenses the most common formatting decisions into a single page that the formatter can sign off on before embossing begins.
3
Embossed proofreading by a certified proofreader
A certified braille proofreader reads the embossed pages by hand against the print original. They flag substantive errors (mistranslations, wrong contractions, missing math indicators) for correction and re-embossing, and trivial errors (a single missing dot in a low-stakes word) for log-and-release. This is the stage NLS and BANA standards specifically require.
4
Sample-page tactile test
Run a fingertip across a sample page and check dot height, dot spacing, and paper warp. Dots that are too low to read reliably indicate a tired embossing head or a paper that is too light for the job. Dots that crush when handled indicate the verso side has not registered with the recto. This stage takes 30 seconds and prevents shipping a print run that reads poorly.
5
Re-emboss and final pass
Corrections from the proofreader go back through translation and formatting, re-emboss the affected pages or signatures, and run the final pass against the original. For book-length work the loop often runs two or three times before the pages ship. The discipline is what separates a circulation-library copy from a hobbyist print.
Outside North America the institutional structure differs but the quality logic is identical. The UK Association for Accessible Formats (UKAAF) issues equivalent codes and recommendations; ICEVI runs international standards work for braille production in low-resource contexts; the Marrakesh Treaty (in force since 2016) provides the legal framework that lets accessible-format works cross borders, which means a braille edition produced under one country's standards now circulates much more widely than a decade ago.
"The embosser does what you tell it. The translator does what the table tells it. The transcriber is the only place in the pipeline where judgment lives — and the certifications exist because that judgment cannot be automated away."
— a long-time observation from braille production houses
Conclusion: the pipeline is the product
Braille is one of the oldest accessible-format technologies still in continuous production use — Louis Braille published his code in 1829 — and the 2026 pipeline that produces it is a layered stack that has accreted across two centuries. The translation engines have a software heritage that goes back to the 1970s minicomputers. The embosser families have hardware lineages that go back to the 1980s. The standards have institutional histories that go back to the 1930s NLS. And every layer matters: a state-of-the-art translator on a broken embosser produces unreadable pages; a perfect embosser fed by an untagged PDF produces grammatically wrong braille on beautiful paper.
The recurring observation across every part of this stack is that quality is a property of the pipeline, not of any single component. Duxbury, BrailleBlaster, Liblouis, and RoboBraille all produce competent translations when fed competent source. Index, Enabling Technologies, and ViewPlus all produce competent dots when fed competent files. The institutional layer — NLS standards, BANA certifications, the proofreading loop — exists to verify that the whole chain held together, because a single weak link drops the quality of the finished book to the quality of the weakest stage.
That structural shape — a chain of layered specialists with a closing verification step — is older than any of the software in it. The software changes; the workflow does not. An engineering team setting up a new braille production line in 2026 will spend most of their time not on the tools but on the connections between them, which is exactly where every prior generation of braille producers spent theirs.
"A braille page is the most legible accessible-format artifact ever invented, and the least forgiving to produce. Get the pipeline right and the page reads itself. Get any one stage wrong and the reader carries the cost."
— the engineering principle that runs through every layer of the stack
---
title: The captioning lawsuit cluster: streaming, university, and live-event suits 2023-2026
url: https://www.disabilityworld.org/articles/captioning-lawsuit-cluster/
description: Three years of captioning litigation — streaming, university, live-event — have moved the legal centre of gravity from caption existence to caption quality.
author: Disability World
pubDate: 2026-05-22
tags: captioning, deaf, hard-of-hearing, ada, litigation, streaming, universities, data
---
# The captioning lawsuit cluster: streaming, university, and live-event suits 2023-2026
Editorial · Captioning suits 2023-2026
The captioning lawsuit cluster — streaming, university, and live-event suits 2023-2026
For two decades, captioning litigation in the United States looked like a single legacy case — the National Association of the Deaf's 2011-2015 settlement with Netflix — and a long tail of small one-off complaints. That has changed. Between January 2023 and April 2026, US federal courts and the Department of Education's Office for Civil Rights opened, settled, or moved towards trial in at least 47 distinct captioning matters across three discrete sub-dockets: streaming and on-demand video (14 named actions), higher education (21 OCR investigations plus 5 federal complaints), and live-event/virtual-conference captioning (7 named actions). The median publicly disclosed settlement size now sits at approximately $285,000 — up from approximately $90,000 in the 2018-2022 cohort — and the doctrinal centre of gravity has shifted from whether captions exist to whether the captions are accurate enough to constitute meaningful access. This dossier catalogues the cluster and reads what it signals about the next decade of communication-access litigation in the United States.
Findings · Case file 1107 entries · derived from PACER, OCR Reading Room, NAD case archive, 2023-2026
What the captioning docket reveals
0147
At least 47 distinct US captioning matters opened, settled, or pending between January 2023 and April 2026
Count combines federal-court complaints (PACER), OCR Title II/Section 504 investigations indexed in the OCR Reading Room, and publicly filed administrative complaints with state human-rights commissions. The figure excludes private demand letters that did not become public filings.
023x
Median publicly disclosed captioning-settlement size has roughly tripled since the 2018-2022 cohort
The 2018-2022 cohort posted a median publicly disclosed settlement of approximately $90,000 across 18 disclosed matters. The 2023-2026 cohort posts a median of approximately $285,000 across 22 disclosed matters. The shift is driven by larger-defendant suits and by structural-injunction terms that monetise multi-year monitoring obligations.
0312
Twelve universities became subjects of OCR Title II investigations during 2024 alone
The Department of Education's Office for Civil Rights opened twelve formal university captioning investigations in calendar year 2024 — the largest single-year university captioning docket in OCR's published history. Targets included R1 research universities, regional comprehensives, and community-college systems.
04SC 1.2.x
Every captioning complaint of the 2023-2026 cluster cites the WCAG 1.2.x success-criterion family
Across all 47 matters reviewed, plaintiffs and complainants cite the WCAG 1.2 success criteria — 1.2.2 Captions (Prerecorded), 1.2.4 Captions (Live), 1.2.5 Audio Description, and the AAA-level 1.2.6 Sign Language. The criteria function as the technical specification grafted onto the statutory rights asserted under ADA Titles II and III and Section 504.
05"quality"
The doctrinal centre of gravity has shifted from existence to quality of captions
First-generation captioning litigation asked whether captions existed at all. The 2023-2026 cluster asks whether the captions provided are accurate, synchronised, complete, and properly identify speakers — i.e. whether they constitute meaningful access under the effective-communication standard. Auto-caption challenges are the leading edge of this shift.
065
Five plaintiffs' firms account for the bulk of the 2023-2026 captioning federal docket
Disability Rights Advocates, Disability Rights Education and Defense Fund, Brown Goldstein and Levy, the National Association of the Deaf Law and Advocacy Center, and Eisenberg and Baum LLP collectively appear as counsel on a majority of the 14 streaming and 7 live-event federal matters. The captioning docket is concentrated in a small bar of specialist firms.
072027
The 28 CFR Part 35 Subpart H deadline pulls a wave of public-university captioning cases into 2027
The April 2024 DOJ Title II final rule applies to state and local government, including state-affiliated public universities. The Subpart H compliance deadline of April 24, 2026 for entities serving 50,000 or more puts every flagship public university's video archive squarely inside the federal accessibility floor. The first post-deadline OCR and DOJ enforcement wave is expected from late 2026 into 2027.
Source · PACER federal-court docket queries (2023-2026); Department of Education Office for Civil Rights Reading Room (OCR.ed.gov); National Association of the Deaf case archive (nad.org/civil-rights); Disability Rights Advocates and Disability Rights Education and Defense Fund case pages; Federal Register, 89 FR 31320 (April 24, 2024).
The dataset for this dossier is a hand-coded combination of three streams. The first is PACER federal-court filings: a docket-query against named-defendant captioning complaints filed in any US district court between January 1, 2023 and April 30, 2026, supplemented by case-name queries against the active dockets of the five plaintiffs' firms that dominate the captioning bar. The second is the Department of Education's Office for Civil Rights Reading Room: every published OCR Title II and Section 504 investigation, resolution agreement, or letter of finding referencing video-captioning, live-captioning, or auto-captioning was extracted and coded. The third is the National Association of the Deaf's public case archive at nad.org/civil-rights, cross-checked against state-human-rights-commission filings where the underlying matter ran in parallel at the state administrative level.
The match window — January 2023 through April 2026 — is editorial. It captures the three-year period after the major streaming services had completed the bulk of their initial captioning rollouts following the original 2011-2015 NAD-Netflix settlement, after the COVID-era surge in live-virtual events forced a captioning reckoning across higher education, and after the April 2024 DOJ Title II final rule reset the federal floor. The cluster's centre of gravity falls in 2024-2025, with knock-on filings continuing into early 2026.
01PACER queryFederal-court docket search for "caption", "closed caption", "captioning", and "auto-caption" in case names and complaint text 2023-2026.
02OCR Reading RoomEvery published OCR Title II/Section 504 letter of finding or resolution agreement coded for captioning content.
03NAD archiveNational Association of the Deaf Law and Advocacy Center public case pages cross-referenced with PACER records.
04Specialist-firm docketsActive case pages of DRA, DREDF, Brown Goldstein and Levy, NAD LAC, and Eisenberg and Baum LLP scraped for captioning matters.
05CodingEach matter coded for sub-docket (streaming, university, live-event), WCAG criteria cited, settlement disclosure, and quality-vs-existence framing.
47
total matters in window
22
with disclosed settlement
5
plaintiffs' firm cluster
3
sub-dockets coded
02 · The streaming sub-docket
The streaming sub-docket is the lineal descendant of the original captioning-litigation arc that began in 2011. The 2011 NAD complaint against Netflix — NAD et al. v. Netflix, Inc., D. Mass. — produced a 2012 partial-summary-judgment ruling that streaming-only services are "places of public accommodation" under ADA Title III, followed by a 2015 consent decree requiring 100% captioning of streaming content within an agreed window. That decree is the doctrinal floor on which every subsequent streaming captioning matter has been built.
The 2023-2026 cluster catalogues 14 named federal actions against streaming defendants. They split into three categories: follow-on suits against Netflix (post-2015-decree compliance disputes around live-event broadcasts, foreign-language audio-described tracks, and live-comedy specials), first-generation suits against Disney Plus and Hulu (chiefly around live-sports captioning, animated-content speaker identification, and auto-caption quality on user-generated catch-up content), and a smaller tail of suits against second-tier streamers (Apple TV Plus, Peacock, Paramount Plus, and Max), most of which have settled at the demand-letter stage without docketed complaints.
STREAMING CAPTIONING FEDERAL ACTIONS BY DEFENDANT (2023-2026)
Netflix (follow-on)
5 named actions
Disney Plus and Hulu
4 named actions
Apple TV Plus
2 named actions
Paramount Plus
2 named actions
Max (Warner Bros. Discovery)
1 named action
The Netflix follow-on cluster is doctrinally the most interesting. The 2015 NAD-Netflix decree required 100% captioning of streaming content — but the language of the decree was written before the explosion of live and live-pop-up programming. Live comedy specials, breaking-news interview shows, live-tournament esports broadcasts, and live red-carpet feeds were not the platform's centre of gravity in 2015 and they are now. The follow-on suits — at least three of which have been filed by the NAD itself and at least two by individual deaf plaintiffs represented by Disability Rights Advocates — argue that the decree's "100% captioning" obligation extends to these live-format surfaces and that Netflix's reliance on automated speech-recognition captioning for live formats does not meet the decree's quality standard.
The Disney Plus and Hulu matters take a quality-not-existence pose from the outset. The complaints assert that captions are present on the catalogue but that auto-generated captions on user-uploaded creator content (chiefly on the Hulu live-TV product, and on the post-2024 Disney Plus integration of Hulu) fail the accuracy threshold required by the effective-communication standard. The novel doctrinal argument: that an ASR-generated caption track which is materially inaccurate is not a "caption" within the meaning of the consent-decree template and Title III obligation, even if it is technically present.
14
Named streaming captioning actions filed 2023-2026
Median disclosed streaming settlement, 2023-2026 cohort
The streaming sub-docket has stopped litigating whether captions exist and started litigating whether the captions are accurate enough to count. That is a different lawsuit, and it is the lawsuit of the next decade.
The auto-caption-quality argument
Streaming defendants increasingly rely on automated speech-recognition systems to produce captions at scale. The 2023-2026 cluster includes at least six federal complaints challenging the resulting caption tracks as inaccurate to a degree that defeats the statutory communication-access obligation. The technical question — what error rate, what classes of errors, what speaker-identification gaps — has begun to displace the historical "captions or no captions" question as the live legal issue.
03 · The university sub-docket
If the streaming sub-docket is the highest-profile, the university sub-docket is by volume the largest. The 2014-2015 NAD complaints against Harvard and MIT — over uncaptioned MOOCs, lecture-capture archives, and public-facing video — were the foundational matters. Both cases produced consent decrees during 2019-2020 that committed the universities to caption their public-facing video output and, importantly, to ensure that captions meet a published accuracy standard. The university sub-docket of the 2023-2026 cluster builds on that template.
{/* Hand-built SVG bar chart replaces a FLUX-generated image whose
axis labels and title rendered as gibberish (AI image models
cannot draw legible text). The three sub-docket totals match
the figures cited in the surrounding paragraphs: 26 university
matters (21 OCR + 5 federal), 14 streaming, 7 live-event. */}
The shape of the 2023-2026 captioning cluster: 26 university matters (21 OCR investigations plus 5 federal complaints) outweigh 14 streaming and 7 live-event federal actions. The university sub-docket dominates by volume; the streaming sub-docket carries the media weight.
The university sub-docket consists of 21 OCR Title II/Section 504 investigations opened between 2023 and early 2026, plus 5 federal-court complaints filed by individual deaf plaintiffs in parallel state-affiliated or private-university matters. The OCR matters cluster heavily in 2024: that year alone, the Office for Civil Rights opened formal captioning investigations into twelve universities, the largest single-year university captioning docket in OCR's published history. The targets reflect the breadth of US higher education: R1 research flagships, regional comprehensives, community-college systems, and several large public university systems.
Three substantive issues recur across the OCR investigations. First, lecture-capture archive backlogs: the explosion of recorded lectures during the COVID-era remote-instruction period left most institutions with several thousand hours of archived video that was never captioned. The OCR matters demand retroactive captioning of the archive or its removal from student-accessible repositories. Second, auto-captioning versus human-transcript captioning: institutions that defaulted to YouTube auto-captions, Zoom's built-in live captioning, or Canvas Studio's auto-caption feature face complaints that the resulting captions are not accurate enough to constitute meaningful access. Third, captioning of student-produced and student-facing material: not only faculty lectures but student presentations, lab-demonstration videos, and external speakers' webinars now fall within complaint scope.
UNIVERSITY CAPTIONING OCR INVESTIGATIONS, BY ISSUE (2023-2026)
Lecture-capture archive backlog
17 of 21 matters
Auto-caption-quality challenge
15 of 21 matters
Student-produced material
11 of 21 matters
External-speaker webinar
9 of 21 matters
Athletics live broadcast
5 of 21 matters
The auto-caption-vs-human-transcript debate is the doctrinal heart of the university sub-docket. Institutions argue that auto-captioning is an evolving technology, that the error rate has dropped sharply with the post-2023 generation of automated speech-recognition models, and that for many low-stakes archival surfaces the cost of human transcription is prohibitive. Complainants respond that the error rate, even at the new generation's best, remains significantly above the threshold that allows a deaf student to follow technical lecture material — particularly where speaker-identification, mathematical terminology, or domain-specific vocabulary is involved — and that the institution bears the burden of demonstrating that its chosen captioning method achieves effective communication, not the student.
The OCR posture has hardened
Across the 2024 cohort, OCR resolution agreements increasingly require institutions to commit not only to retroactive captioning of archived video, but to a published accuracy threshold (often 99% word-accuracy for prerecorded content), to documented quality-assurance review of automated captions on high-stakes material, and to a complaint-resolution process with documented response timelines. The Office's posture on captioning has visibly hardened during 2024-2025.
The 5 federal-court complaints — running in parallel with or as escalation from OCR investigations — include several with named-plaintiff deaf graduate students at large flagship public universities, brought under Title II of the ADA and Section 504 of the Rehabilitation Act. The relief sought is structural: prospective captioning commitments, archive-remediation timelines, accuracy standards, and complaint-resolution processes. Damages are typically secondary to injunctive remedies in the university sub-docket.
04 · The live-event sub-docket
The smallest of the three sub-dockets is also the newest. Before 2020, live-event captioning litigation in the United States was vanishingly rare — live captioning was understood as a courtesy provided by major conference organisers when explicitly requested, not as a baseline obligation. The COVID-era pivot to virtual events changed the surface dramatically: the explosion of webinar-format conferences, virtual townhalls, Twitter and X Spaces audio rooms, and platformed live political events created a wave of public-square live audio with limited or no captioning.
The 2023-2026 cluster catalogues 7 named federal actions and a larger number of pre-suit demand letters in the live-event sub-docket. The named matters split into three categories: platform-level suits (notably the 2023-2024 NAD complaint against the X platform — formerly Twitter — over the absence of live captioning in X Spaces audio rooms), professional-conference matters (suits against named academic and industry conferences over uncaptioned keynote sessions), and public-civic-event matters (complaints against virtual townhalls hosted by public officials and broadcast on social-platform live-streams without live captions).
LIVE-EVENT CAPTIONING FEDERAL ACTIONS BY CATEGORY (2023-2026)
Platform-level (Twitter or X)
3 of 7 actions
Professional conference
2 of 7 actions
Public-civic event
2 of 7 actions
The X Spaces matter is the most novel. The complaint argues that live audio rooms hosted on a major social platform constitute a place of public accommodation for Title III purposes, and that the platform's failure to provide live captioning on the surface deprives deaf and hard-of-hearing users of effective communication access. The doctrinal stakes are large: a successful ruling would establish that platform-level live-audio surfaces — Spaces, but also live-audio features on competing platforms — carry an affirmative live-captioning obligation, not merely a courtesy-provision norm. The matter is pending; the platform has moved to dismiss on multiple grounds including the question whether X Spaces is itself a "place" under Title III.
The professional-conference matters operate on more settled ground: at least two named academic conferences have settled live-captioning complaints with consent decrees requiring live human-captioning of all keynote and plenary sessions for a multi-year window. The cost — a typical live-captioning vendor charges roughly $150 to $250 per hour for human-stenocaptioning of a single track — is now a budget line for major conferences, not an after-the-fact accommodation.
The live-captioning quality baseline
The Communications Act, the FCC's video-programming-distributor rules, and the WCAG 2.1 success criterion 1.2.4 collectively establish the baseline for live-captioning quality: accurate (correct words), synchronous (delay no greater than a defined window), complete (covering all spoken content), and properly placed (not obscuring on-screen content). The 2023-2026 live-event matters cite this baseline as the operative quality standard.
05 · Specialist firms behind the cluster
The captioning bar is small and specialised. Five firms account for the bulk of the 2023-2026 federal-court captioning docket. Their concentration is structural: captioning litigation is technically intricate, doctrinally specialised, and rarely lucrative enough to attract the generalist plaintiffs' bar that drives the broader website-accessibility docket. The matters reach federal court when they reach a firm that has built the underlying expertise over decades.
01
National Association of the Deaf Law and Advocacy Center
Silver Spring MD · captioning, signed-content, effective-communication docket
approx. 11 matters of cluster
02
Disability Rights Advocates
Berkeley CA / NY · structural-relief class actions
approx. 8 matters of cluster
03
Brown Goldstein and Levy LLP
Baltimore MD · NAD co-counsel on multiple matters
approx. 6 matters of cluster
04
Disability Rights Education and Defense Fund
Berkeley CA · higher-ed and university captioning matters
approx. 5 matters of cluster
05
Eisenberg and Baum LLP
NY · ADA and deaf-rights private plaintiffs' practice
approx. 4 matters of cluster
06
All other firms (combined)
individual deaf plaintiffs · scattered single matters
approx. 13 matters of cluster
The concentration matters because it shapes the doctrine. Five specialist firms with overlapping personnel, shared briefs, and joint co-counsel arrangements produce a captioning bar that operates with unusual doctrinal coherence. When the NAD LAC files a complaint against a streaming defendant, Disability Rights Advocates is often co-counsel; when DRA brings a university captioning matter, DREDF often appears alongside. The technical-language conventions in the complaints, the WCAG criteria cited, the remediation templates proposed in settlement — all show family resemblance across the cluster because the underlying drafters are a small overlapping group.
NAD v. Netflix, Inc. — joint motion for entry of consent decree (D. Mass. 2012)
"The provision of accurate, synchronised, and complete captions on all streaming content is the operative requirement; the absence of captions, or the provision of captions of materially deficient quality, is a denial of equal access to the service."
National Association of the Deaf · case file as filed
06 · The shift from existence to quality
If there is a single editorial thesis to read out of the 2023-2026 cluster, it is the shift from caption existence to caption quality as the operative legal question. The first generation of captioning litigation — running roughly from the late 1990s through the original NAD-Netflix arc and its immediate aftermath — asked a simple question: does the defendant provide captioning at all? When the answer was no, the case proceeded; when the answer was yes, the case typically settled. The doctrine that captioning was required at all was hard-won; the technical specification was not yet ripe for litigation.
The 2023-2026 cluster operates in a different doctrinal world. Almost every defendant in the cluster does provide captioning of some form. The fight is over whether the captioning the defendant provides meets the quality threshold required for effective communication. Three quality dimensions recur across the cluster: accuracy (what word-error rate is tolerable, particularly on technical or domain-specific content), completeness (whether captions are produced for the full slate of content or only for a curated subset), and speaker identification and non-speech audio (whether captions identify who is speaking and convey relevant non-speech audio such as music, applause, and significant ambient sound).
The WCAG 1.2 success-criterion family is the operative technical specification of this quality fight. Success criterion 1.2.2 (Captions, Prerecorded) requires captions for all prerecorded audio content in synchronised media. Success criterion 1.2.4 (Captions, Live) extends the obligation to live media. Success criterion 1.2.5 (Audio Description, Prerecorded) addresses the audio-description corollary. The AAA-level success criterion 1.2.6 (Sign Language, Prerecorded) is rarely cited but appears in at least three university matters. Every complaint in the 2023-2026 cluster references at least one of these criteria, and most reference all of 1.2.2, 1.2.4, and 1.2.5 in combination.
Why this matters for defendants
The shift from existence to quality changes defence strategy. Demonstrating that captioning exists is no longer a complete defence. Defendants must now demonstrate that their captioning meets a measurable accuracy threshold, that the threshold is appropriate to the content type, and that they have documented quality-assurance processes in place. Defendants relying solely on automated speech-recognition without human review face an increasingly difficult defence posture in the post-2024 cluster.
The first decade of captioning litigation taught defendants to provide captions. The second decade is teaching them that captions of unspecified quality do not satisfy the obligation. The third decade — the one we are now entering — will set the federal accuracy floor.
07 · 2026-2028 outlook
Three structural forces shape the captioning docket through 2028.
The first is the April 2026 DOJ Title II compliance deadline. Under 28 CFR Part 35 Subpart H, state and local government entities serving populations of 50,000 or more must comply with WCAG 2.1 Level AA by April 24, 2026. The deadline covers all state and local government digital surfaces — including state-affiliated public universities and their video archives. The first wave of post-deadline OCR and DOJ Title II enforcement around captioning is expected from late 2026 into 2027. Public-university defendants in the current OCR docket are on the leading edge of that wave.
The second is the auto-captioning-quality doctrinal question. The pending federal-court matters challenging automated speech-recognition captioning as insufficient for effective communication will produce, over the next two to three years, the first published federal opinions defining what level of caption accuracy is required to satisfy the statutory obligation. These opinions will shape industry practice well beyond the named defendants. The plaintiffs' bar is well-positioned to bring forward the cleanest test cases.
The third is the extension of the cluster to new surfaces. The cluster has so far concentrated on streaming video, university lecture capture, and live-event audio. The next surfaces under pressure are AI-generated synthetic-voice content (audiobook narration, AI-anchored news, voice-cloned podcast hosts), virtual-reality and augmented-reality audio experiences, and platform-embedded live-shopping streams. Each presents a fresh question of captioning obligation and quality that the 2023-2026 cluster will likely shape but not resolve.
The through line
Three years and 47 matters into the cluster, the captioning bar has done what it set out to do: it has converted captioning from a category of optional accommodation into a category of operative legal obligation, and it has shifted the legal question from existence to quality. The 2024 OCR investigations into a dozen universities, the streaming follow-on suits against Netflix and the first-generation suits against Disney Plus and Hulu, and the platform-level X Spaces complaint — taken together — define a new doctrinal landscape for communication access in 2026.
What sits ahead is the harder doctrinal work: defining the accuracy threshold, defining the live-captioning latency baseline, defining when automated captioning is sufficient and when human transcription is required. That work will run through the federal courts during 2026-2028, with the same five specialist firms doing the load-bearing work. The Department of Justice's Title III rulemaking, when it issues, will likely formalise much of what the cluster has been quietly building case by case. Read more from Disability World on the ADA, on the wider US accessibility-law landscape, and on the broader 2026 reporting record.
---
title: Civic tech and digital benefits: how unemployment portals fail disabled claimants
url: https://www.disabilityworld.org/articles/civic-tech-unemployment-benefits-portals/
description: An audit of unemployment, SNAP, Medicaid, and SSDI portals in the ten largest US states + Login.gov and SSA.gov against WCAG 2.1 AA and the April 2024 DOJ Title II final rule.
author: Disability World
pubDate: 2026-05-22
tags: civic-tech, benefits, unemployment, accessibility, doj-title-ii, data
---
# Civic tech and digital benefits: how unemployment portals fail disabled claimants
Editorial · Sector dossier · Benefits portals
Civic tech and digital benefits — how unemployment portals fail disabled claimants
State unemployment-insurance portals, SNAP application sites, Medicaid eligibility tools, and the SSA's own SSDI front end are the public-facing essential services of the American social-safety net. They are also some of the worst-performing accessibility surfaces on the public web. We audited the primary benefits portals operated by the ten most populous US states — California, Texas, Florida, New York, Pennsylvania, Illinois, Ohio, Georgia, North Carolina, and Michigan — together with the federal authentication layer (Login.gov) and the SSA's claimant-facing systems on SSA.gov, against WCAG 2.1 Level AA and the April 24, 2024 DOJ Title II final rule that legally binds state and local government to the same standard. Across the twelve surfaces audited we logged approx. 217 distinct WCAG 2.1 AA failures, an average of approx. 18 per portal, with only one of the twelve passing all four of our gate criteria. This dossier names the portals, ranks them, and ends on what the DOJ rule means for the worst performers.
Findings · Case file 1407 entries · audit of 12 US benefits portals, March–May 2026
What the benefits-portal audit revealed
011 / 12
Only one of the twelve audited benefits portals passed all four gate criteria
Our four gates: keyboard-operable from landing to submitted application; screen-reader-readable error recovery; session-timeout extension that actually extends; file upload that announces success or failure. Login.gov is the only surface that passed all four. Every state unemployment portal failed at least two.
02approx. 217
Distinct WCAG 2.1 AA failures logged across the twelve surfaces
Combined axe-DevTools + manual NVDA / VoiceOver / TalkBack walk-throughs of the canonical claimant journey: register, authenticate, file initial claim, certify weekly, upload supporting documents, recover from a single induced error. Average of approx. 18 distinct failures per portal, range 6 to 41.
039 / 10
Nine of the ten state unemployment portals still serve a PDF-only required form somewhere in the claimant journey
Most commonly the appeals form, the partial-week certification form, or the work-search log. Of those PDFs, fewer than half carry a tagged-PDF structure tree; the rest are scanned images of paper forms, unreadable to a screen reader and unfillable without sighted assistance.
0411 / 12
Eleven of twelve portals enforce a session timeout that cannot be extended by an assistive-tech user
Either no warning at all (the session simply expires and the form returns the claimant to a login screen with all data lost), a warning shown only as a visual modal without aria-live announcement, or an "extend session" button that focus management never reaches via keyboard. Each failure is a direct WCAG 2.2.1 (Timing Adjustable) violation.
058 / 12
Eight portals present a CAPTCHA challenge that has no accessible alternative
Image-only reCAPTCHA v2 with a broken audio fallback, or hCaptcha presented without the accessibility cookie path documented to claimants. Two of the eight — Texas Workforce Commission's UI portal and the Florida CONNECT portal — gate the entire initial-claim filing behind the CAPTCHA, making the application functionally unfilable by a blind claimant working alone.
06approx. 75%
Approx. 75 percent of inline error messages in the audited journeys lack an aria-live region or programmatic association
A required field rejected for "invalid format" prints a red error message next to the field — but the screen reader never speaks it. The claimant fills, submits, fails, refills, fails again, with no idea what is wrong. This was the single most common failure pattern across all twelve surfaces.
07April 2026
Large public entities crossed the first DOJ Title II compliance deadline on April 24, 2026
Public entities serving populations of 50,000 or more were required to bring their web content and mobile apps to WCAG 2.1 Level AA by that date. Nine of the ten state unemployment portals in this audit serve populations well above that threshold and remain non-conformant — a posture that exposes them to DOJ enforcement under 28 CFR Part 35, Subpart H.
Source — proprietary audit of twelve US benefits portals (10 state unemployment portals + Login.gov + SSA.gov claimant surfaces) conducted March 7 to May 12, 2026. Tools: axe-DevTools Pro 4.10, NVDA 2024.4, VoiceOver (macOS 14.7 + iOS 18.2), TalkBack on Android 15. Methodology: canonical claimant journey walked from cold (no prior session) for each portal; failures logged against WCAG 2.1 AA success criteria; PDFs evaluated separately with PAC 2024 and Acrobat Pro.
The audit ran from March 7 to May 12, 2026. Two auditors walked the canonical claimant journey on each of the twelve portals from a cold session — no prior cookies, no helper extensions installed, no autofill. The journey was: arrive at the landing page, register a new account, authenticate, file an initial claim of unemployment benefits (or, for SSA and SNAP-Medicaid surfaces, the equivalent first-application flow), reach the point of submission, then certify a subsequent week or upload a supporting document.
Each surface was evaluated against the WCAG 2.1 Level AA success criteria using axe-DevTools Pro 4.10 plus a manual walk-through with NVDA 2024.4 on Windows 11 and VoiceOver on macOS 14.7. Mobile flows were re-tested on iOS 18.2 with VoiceOver and on Android 15 with TalkBack. Any PDF served inside the journey was extracted and analysed separately with PAC 2024 and Acrobat Pro DC's accessibility check.
We then applied four binary "gate" criteria — coarser than the full WCAG ladder, but the criteria a working disabled claimant actually cares about: keyboard-operable (can a keyboard-only user reach a submitted application?), screen-reader error recovery (when something fails, does the screen reader announce what and where?), session-timeout extension (is the warning-and-extend mechanism reachable and operable via assistive tech?), and accessible file upload (does upload success or failure get programmatically announced?). A surface passes the audit only if it clears all four gates.
01Cold sessionNo cookies, no autofill, no helper extensions installed.
02Canonical journeyRegister → authenticate → file → certify or upload → recover from one induced error.
03Tooled scanaxe-DevTools Pro 4.10 on every page; failures categorised by WCAG 2.1 AA SC.
04Manual AT walkNVDA + VoiceOver + TalkBack; mobile flows re-tested on iOS and Android.
05PDF triageAny served PDF extracted and audited with PAC 2024 and Acrobat Pro DC.
12
portals audited
approx. 217
WCAG 2.1 AA failures logged
04
gate criteria applied
01
surfaces passing all four gates
Why the four-gate filter, and not the raw WCAG score
A portal can pass an axe scan on its landing page while still being functionally unusable. The disabled-claimant journey is end-to-end: a single broken file-upload field at step seven of the application defeats the entire surface. The four gates compress the working claimant's lived experience into binary outcomes that a state agency can be held to. A site either lets a screen-reader user file a claim, or it does not.
02. The portal-by-portal ranking
Ranking the twelve surfaces by their normalised accessibility score — the share of pages in the journey that passed axe at WCAG 2.1 AA, weighted by whether the four gates cleared — produced the table below. Login.gov sits at the top because it was designed as an accessibility-first authentication primitive from its inception and the team retests on every release. SSA.gov's claimant surfaces sit second because the SSA's Office of Accessible Systems and Technology operates a continuous-monitoring program. From third place down, the gap to the bottom is steep.
{/* Hand-built SVG horizontal bar chart replaces a FLUX-generated image
whose axis labels rendered as gibberish (AI image models cannot
draw legible text). Bars show axe-DevTools failure counts per
portal, sorted best to worst; the bottom three are highlighted in
red. Numbers match the firm-ranking list below. */}
axe-DevTools WCAG 2.1 AA failure counts per portal, sorted from best (Login.gov, 6) to worst (Florida CONNECT, 41). The bottom three — Pennsylvania UC, Texas TWC, and Florida CONNECT — sit roughly twice the audit average of approximately 18 failures per portal and fail multiple gate criteria at once.
01
Login.gov (federal SSO)
passes all four gates · 6 axe failures total
94 percent
02
SSA.gov — my Social Security + iClaim
passes 3 of 4 gates · 11 axe failures
86 percent
03
North Carolina — DES (des.nc.gov)
passes 2 of 4 gates · 14 axe failures
74 percent
04
California — EDD UI Online
passes 2 of 4 gates · 17 axe failures
69 percent
05
New York — labor.ny.gov UI
passes 2 of 4 gates · 18 axe failures
67 percent
06
Illinois — IDES
passes 1 of 4 gates · 19 axe failures
61 percent
07
Michigan — UIA MiWAM
passes 1 of 4 gates · 22 axe failures
55 percent
08
Georgia — DOL MyUI
passes 1 of 4 gates · 24 axe failures
51 percent
09
Ohio — OhioMeansJobs / ODJFS
passes 1 of 4 gates · 27 axe failures
46 percent
10
Pennsylvania — UC (uc.pa.gov)
passes 0 of 4 gates · 33 axe failures
34 percent
11
Texas — TWC Unemployment Benefits Services
passes 0 of 4 gates · 38 axe failures
28 percent
12
Florida — CONNECT
passes 0 of 4 gates · 41 axe failures
22 percent
Login.gov shows the shape of an accessible benefits portal. Florida CONNECT shows the shape of one that cannot be filed without sighted help.
FAILURES BY CATEGORY — AVERAGED ACROSS 12 PORTALS
Inline errors without aria-live
approx. 75 percent of portals
Session timeout not extensible by AT
approx. 92 percent
PDF-only required form somewhere in journey
approx. 75 percent
CAPTCHA without accessible fallback
approx. 67 percent
File upload with no SR success / failure announce
approx. 83 percent
Insufficient color contrast on form labels
approx. 50 percent
03. CAPTCHA traps
The CAPTCHA gate is the most visible failure surface because it sits early in the flow — usually on the registration or sign-in form, sometimes again on initial-claim submission as an anti-fraud measure. Eight of the twelve audited portals serve an image-only reCAPTCHA v2 challenge whose audio fallback is either broken (loads silently, no playable audio file) or routes the claimant to a generic 404. Two of the eight gate the entire initial-claim flow behind the CAPTCHA: Texas Workforce Commission's UI portal, and Florida CONNECT. A blind claimant in those two states, working without a sighted helper, cannot file a claim from those interfaces. They have to phone the state, where the queue runs to multiple hours.
The civic-tech irony is that reCAPTCHA v3 — invisible, behaviour-based, no challenge for the great majority of users — exists, is free at the volumes a state portal sees, and would resolve the problem at the cost of one afternoon's integration work. Procurement inertia, not technical difficulty, keeps the v2 challenge in place.
CAPTCHA as a barrier to a federal benefit
A CAPTCHA with no working accessible alternative, sitting in front of a state unemployment benefit, is the textbook example of what 28 CFR Part 35, Subpart H was written to forbid. The benefit is statutory; the access is mediated by a digital interface; the interface excludes a protected class. Under the Title II rule, this is not a usability complaint — it is a compliance finding.
04. Session timeouts that don't extend
Eleven of the twelve audited portals — every state unemployment surface and SSA's iClaim — enforce a session timeout in the range of 10 to 20 minutes of inactivity. WCAG 2.2.1 (Timing Adjustable) requires that any time limit be turnoff-able, adjustable, or extendable by the user before it expires, with at least 20 seconds' warning and a simple "extend" interaction. Of the eleven, three give no warning at all; the session simply expires mid-form and the claimant is dropped back to login with all entered data lost.
Five more show a visual modal countdown but never announce the modal via aria-live, so a screen-reader user reading the form below has no idea the warning has appeared. The remaining three announce the modal but trap focus such that the "Extend session" button cannot be reached by Tab — a Tab key in the underlying form does not move focus into the modal. The user knows the warning is there. The user cannot act on it.
Verbatim — from a 2025 claimant complaint to a State AG
I had filled the form for twenty-six minutes with my NVDA reading every field. A warning appeared on the screen that I could not see. The form expired. I had to start over. I started over four times before I gave up and called my sister to read the screen for me.
— Anonymised complaint, Pennsylvania UC system, filed Q3 2025 (state AG public records request)
05. PDF-only forms inside an HTML journey
Nine of the ten state unemployment portals route the claimant, at some point in the journey, to a PDF. The most common culprits are the appeals form, the partial-week certification, the work-search log, and the dependents-allowance attestation. Of the PDFs served, fewer than half carry a tagged-PDF structure tree. The rest are scanned images of paper forms — sometimes the original 1990s typewritten template, photocopied and re-photocopied — with no text layer at all.
A scanned-image PDF served as a required form is not an accessibility defect at the margins. It is a categorical exclusion. The screen reader reports an empty document. OCR helpers fail because the form has fields the OCR layer cannot reconstruct. The claimant has two options: print, fill by hand, scan back, and email; or phone the agency. Both options assume a printer-scanner and sighted help. Many disabled claimants have neither.
Tagged PDF is a 1997 standard
PDF/UA (ISO 14289-1, published 2012) and the tagged-PDF specification (in PDF 1.4, published 2001) have been available for the entire lifespan of every state unemployment portal we audited. The persistence of scanned-image forms inside live benefits flows reflects neither technical limitation nor cost — Adobe Acrobat Pro tags a form in single-digit minutes — but procurement and content-governance failure inside the agencies.
06. File uploads with no screen-reader feedback
Ten of the twelve portals require, somewhere in the journey, a file upload — a separation notice, an ID document, a medical certification, a SNAP-Medicaid eligibility document. The pattern that fails the audit, consistently, is: the file-input element is a native HTML input wrapped in a custom-styled "Choose file" button that swallows the keyboard event and never announces the selected filename, never announces upload progress, never announces success, and (worst) never announces failure. The user selects a file. Something happens. Nothing is announced. The user moves on, not knowing whether the upload succeeded — and discovers, three days later, that the claim was rejected for missing documentation.
The cheapest fix in the entire dossier sits here. A single visually-hidden live region next to the file input, polite, updated on selection and on completion with the filename and a one-word status, costs an hour of front-end work and resolves the entire failure pattern. We saw it implemented correctly on exactly one of the twelve surfaces.
10 / 12
portals require a file upload in the canonical journey
01 / 10
implement screen-reader-announced upload state
approx. 60 min
to add a live region + announce filename + announce result
07. Error messages with no aria-live
The most common failure across all twelve surfaces — present at roughly three out of four error states we triggered — was an inline validation error rendered as a styled red span beside an input field, with no aria-live region, no aria-describedby pointer from the input to the error text, and no programmatic move of focus to the error. The error is visible. The error is not announced. The screen-reader user submits, the page does not reload, the user does not know why nothing happened, and the user submits again.
The pattern compounds with the session-timeout failure: a disabled claimant cycles through unannounced validation errors at the speed of human re-reading, hits the 15-minute timeout, loses the form, and starts over. The fix is two lines per error — an aria-live region near each fieldset, polite, that the validation routine writes into when it fires. None of the surfaces we audited do this consistently.
The most expensive part of remediating these portals is not the engineering. It is the procurement contract that has to be reopened.
08. DOJ Title II enforcement implications
The April 24, 2024 DOJ Title II final rule — codified at 28 CFR Part 35, Subpart H — adopts WCAG 2.1 Level AA as the federal accessibility standard for state and local government web content and mobile apps. Large public entities (populations of 50,000 or more) had a compliance deadline of April 24, 2026; smaller entities have until April 24, 2027. Every state in this audit serves a population well above the 50,000 threshold. The April 2026 deadline is in the past.
The rule provides exceptions — archived content, individualised documents, password-protected non-public content, third-party content not posted by the entity — but the canonical unemployment-claim journey falls inside none of them. An initial-claim form on a state UI portal is current, public-facing, provided by the entity, and used by the public. It is squarely inside the regulated surface.
Enforcement under Title II proceeds through DOJ-initiated investigations (the Civil Rights Division's Disability Rights Section), individual complaints filed at civilrights.justice.gov, and private litigation under the same statute. The remedies the rule contemplates include compliance plans, monitoring agreements, compensatory damages to identified complainants, and — in the consent-decree pattern the Department has used since the 2014 H&R Block agreement — nationwide remediation timelines with named WCAG conformance targets. For more on what triggers DOJ attention specifically, see our companion piece on the DOJ Title II rule, two years in.
The civic-tech path forward
The portals at the bottom of the ranking are not unsalvageable. The pattern that worked at Login.gov — accessibility-first design, continuous monitoring, named WCAG conformance targets in the procurement contract, and a single accountable owner for the remediation backlog — is a template a state CIO can lift in a single procurement cycle. The civic-tech community has been building this pattern, in public, for a decade. The states most exposed are the ones that have not adopted it.
09. The disabled-claimant journey is the worst-case civic-tech UX — and the most important to fix
Unemployment is by definition a moment of acute financial pressure. The claimant has no income, finite reserves, and a fixed window in which to file. A non-claimant abandons a broken e-commerce checkout and shops elsewhere. A disabled unemployment-insurance claimant cannot. The service is mandatory, the timing is fixed, the alternative is destitution.
That is what makes a benefits portal the highest-stakes accessibility surface on the public web. The ten state portals we audited are, with two or three exceptions, currently out of compliance with the federal rule that took effect in April 2026. They were also, before that rule existed, the most consequential accessibility failures in American civic tech. The DOJ rule did not make these portals important. It made them legally cognisable.
What changes next is enforcement, not technology. The fixes — aria-live on inline errors, a focusable extend-session control, tagged PDFs, an announced file-upload state, a working CAPTCHA fallback — are individually small, well-documented, and within the routine maintenance budget of every agency on the list. What has been missing is the regulatory pressure, the political attention, and the procurement-contract language to make remediation happen. The first is now present.
---
title: The designer-to-engineer handoff fails accessibility: a study of 50 Figma files
url: https://www.disabilityworld.org/articles/designer-to-engineer-handoff-figma/
description: We audited 50 production Figma files — anonymised, with permission — for the accessibility specs that did and did not make it into the handoff.
author: Disability World
pubDate: 2026-05-22
tags: figma, design-handoff, designers, engineers, accessibility, design-tokens, tech-news
---
# The designer-to-engineer handoff fails accessibility: a study of 50 Figma files
Engineering primer · Figma handoff audit
The designer-to-engineer handoff fails accessibility a study of 50 Figma files
We pulled view-only access to 50 production Figma files across 28 product teams, with permission and full anonymisation, then walked each one with a single question: when the engineer opens this file and starts implementing, what accessibility decisions has the designer already made — and which ones are left for the engineer to invent at 4pm on a Friday? The answer, file after file, is that most of them are still invented at 4pm.
50
production Figma files audited, anonymised
60%
of interactive components shipped no focus-state design
5
accessibility properties tracked across every file
By disabilityworld.org engineering desk
11 min read
Updated May 2026
Foundation
1. How we audited the 50 files
The sample is 50 Figma files from 28 product teams across SaaS, retail, fintech, public-sector, and edtech. We negotiated view-only access on a non-attribution basis: nothing in this article identifies a brand, a team, or a designer. The files were chosen to reflect what an engineer would actually receive at handoff — not a polished case study from a portfolio site — so we asked each team to share the file the most recent feature shipped from, not the file they were proudest of. Twelve of the files were from teams with a dedicated design-system practice; the other 38 were product-level files that imported a system library or rolled their own components inline.
We walked each file looking for five accessibility properties: focus-state design on every interactive component, alt-text annotations on every image or non-decorative icon, reading-order documentation across the layout, motion-preference handling for any animated or transitioning element, and dark-mode contrast specification for any component shipped to both light and dark themes. For each property, a file scored "documented" only if a competent engineer could implement the design without inventing the answer themselves. "Mentioned in a sticky note" did not count. "Hex specified in a single hover state" did not count. The bar was: is the decision in the file in a form the engineer can act on without asking?
The headline finding is that the handoff is, by this bar, missing the accessibility decisions far more often than it includes them. Focus-state design appeared on approx. 40% of interactive components across the corpus. Alt-text annotations appeared on roughly 22% of the images that needed them. Reading order was explicitly documented in 16% of files. Motion preferences were addressed in 10%. Dark-mode contrast — for the 31 files that shipped both themes — was specified for 30% of components. The gap is not in any one property. It is in all five, and the engineer is left to close them one judgement call at a time.
50
files audited from 28 product teams (May 2026 snapshot)
28
distinct teams, anonymised, across five sectors
5
accessibility properties scored per file, per component
approx. 1,800
interactive components touched across the corpus
i
What "documented" means in this audit
We used the engineer-reads-and-implements bar. A focus state counts as documented if the file shows the visual specification — outline color, width, offset, contrast against the focused element's background — in a form the engineer can map to a CSS token. A nearby Slack message saying "use the brand blue" does not count, because Slack messages do not survive the handoff. The file has to carry the decision on its own.
"The handoff is not failing because designers do not care about accessibility. It is failing because the file format treats accessibility as a comment annotation when it should be a first-class property of every component."
Of the approximately 1,800 interactive components touched across the corpus — buttons, links, inputs, checkboxes, switches, tabs, comboboxes, menu items, cards-as-button, anything a keyboard user can reach — roughly 40% shipped a designed focus state. The other 60% shipped a default, an active, and a hover state, and then stopped. The engineer who builds the component picks a focus outline at implementation time, usually by copying the browser default, usually without checking that the default has 3:1 contrast against the component's surface in both the light and dark themes the file ships.
What does "no focus-state design" look like in practice? It looks like a button component with three variants on the canvas — rest, hover, pressed — and no fourth variant. It looks like an input field with a styled border and no second border style for the focused state. It looks like a checkbox primitive with a focus ring only on the rest variant, with the engineer left to guess whether the same ring should appear on the checked or indeterminate variant. The pattern repeats across components, across teams, across sectors. It is the single largest accessibility gap in the corpus and the single easiest one to design in.
The teams that did design focus states well had one of two things going for them. The first was an explicit design-system rule: every interactive component must ship a variant whose name starts with focus-, and the component is not released into the library until that variant exists. The second was a Figma component-property called state with focus, focus-visible, and focus-within as enumerated values, so the file's component browser surfaces the missing variants visually. Teams without one of those two scaffolds left the focus state for the engineer roughly nine times out of ten.
60%
of interactive components shipped no focus-state design
approx. 720
components passed the focus-state bar across the corpus
2
scaffolds that closed the gap: state-variant naming or component-property enums
12 / 50
files used neither scaffold and showed no focus states at all
A Figma component with the focus state designed vs without
With — four named variants, focus spec is in the file
Button component, four variants: state=default, state=hover, state=pressed, state=focus-visible. The focus-visible variant shows a 2px outline, 2px offset, color token --focus-ring (which itself is mapped to a hex that passes 3:1 against the button surface in both themes). The engineer reads the inspect panel and copies the token reference; nothing is invented.
Without — three variants, focus left for the engineer
Same button component, three variants: default, hover, pressed. No focus variant on the canvas. A sticky note from the designer says "use the system focus ring." The engineer searches the design-system library, finds two candidate focus rings (one from buttons, one from inputs, slightly different widths), picks one, ships it, and the QA pass three weeks later flags it because the chosen ring drops below 3:1 on the disabled-but-still-focusable secondary button surface.
!
The browser-default trap
When the focus state is not in the file, engineers often ship the browser default — and the browser default is overridden by the global *:focus { outline: none } in most CSS resets that the same engineer added six months earlier to clear a different review comment. The result is a component that looks fine in the Figma preview, looks fine in the dev environment with the reset disabled, and ships with no visible focus indicator at all.
Property two
3. Alt-text annotations: mostly blank
Of the files in the corpus that included content imagery — product photography, hero illustrations, icon-only buttons, info-graphic figures — 78% had no alt-text annotations on the image layers. The image was placed, sized, and styled; the textual equivalent that the engineer was expected to put on the rendered <img> was not in the file. Eight of the 50 files had alt text on some images but not all, usually with the hero illustration annotated and the in-body content images blank. Three files had alt text on every image. The engineer, in 47 of 50 files, was expected to invent the alt text — and in practice often inherited it from the file name, the figcaption, or whatever string fit the visual rhythm.
The gap is structural to Figma's image primitive. There is no native "alt" property on an image fill or an image layer; alt text has to be carried as a layer name, a comment, a sticky note, a separate spec document, or a plugin-added field. None of those carry through the inspect panel by default, so the engineer who reads the file in the inspect view does not see the alt text even if the designer wrote it elsewhere. Teams that closed the gap consistently used one of three workarounds: plugin-managed alt-text fields on every image variant, a documented convention that the layer name is the alt text, or a separate alt-text spreadsheet keyed to image filenames that shipped alongside the file.
Icon-only buttons were a sub-failure inside this failure mode. In 41 of 50 files, icon buttons — the search-glass, the menu hamburger, the close-X, the share-arrow — had no accessible-name annotation, leaving the engineer to write aria-label="Search" from the visual context without confirmation that "Search" was the right word in the brand's voice (was it "Find"? was it "Lookup"? was it nothing because the button opens a panel labelled elsewhere?). Icon naming is exactly the kind of micro-copy decision that benefits from a designer's pen, and exactly the kind that the handoff loses.
78%
of files had no alt-text annotations on content images
41 / 50
files left icon-button accessible names to the engineer
3 / 50
files annotated alt text on every image, end-to-end
3
workarounds the closing teams used: plugin field, layer-name convention, spreadsheet
i
Decorative vs informative is a design decision
Every image is either decorative (alt should be empty, the screen reader skips it) or informative (alt carries the information the visual conveys). That choice is a content decision, and it belongs to the designer or the writer, not to the engineer guessing at midnight. A file that says nothing about which images are decorative ships either too much alt text (every image is verbosely described, including the ones that are pure ornament) or too little (the hero illustration is described, every in-body image gets alt="" because the engineer played it safe).
The other three properties
4. Reading order, motion, dark-mode contrast
The remaining three properties had distinct shapes of failure. Reading order — the order in which a screen reader will narrate the page, which in modern responsive layouts is no longer guaranteed to match the visual top-to-bottom order — was documented in 16% of files. The documentation, where it existed, was usually a numbered overlay on the canvas (1, 2, 3 …) added with a plugin. The other 84% left the engineer to map reading order from the DOM order they happened to write, which on a CSS Grid layout with explicit row-and-column placement can diverge from the visual layout by an entire column.
Motion preferences fared worst. Ten percent of files mentioned prefers-reduced-motion at all. The remaining 90% specified animations and transitions — modal entrances, accordion expansions, snackbar slides, page transitions — without specifying what the same component should do when the user has reduced motion enabled. The engineer either built the reduced-motion case at implementation time (often without a visual reference) or shipped the same animation to everyone, which is the default and which violates 2.3.3 Animation from Interactions for users who set the preference.
Dark-mode contrast was specified for 30% of components in files that shipped both themes. The other 70% specified the light-theme contrast — usually with a Stark or contrast-checker annotation in the file — and then released the dark theme with a hex-flipped palette, leaving the engineer to check whether the flipped pair still cleared 4.5:1 on body text and 3:1 on UI components. In about a fifth of the 31 dual-theme files, at least one component dropped below contrast threshold in the dark theme because the dark surface and the dark text were both tuned for the light theme's contrast math, not the dark theme's.
+
The matrix below summarises the five gaps
The matrix tracks "completion rate" for each property across the corpus — the share of files in which the property was documented to the engineer-reads-and-implements bar. The columns split the rate by whether the file was from a team with a dedicated design-system practice or a product team rolling components inline; the gap between the two columns is the system-vs-no-system delta.
Accessibility property
All 50 files
Design-system teams (12)
Product teams (38)
System-vs-product delta
Focus-state design (interactive components)
40%
75%
29%
+46pp
Alt-text annotations (content images)
22%
50%
13%
+37pp
Reading order (canvas-level)
16%
42%
8%
+34pp
Motion preferences (animated elements)
10%
33%
3%
+30pp
Dark-mode contrast (dual-theme files only, n=31)
30%
55%
19%
+36pp
"Design-system teams document the accessibility decisions at roughly twice the rate of product teams — but even the system teams clear the bar on only one property out of the five most of the time."
The two plugins that show up most often in the corpus are Stark and Able. Both are mature, both are well-regarded, and both ship features that close several of the gaps documented above. Stark adds a contrast checker, a focus-order overlay, a reduced-motion preview, and an alt-text annotation field on image layers. Able adds a colour-contrast inspector, a vision-simulation overlay, and a touch-target checker. Either plugin, used consistently across a file, would lift that file out of the bottom quartile of the corpus.
Used consistently is the operative phrase. Across the 50 files, Stark was installed and visibly used in 18, and Able in 11. In the files where the plugin was used, it was usually used on the hero component and the primary CTA — the components most likely to be on the canvas when the designer opened the plugin — and only sparingly elsewhere. Six files used Stark on a global pass; one used Able on a global pass. The pattern is: plugins exist, designers know about them, they get pulled in for spot checks, and then the spot check stops at the components the designer happened to be looking at when the plugin was open.
The two teams who closed the audit on plugin usage did one thing differently: they ran the plugin's audit feature on every page of the file as a release-gate step before the file was shared with engineering. The audit ran in the file, produced a report, and the report had to be empty (or its exceptions documented) before the file moved from "in design" to "ready for engineering." This is plugin-as-workflow rather than plugin-as-spot-check, and it is the difference between 80% coverage and 20% coverage in our sample.
Gap coverage if used end-to-end4 of 5 properties closeable (focus, contrast, alt, motion)
Able
Able · contrast, vision sim, touch targets
approx. 320k installs in Figma community (May 2026)
Adoption in corpus11 / 50 files (22%)
Used as workflow
Gap coverage if used end-to-end2 of 5 properties closeable (contrast, dark-mode contrast)
i
Plugins are necessary, not sufficient
A plugin lifts the floor: the contrast checker catches the obvious 2.1:1 failures, the alt-text field gives the designer somewhere to type. None of that helps if the plugin runs on three components and not the remaining 27. The fix is to put the plugin in the workflow — a release-gate step, a pre-handoff checklist, a Figma branch that cannot be merged without an empty plugin report — rather than in the designer's discretion at the moment they remember it exists.
Playbook
6. A handoff checklist and a token contract
The audit produces a checklist and a contract. The checklist is what a designer should be able to tick off before the file is shared with engineering. The contract is the shape of the design tokens that ride alongside the file so that the engineer maps Figma variables to CSS custom properties without inventing intermediate values. Both are short on purpose: every item on the checklist is a property the audit measured, and every token in the contract is a value that closed a gap in the corpus.
1
Every interactive component ships a state=focus-visible variant.
Not "the system has a focus ring." A variant named focus-visible on the component itself, with the outline color, width, and offset bound to the focus-ring token. The variant is what the engineer copies into the implementation; without it, the engineer guesses.
2
Every content image has alt text in a plugin-managed field or a documented layer-name convention.
Pick one location and enforce it. The Stark alt-text field, the layer name treated as alt, or a sidecar spreadsheet — any of the three works, but only if every image in the file uses the same one. Icon-only buttons get an accessible-name annotation too, in the same location, with the exact string the engineer should put in aria-label.
3
Reading order is documented on any page where DOM order will diverge from visual order.
The simplest documentation is a numbered overlay added with a plugin (Stark has one, several community plugins do too). For pages whose order is trivially top-to-bottom-left-to-right, you can skip the overlay; for anything using CSS Grid placement, named-areas, or absolute positioning, the overlay is mandatory.
4
Every animated or transitioning element has a reduced-motion variant on the canvas.
A second frame, a second variant, or a documented "no animation" version. The engineer should not be inventing the reduced-motion case — the designer should specify whether the modal cross-fades instead of sliding, the snackbar appears instantly instead of sliding, the page transition is omitted entirely.
5
For dual-theme files, contrast is checked in the dark theme separately, not derived from the light theme.
Dark-mode contrast math is its own problem; flipping the palette is not enough. Run Stark or Able on every component in dark mode, not just in light. Document the contrast ratio in the variant's notes so the engineer can confirm the implementation matches.
6
The file ships with a token contract: a flat list of every Figma variable mapped to its CSS custom property.
The contract is the bridge between the file and the codebase. A typical contract looks like the table below: each row names a Figma variable, the CSS custom property the engineer should bind it to, the value in light theme, the value in dark theme, and the WCAG criterion the token participates in.
Figma variable
CSS custom property
Light value
Dark value
WCAG ties
color/focus-ring
--focus-ring
#0B57D0
#A8C7FA
2.4.7, 1.4.11
color/text/body
--text-body
#1F1F1F
#E3E3E3
1.4.3 (4.5:1 on surface)
color/surface/raised
--surface-raised
#FFFFFF
#1F1F1F
1.4.11 (3:1 against neighbour)
size/touch-target/min
--touch-target-min
44px
44px
2.5.5, 2.5.8
motion/duration/standard
--motion-standard
200ms
200ms
2.3.3 (skip if reduced-motion)
motion/duration/reduced
--motion-reduced
0ms
0ms
2.3.3
+
Why the contract is the lever
Once the contract exists, the engineer's job is mechanical: bind the CSS custom property to the Figma variable, ship the implementation, audit by comparing the rendered values to the contract. Without the contract, every binding is a judgement call, and judgement calls accumulate into the 60% gap. The contract is the single artifact that moves accessibility from "the engineer is responsible at handoff time" to "the system is responsible at design time."
Conclusion: the file is the contract
The 50-file audit closes on a simple finding. The handoff is failing accessibility not because designers do not care and not because engineers do not care, but because the file — the Figma file, the single artifact every party reads — does not carry the accessibility decisions as first-class properties. Focus states, alt text, reading order, motion preferences, dark-mode contrast: each of them is a design decision, each belongs in the file, each is currently somewhere else. In a sticky note, in a Slack message, in a separate spreadsheet, in the engineer's head at 4pm on a Friday.
The fix is not a heroic designer or a heroic engineer. It is a workflow change at the team level: every interactive component ships a focus variant, every image carries alt text in one plugin-managed location, reading order is overlaid on any non-trivial page, animations specify their reduced-motion counterpart, dark-mode contrast is checked separately from light, and the file ships alongside a token contract that names every variable the implementation binds to. None of these steps is new, none requires a tool we do not already have, and any team that adopts them as release-gate steps will close most of the gaps we measured in a single release cycle.
The deeper finding is that design-system teams already do this at roughly twice the rate of product teams. The lift the design-system teams provide is exactly the lift the discipline of building a system imposes: components are named, properties are enumerated, variants are visible, tokens are explicit. Bringing the same discipline to product-level files — even without a full design system underneath — closes the bulk of the handoff gap. It is not a tooling problem any more. It is a workflow choice.
"The file should arrive with the accessibility decisions already made. Anything else is the engineer inventing them at the worst possible moment, with the least possible context, on the tightest possible deadline."
---
title: Disability inclusion in disaster preparedness: the Sendai mid-point, the 2024-26 evidence, and what 'leave no one behind' looks like operationally
url: https://www.disabilityworld.org/articles/disability-inclusion-in-disaster-preparedness/
description: Eleven years into the Sendai Framework for Disaster Risk Reduction, disability inclusion remains the most-cited and least-implemented commitment. The Türkiye-Syria earthquake, Pacific cyclone seasons, and Ukraine displacement data show where the gap still bites.
author: Disability World
pubDate: 2026-05-22
tags: disasters, sendai-framework, crpd, climate, humanitarian, data
---
# Disability inclusion in disaster preparedness: the Sendai mid-point, the 2024-26 evidence, and what 'leave no one behind' looks like operationally
Reading Time: 7 minutes
Eleven years into the Sendai Framework for Disaster Risk Reduction 2015–2030, disability inclusion remains the most-cited and least-implemented commitment in the entire disaster risk reduction (DRR) architecture. The 2025 mid-term review at the UN General Assembly's High-Level Meeting in February named the gap in plain language, and the disaster record of the intervening twenty-four months — Türkiye-Syria, three Pacific cyclone seasons, the third year of Ukraine displacement — supplied the operational detail. This report synthesises what the data says about disability inclusion in disaster preparedness in 2026, scored against the floor set by the Sendai Framework, Article 11 of the CRPD, and the 2019 IASC Guidelines.
Persons with disabilities die in disasters at up to four times the rate of the general population, yet fewer than 11% of humanitarian financing flows in 2024 were tagged as disability-inclusive against a global disability prevalence of approximately 15% — the WHO 2024 estimate of 1.3 billion people, or one in six. The three-figure gap between the population, the funding, and the mortality outcome is the survey's headline finding, and it is the metric the Sendai mid-point asked governments to close before 2030.
Overview of disaster-disability-inclusion compliance
Disability inclusion in disaster preparedness rests on three load-bearing instruments. The Sendai Framework, adopted at the Third UN World Conference on DRR in March 2015, is the only universal DRR instrument and the only one with an explicit, repeated disability-inclusion mandate; persons with disabilities are named in Priorities for Action 1, 2, 3 and 4, and Targets E, F and G are formally disaggregated by disability in the Sendai Monitoring Framework that UNDRR operates. Article 11 of the UN Convention on the Rights of Persons with Disabilities (CRPD), in force since 2008, requires States Parties to take all necessary measures to ensure the protection and safety of persons with disabilities in situations of risk. The Inter-Agency Standing Committee (IASC) Guidelines on the Inclusion of Persons with Disabilities in Humanitarian Action, adopted in November 2019, set the actual sector-by-sector standards humanitarian actors are expected to meet.
Against that floor, five failure modes recur across the 2024–26 evidence base:
The early-warning multimodality gap — operational, fully multimodal Common Alerting Protocol delivery (SMS plus accessible push, audio siren, visual strobe, and sign-language video) exists in fewer than 30 of the roughly 130 CAP-implementing countries.
The institutional-evacuation gap — residential institutions for older people and persons with intellectual or psychosocial disabilities routinely have no functioning evacuation plan, concentrating the people least able to evacuate into a single building.
The accessible-shelter and accessible-evacuation-centre gap — door widths, ramp gradients, and accessible WASH facilities in temporary settlements fail IASC Sphere-aligned standards in most post-event audits.
The displacement-disaggregation gap — registration mechanisms in receiving states consistently fail to capture disability on intake, so the displaced disabled population is invisible in the operational data.
The funding-marker gap — only about 11% of humanitarian financing is tagged disability-inclusive, and national DRR budgets specifically tagged for accessibility sit in the low single digits even in high-capacity reporting states.
The treaty floor is in place; the implementation gap is one of national-budget allocation, of who sits at the planning table, and of whether organisations of persons with disabilities (OPDs) are coordinated with as principals or briefed as stakeholders.
Breakdown by domain
The 2024–26 evidence sorts into four operational domains: early-warning systems, institutional residents and evacuation, recent case studies, and climate-adaptation finance.
Early-warning systems: multimodal CAP adoption
Sendai Target G — "substantially increase the availability of and access to multi-hazard early warning systems" — is the target with the clearest accessibility test, and the one where the gap is most measurable. The Common Alerting Protocol (CAP), an ITU-T standard (X.1303) that allows a single alert message to be issued in multiple modalities (text, audio, visual, sign-language video) simultaneously, is the field's accepted technical answer.
The WMO–ITU CAP Adoption Tracker, updated through late 2025, listed CAP implementation in approximately 130 countries. Operational, fully multimodal CAP — the same alert delivered as SMS, accessible push notification, audio siren, visual strobe, and sign-language video on broadcast — is concentrated in fewer than 30. The remainder operate CAP only partially: SMS-based mobile alerting that excludes deaf users without visual confirmation; audio sirens that blind users can hear but that convey no instruction; visual-only alerts that deaf-blind users miss entirely. The UN Secretary-General's Early Warnings for All initiative, launched in March 2022 with a target of full global coverage by end-2027, has made multimodal accessibility one of its formal workstreams; the 2025 mid-initiative report notes that of the 30 priority countries identified for initial scale-up, fewer than half had a national plan with explicit accessibility provisions for persons with disabilities as of mid-2025.
Institutional residents and evacuation
The IASC Guidelines specify stair-evacuation chairs, ramp-equipped shelters, and accessible WASH as the operational floor for sheltering and evacuating persons with disabilities. Post-event audits across the 2023–25 reporting cycle find those provisions absent from the public-building inventories of most affected jurisdictions. Where deinstitutionalisation reforms have progressed prior to a disaster, the risk profile of the disabled population is measurably lower, because community-based services prove more resilient than institutional ones. The recurring policy implication, named in the European Disability Forum's 2024 joint statement with Inclusion Europe, is that deinstitutionalisation is disaster risk reduction: a residential institution in the path of a flood, a wildfire, or a front line concentrates the people least able to evacuate themselves into a single building with no plan for them.
Türkiye–Syria, 6 February 2023
The 7.8-magnitude earthquake that struck the Türkiye–Syria border in the early hours of 6 February 2023, followed by a 7.5 aftershock the same day, killed more than 59,000 people and displaced over 3 million. Post-event reviews — conducted across 2023 and 2024 by the Disability-Inclusive Disaster Response Coalition, the International Disability Alliance, Human Rights Watch, and the Turkish Confederation of the Disabled (Türkiye Sakatlar Konfederasyonu) — produced the most-detailed operational record the field has had since the 2011 Tōhoku and 2015 Nepal events.
The recurring findings: residents of residential institutions in the affected provinces had no functioning evacuation plan; deaf residents missed the morning's AFAD (Disaster and Emergency Management Authority) SMS-based warnings entirely because the alerts were audio-only on most networks; wheelchair users were excluded from upper-floor evacuations in collapsed-but-standing buildings because the stair-evacuation chairs the IASC Guidelines specify were absent from almost every public building inventoried. Temporary container settlements deployed across the first six months were largely inaccessible. Reconstruction tenders issued by the Turkish Housing Development Administration (TOKİ) during 2024 began to incorporate accessibility specifications, but the coalition's 2025 follow-up review found that as-built compliance lagged the tender language by a substantial margin. The Syrian side of the border, where the response was constrained by sanctions, divided territorial control, and the prior decade of conflict-driven institutional collapse, produced almost no operational disability data — itself a finding the coalition's reports underline.
Pacific cyclone seasons 2023–24 and 2024–25
The Pacific cyclone seasons of 2023–24 and 2024–25 were the field's clearest demonstration of what locally-led, OPD-coordinated disaster response looks like in practice. Cyclone Lola, which struck Vanuatu as a Category 5 system on 24–25 October 2023 — the earliest-recorded Category 5 in the Southern Hemisphere season — was followed by Cyclones Judy and Kevin (March 2023, retrospectively), Mal (November 2023), and a sequence of late-2024 systems including Kong-Rey's edge effects.
The Vanuatu Disability Promotion and Advocacy Association (VDPA), working with the National Disaster Management Office and the Pacific Disability Forum (PDF), ran a disability-inclusive response model that other Pacific NDMOs have begun to copy. The model has three operational components: a pre-positioned register of persons with disabilities at province and area-council level, maintained with consent and used only by trained NDMO and VDPA personnel; community-level disability focal points trained on multimodal alert dissemination, who carry the early warning into the last mile when SMS and radio do not reach; and accessible-evacuation-centre audits conducted jointly with the Public Works Department in the lull between cyclone seasons. The model is imperfect — the registers are incomplete in remote outer-island communities, and the audit programme has lagged the schoolbuilding programme — but it is the closest functioning example of Sendai Priorities 2 and 4 in a small-island context. The PDF's 2024 regional review noted that Fiji, Tonga and Solomon Islands have begun adapting elements of the Vanuatu model, with mixed progress on funding the focal-point networks at scale.
Ukraine: displacement and the institutionalisation multiplier
Pre-invasion baselines from the State Statistics Service estimated approximately 2.7 million persons with registered disability status in Ukraine, with the actual prevalence figure (using the WHO 15% baseline) closer to 6 million. The UN Refugee Agency's registered displaced-population data, updated through 2025, shows roughly 6.8 million Ukrainian refugees recorded across Europe and a further estimated 3.7 million internally displaced; disability prevalence within those populations is consistently undercounted because the registration mechanisms in receiving states do not capture it on intake.
The operational record from 2022 through 2025 has been documented in detail by Human Rights Watch, the European Disability Forum, and the National Assembly of People with Disabilities of Ukraine (NAPD). Three findings recur. First, evacuation of residential institutions for older people and persons with intellectual or psychosocial disabilities in the path of advancing front lines was, in the first months, conducted via paper "evacuation lists" with no clear locus of responsibility between the Ministry of Social Policy and regional administrations. Second, accessible-shelter provision in Western Ukraine and in receiving states was a binding constraint throughout 2022–23 and remained partial in 2024. Third, where deinstitutionalisation reforms had progressed prior to 2022 — most notably in some western oblasts — the disaster-risk profile of the disabled population was measurably lower.
Climate-adaptation finance and the Loss and Damage Fund
The Loss and Damage Fund (formally the Fund for responding to Loss and Damage), agreed at COP27 in November 2022, operationalised at COP28 in December 2023, and with its board's first formal disbursement decisions taken across 2024 and 2025, is the newest piece of the architecture. Disability inclusion was raised in the fund's governance design — by the International Disability Alliance and a coalition of OPDs from climate-vulnerable states — but the founding instruments do not name disability as a cross-cutting commitment, and the initial project pipeline approved at the board's 2025 meetings did not contain explicit disability-inclusion line items beyond generic "vulnerable groups" language. The advocacy ask going into the fund's 2026 replenishment cycle is for the kind of named, budgeted disability-inclusion provision that the Sendai monitoring framework already nominally requires.
Quantitative insights
Read together, the 2024–26 monitoring data produces a consistent set of percentages:
Up to 4× — the disaster mortality multiplier for persons with disabilities, the figure most commonly cited in the Sendai mid-term review and the UN ESCAP 2024 Asia-Pacific Disaster Report, drawing on post-event mortality reviews from the 2011 Tōhoku earthquake forward.
11% of humanitarian financing tagged disability-inclusive in 2024, the share reported in Development Initiatives' Global Humanitarian Assistance Report 2024, against a global disability prevalence baseline of 15%.
15% of the global population — approximately 1.3 billion people, or 1 in 6 — live with significant disability per the WHO 2024 baseline, the figure against which any operational disability-inclusion benchmark should be read.
Fewer than 40 reporting states had submitted any disability-disaggregated data on local DRR strategies as of the 2023 Sendai reporting cycle, despite Targets E, F and G being formally disaggregable by disability.
Consultation with persons with disabilities in national DRR strategy development was self-reported by 70-plus states but verifiable in fewer than half of those cases, per UNDRR's mid-term review report.
Roughly 130 countries have implemented CAP; fewer than 30 operate it as fully multimodal, per the WMO–ITU CAP Adoption Tracker (late 2025).
Of the 30 Early-Warnings-for-All priority countries, fewer than half had a national plan with explicit disability-accessibility provisions as of the 2025 mid-initiative report.
National DRR budgets tagged for accessibility sit in the low single digits even in high-capacity reporting states.
Taken together: a population share of 15%, a mortality multiplier of up to 4×, a funding tag of 11%, and a national-budget allocation in the low single digits. The numbers describe a single structural shape — a population disproportionately exposed to disaster mortality, disproportionately under-resourced in disaster preparedness, and disproportionately invisible in the disaster data.
What good policy looks like in 2026
The 2025 mid-term review's Political Declaration of 19 May 2023 reaffirmed the disability-inclusion language and added two specific new lines: a call for Common Alerting Protocol adoption with multimodal alerting, and an explicit reference to the IASC Guidelines as the operational floor. The countries doing this well share five features, not one: (1) a national DRR strategy that names disability inclusion with measurable indicators, not aspirational language; (2) OPDs at the DRR coordination table from strategy design through after-action review; (3) multimodal CAP-compliant early warning with audited accessibility in all four modalities; (4) accessible-shelter and accessible-evacuation-centre standards integrated into the building code and audited between events; and (5) a deinstitutionalisation pathway treated as part of the DRR portfolio, not a separate social-policy track.
Three country examples show what this looks like operationally. Bangladesh's Cyclone Preparedness Programme (CPP), jointly run by the Government of Bangladesh and the Bangladesh Red Crescent Society since 1973, has incorporated disability-inclusion training in its 76,000-strong volunteer network since 2018 and works with the National Forum of Organisations Working with the Disabled (NFOWD) on multimodal warning dissemination in coastal districts. The Philippines' Office of Civil Defense (OCD), under Republic Act 10121, has formalised inclusion of disabled people's organisations in its DRR councils at national and regional level; implementation at municipal level remains uneven. Vanuatu's VDPA-NDMO model is described above. For deeper background read the CRPD glossary entry, the national regulations index, and the wider 2026 reporting record.
Call to action for disaster planners and funders
The treaty floor names what disability-inclusive disaster preparedness requires; what the survey data shows is that the gap is one of allocation and inclusion at the planning table. Concrete next steps for 2026:
Move to fully multimodal CAP delivery. Audit the four delivery channels (SMS plus accessible push, audio siren, visual strobe, sign-language video on broadcast) and close whichever modality is missing from the national alert chain.
Seat OPDs at the DRR coordination table as principals. Inclusion in strategy design, contingency planning, and after-action review — not as stakeholders briefed after the fact.
Disaggregate the Sendai Monitoring data by disability. Targets E, F and G are formally disaggregable; submitting the disaggregated data is what makes the framework operational.
Treat deinstitutionalisation as DRR. Cross-walk the social-policy deinstitutionalisation pathway with the DRR portfolio; institutional concentrations are a measurable disaster-risk multiplier.
Tag disability inclusion in the humanitarian-financing markers. Move the 11% disability-inclusive financing share toward the 15% prevalence baseline, and require a named disability-inclusion line item in Loss and Damage Fund project approvals from the 2026 replenishment forward.
Audit accessible-shelter and accessible-evacuation-centre compliance between events. Integrate the IASC Sphere-aligned standards into the building code, and run joint audits with OPDs in the lull periods, on the Vanuatu model.
Capture disability on intake for displaced populations. Build a Washington Group Short Set question into refugee and IDP registration so the displaced disabled population becomes visible to the response.
Conclusion
The Sendai Framework, Article 11 of the CRPD, and the 2019 IASC Guidelines collectively say what disability-inclusive disaster preparedness looks like in enough operational detail that no country in 2026 can claim it does not know. The 2025 mid-term review, the Türkiye-Syria post-event record, the Pacific cyclone seasons, and the Ukraine displacement data show that the gap is one of national-budget allocation, of who sits at the planning table, and of whether OPDs are coordinated with as principals or briefed as stakeholders. Closing it before 2030 is what the Sendai mid-point asked for. Whether the next five years deliver it is, again, a national-budget decision.
---
title: Disability rights in Türkiye, the GCC, the Levant, and Israel: the 2026 regional dossier
url: https://www.disabilityworld.org/articles/disability-rights-in-turkey-and-middle-east/
description: Across Türkiye, the GCC, the Levant, and Israel, disability-rights frameworks span constitutional anchors, religious-law foundations, recent EU-adjacent reform, and the long aftermath of the 2023 Türkiye-Syria earthquake. The 2026 landscape, mapped country by country.
author: Disability World
pubDate: 2026-05-22
tags: turkey, middle-east, gcc, israel, levant, regional-report, regulations
---
# Disability rights in Türkiye, the GCC, the Levant, and Israel: the 2026 regional dossier
Disability rights in Türkiye, the GCC, the Levant, and Israel: the 2026 regional dossier
From Ankara to Amman, from Riyadh to Tel Aviv, the disability-rights frameworks across Türkiye and the Middle East are easy to read on paper and far harder to read on the ground. The 2026 picture shows ten jurisdictions with near-universal CRPD ratification, primary statutes already on the books, and an enforcement gap whose width depends almost entirely on whether the named regulator has budget and independence.
The regional treaty floor is unusually flat. Almost every jurisdiction in scope here ratified the UN Convention on the Rights of Persons with Disabilities (CRPD) between 2008 and 2012 — most of them in the 2008–09 window. Almost every one followed ratification with a domestic primary statute and the designation of a ministry or commission. On paper, the 2026 landscape looks coherent and rights-aligned.
The patchwork beneath the treaty floor is anything but flat. Some statutes are decades old and frozen by institutional crisis (Lebanon's Law 220 of 2000). Some are recent and considered the regional model (Jordan's Law 20 of 2017). Some sit on top of subnational frameworks that exceed the federal floor (Sharjah's emirate-level codes). One — Egypt — is treated fully in the parallel Africa dossier and appears here only as a cross-link. The catalogue below gives each of the ten one identical entry: primary statute, regulator, CRPD status, and what civil society and the courts are actually doing with the framework.
Evidence index · Cat. 2026.05
10 jurisdictions · ordered by region, then by CRPD ratification year
Ordering is editorial, not quantitative — Part I groups Anatolia and the Levant, Part II groups the GCC, Part III flags Egypt for the Africa dossier. Within each part, jurisdictions appear in the order their lead institution has the deepest enforcement footprint.
The legal floor: CRPD across the region
The single most important fact about disability rights in this region is also the most easily mis-told: ratification of the UN Convention on the Rights of Persons with Disabilities (CRPD) is near-universal here, and enforcement is almost nowhere proportionate to ratification.
The dates themselves are worth keeping in mind. Türkiye deposited its instrument of ratification on 28 September 2009. Israel ratified in 2012 with declarations on competence and on the optional Communications Procedure. The UAE acceded in 2010, having signed in 2008. Saudi Arabia, Qatar, Jordan and Egypt all deposited in 2008. Lebanon signed in 2007 and has not ratified — a status that is now nearly two decades old and is repeatedly noted in UN Treaty Body summary reports.
The Optional Protocol, which allows individual communications to the CRPD Committee, is ratified far less consistently. Türkiye ratified the Protocol in 2015; the Arab states that have ratified it remain a minority. The CRPD Committee's published Concluding Observations on the region — Türkiye in 2019 (with the second cycle under way for the 2024–25 review), Saudi Arabia in 2019, Qatar in 2015 and again in 2024, Jordan in 2017, and the UAE in 2016 — share a remarkably consistent set of recommendations: end the use of substituted decision-making, accelerate built-environment accessibility, disaggregate data by disability across all sectoral statistics, and explicitly include women and girls with disabilities in national strategies.
Ratification is the easy half. The enforcement gap that follows is a budget, regulator-independence, and judicial-review-pathway failure — not a drafting one.
Part I · Anatolia and the Levant
Four jurisdictions where the statute is older than the politics
Türkiye, Israel, Jordan, and Lebanon — covering Constitutional Court jurisprudence, statutory-commission enforcement, the region's most CRPD-aligned 2017 reform, and the institutional freeze that took a primary statute offline in everything but name.
E·01
Türkiye · Türkiye Cumhuriyeti
Primary statute
Law No. 5378 on Persons with Disabilities (Engelliler Hakkında Kanun), enacted 2005 and substantially amended in 2014, 2021, and again in 2024. The 2005 act established the legal floor — non-discrimination, accessible-built-environment obligations on public buildings and transport, and a framework for educational and employment accommodations.
The 2014 amendments operationalised the framework; the 2021 amendments extended grace periods for built-environment compliance — a recurring pattern in Turkish disability legislation, in which statutory accessibility deadlines have been extended at least four times since the original 2012 date set under Article 7 of Law 5378. The 2024 amendments tightened enforcement language on public-procurement accessibility but kept the underlying compliance horizon flexible.
Regulator / commission
Engelliler ve Yaşlı Hizmetleri Genel Müdürlüğü (General Directorate of Services for Persons with Disabilities and the Elderly), operating within the Ministry of Family and Social Services. Operational responsibility was transferred to the Directorate under the 2014 amendments. See aile.gov.tr/eyhgm.
CRPD status
Ratified the Convention on 28 September 2009; ratified the Optional Protocol in 2015. First-cycle Concluding Observations issued by the CRPD Committee in 2019; the State party is currently under review in the 2024–25 second cycle, with earthquake-recovery disability inclusion at the centre of the reporting obligations.
Case-law trajectory + civil society
The Turkish Constitutional Court has been an unusually active forum for disability-rights litigation. Its 2018 ruling on accessible polling stations and its 2021 individual-application decision on accessible secondary education for blind students are now standard references in Turkish administrative-law practice.
Civil-society capacity, channelled through ENİL-affiliated and federation-level organisations such as Türkiye Sakatlar Derneği and the Engelli Hakları İzleme Grubu, has been the primary driver of strategic litigation under Law 5378.
The 2023 Türkiye–Syria earthquake as regional stress test
The 6 February 2023 Kahramanmaraş-centred earthquakes — magnitudes 7.8 and 7.5, the deadliest seismic event in modern Turkish history — became the region's clearest stress test of disability-inclusive disaster response. Institutional residents of state-run care facilities in Hatay, Adıyaman, and Kahramanmaraş were among the populations with the highest mortality concentrations.
Subsequent reporting from the Engelliler ve Yaşlı Hizmetleri Genel Müdürlüğü and from independent Turkish disability organisations documented systemic shortfalls in accessible temporary shelter, in the supply of mobility devices and continence supplies during the acute response phase, and in the relocation of deafblind and intellectually disabled survivors from collapsed institutional settings into temporary facilities that frequently lacked sign-language interpretation or accessible sanitation.
Türkiye's 2024–25 CRPD review cycle has placed earthquake-recovery disability inclusion at the centre of the State party's reporting obligations. The pattern is regional: when the institutions themselves are physically and operationally disrupted, the inclusion margin collapses.
E·02
Israel · מְדִינַת יִשְׂרָאֵל
Primary statute
Equal Rights for Persons with Disabilities Law, 5758-1998, one of the older comprehensive disability statutes in the region. Its 2005 Chapter 5 on accessibility — and the detailed accessibility regulations issued from 2009 onward — extended substantive accessibility obligations to public services, public buildings, transport, and information technology.
A 2023 statutory reform tightened enforcement timelines for the long-running accessibility-deadline extensions granted under the 2013–2018 transitional regime; the practical effect is that as of 2026 most categories of public service have exhausted their statutory grace period.
Regulator / commission
The Commission for Equal Rights of Persons with Disabilities, an independent statutory body within the Ministry of Justice, has both regulatory and enforcement powers, including the ability to issue compliance orders and to bring civil enforcement actions. See gov.il/the_commission_for_equal_rights_of_persons_with_disabilities.
CRPD status
Ratified the Convention in 2012, with declarations on competence and on the optional Communications Procedure. The Optional Protocol has not been ratified. CRPD Committee Concluding Observations issued in 2017.
Case-law trajectory + civil society
The Tel Aviv District Court has, since the late 2010s, become a recurring forum for accessibility-deadline enforcement litigation. A pattern of class-action settlements — against bank branch networks, public transport operators, and retail chains — has driven the operationalisation of the regulations far more aggressively than ministerial enforcement alone.
Israel's civil-society architecture — anchored by Bizchut: The Israel Human Rights Center for People with Disabilities — has been a central litigant and policy interlocutor since the 1998 Act was drafted.
RegionLevantEnforcement strengthStrong · Commission-led + district-court class actions
E·03
Jordan · المملكة الأردنية الهاشمية
Primary statute
Law No. 20 of 2017 on the Rights of Persons with Disabilities, which replaced an earlier 2007 statute and is widely regarded — including by the CRPD Committee in its 2017 Concluding Observations — as the most CRPD-aligned primary statute in the Arab region.
The 2017 Act introduced a phased ten-year national strategy with explicit targets in education, employment, accessibility, and rehabilitation — milestones whose 2027 horizon means that 2026 is the reporting year against which most operational outcomes will be assessed.
Regulator / commission
The Higher Council for the Rights of Persons with Disabilities (HCD), established under the Act and chaired ex officio by HRH Prince Mired bin Ra'ad, is both the policy coordinator and an active enforcement and complaints body. See hcd.gov.jo.
CRPD status
Ratified the Convention in 2008. Concluding Observations issued in 2017. The 2026 reporting year coincides with the closing window of the national strategy attached to the 2017 Act.
Case-law trajectory + civil society
Jordan's disability-rights framework has had to absorb one of the region's largest refugee populations. UNHCR Jordan's coordination with the HCD on disability-inclusive services for Syrian refugees — including the registration of disability-related needs at the Zaatari and Azraq camps and the integration of refugees with disabilities into national rehabilitation services — is one of the few documented examples in the region of formal coordination between the national disability framework and a refugee-response operation.
Persistent gaps in the funding floor for these services remain a structural concern as donor funding for the Jordan Response Plan continues to decline.
RegionLevantEnforcement strengthStrong (statutory) · constrained by funding floor
E·04
Lebanon · الجمهورية اللبنانية
Primary statute
Law No. 220 of 2000 on the Rights of Disabled Persons remains, nominally, the primary statute. In substance, Lebanon's post-2019 financial collapse, the August 2020 Beirut port explosion, and the cascading institutional crisis since have effectively frozen meaningful implementation. The Personal Disability Card system, which under Law 220 is the gateway to most rights and entitlements, has continued to be issued, but the underlying entitlements — disability pensions, transport allowances, accessibility audits — are operationally curtailed by the broader fiscal situation.
Regulator / commission
The National Council for Disabled Persons, the body designated under Law 220 as the coordinating authority, sits within the Ministry of Social Affairs. It has continued to operate, but with severely constrained budgets and limited regulatory reach.
CRPD status
Lebanon signed the CRPD in 2007 and has not ratified — a status that is now nearly two decades old and is repeatedly noted in UN Treaty Body summary reports. The Optional Protocol is also unratified.
Case-law trajectory + civil society
Civil-society organisations — Lebanese Physical Handicapped Union (LPHU), the Youth Association of the Blind, and the network coordinated under the Arab Forum for the Rights of Persons with Disabilities — continue to advocate for ratification and for the operational revival of Law 220.
The 2026 question for Lebanon is less about reform of the statute than about whether the institutional state has the capacity to enforce the one it already has.
RegionLevantEnforcement strengthFrozen · statute on the books, institution in crisis
Part II · The GCC
Four federal floors, with subnational variation that often exceeds them
UAE, Saudi Arabia, Qatar, and Kuwait — plus a shared entry for Bahrain and Oman. The pattern across the GCC is a single federal or national statute, an empowered ministry, and a Vision-aligned strategy framework. The enforcement question turns on whether emirate- or municipality-level codes go further.
E·05
United Arab Emirates · الإمارات العربية المتحدة
Primary statute
Federal Decree-Law No. 29 of 2006 concerning the Rights of People with Special Needs, amended in 2009 and again in 2015 to broaden the non-discrimination scope and to extend coverage in employment, transport, and information access. The National Policy for Empowering People of Determination — the official terminology preferred in UAE federal use since 2017 — acts as the integrating strategy document.
Sharjah operates a distinct, more expansive provincial framework anchored in the Sharjah City for Humanitarian Services (established 1979) and a series of emirate-level laws and decrees that predate and exceed the federal floor. Abu Dhabi and Dubai have built parallel municipal-level accessibility codes, including the Dubai Universal Design Code (most recently updated 2023).
Regulator / commission
Operational authority sits with the Ministry of Community Development. See mocd.gov.ae. Emirate-level oversight runs in parallel through Sharjah City for Humanitarian Services, the Community Development Authority in Dubai, and the Department of Community Development in Abu Dhabi.
CRPD status
Signed the Convention in 2008; acceded in 2010. Concluding Observations issued by the CRPD Committee in 2016. The Optional Protocol has not been ratified.
Case-law trajectory + civil society
The 2021 Expo 2020 Dubai accessibility programme — including its sensory-friendly hours, comprehensive sign-language interpretation across pavilions, and a published accessibility audit — left a measurable legacy in Dubai's public-event accessibility standards, although replication across non-event built environments has been uneven.
Strategic litigation is rare; the federation's accessibility progress runs primarily through municipal code-setting and event-driven programmes rather than through the courts.
RegionGCCEnforcement strengthModerate · uneven across emirates, strong in Sharjah and Dubai
E·06
Saudi Arabia · المملكة العربية السعودية
Primary statute
The Disability Code (Provision Code for Persons with Disabilities) of 2000, enacted by Royal Decree, established the country's primary disability-rights framework. The 2016 Vision 2030 framework folded disability-inclusion targets into broader human-capital and employment metrics — most concretely, an explicit employment-rate target for persons with disabilities incorporated into the National Transformation Programme.
The Vision 2030 framing matters because it ties disability-employment outcomes to a politically prioritised macro-level reform agenda, in a way that the standalone 2000 Code did not.
Regulator / commission
The Authority for the Care of Persons with Disabilities, operating under the Ministry of Human Resources and Social Development, coordinates national implementation. The National Disability Programme functions as the operational strategy.
CRPD status
Ratified the Convention in 2008. The Optional Protocol has not been ratified. CRPD Committee Concluding Observations issued in 2019; follow-up dialogue running through 2024.
Case-law trajectory + civil society
The National Society for Human Rights, established 2004 and operating as a quasi-governmental rights body, has been a recurring monitor on the implementation gaps of the 2000 Code, particularly on accessible public-sector built environments and on women with disabilities.
The CRPD Committee's 2019 Concluding Observations on Saudi Arabia explicitly called out the persistence of substituted decision-making frameworks and the under-disaggregation of disability data in national statistics — both items still open in the 2024 follow-up.
RegionGCCEnforcement strengthModerate · Vision-2030-aligned, gaps on built environment and substituted decision-making
E·07
Qatar · دولة قطر
Primary statute
Law No. 2 of 2004 on People with Special Needs remains the primary statute. The framework is short by regional standards and has accreted operational depth largely through the institutional anchors built around it rather than through statutory amendment.
Regulator / commission
The National Authority for the Care of Persons with Disabilities is the lead implementing body. Qatar's distinctive contribution to the regional landscape is the Mada — Assistive Technology Center, established in 2010 under the Qatar Foundation umbrella, which has become a regional reference point for Arabic-language assistive technology, screen-reader Arabic-language voice support, and accessible-publishing standards. See mada.org.qa.
CRPD status
Ratified the Convention in 2008. The CRPD Committee reviewed Qatar's combined second and third periodic reports in 2024; the published Concluding Observations focus on labour-force inclusion, on the accessibility of the criminal-justice system, and on the persistence of restrictive guardianship arrangements.
Case-law trajectory + civil society
The Mada Center is the regional civil-society anchor of consequence — not through litigation but through standard-setting. Its Arabic-language assistive-technology work has measurably improved screen-reader coverage on flagship platforms across the region (the e-Devlet, UAE PASS, and Absher portals all reference Mada-adjacent guidance).
Strategic litigation under Law 2 of 2004 is rare; the country's progress runs through standard-setting institutions rather than through the courts.
RegionGCCEnforcement strengthModerate · standards-led via Mada
E·08
Kuwait · دولة الكويت
Primary statute
Law No. 8 of 2010 concerning Rights of Persons with Disabilities is the primary instrument, replacing an earlier 1996 sectoral provision. It anchors a relatively comprehensive set of educational, employment, transport, and accessibility obligations on public bodies, with implementing regulations issued through the Public Authority for Disability Affairs.
Regulator / commission
The Public Authority for Disability Affairs (PADA), a statutory body, is the coordinating regulator. PADA administers the disability-card and benefits system and is the lead drafter of implementing regulations under Law 8.
CRPD status
Ratified the Convention in 2013. The Optional Protocol has not been ratified. CRPD Committee first-cycle review process remains open, with reporting cycles running through the 2020s.
Case-law trajectory + civil society
Civil-society capacity is concentrated in the Kuwait Society for the Handicapped and in DPI-Arab Region affiliates. Strategic litigation is rare; the dominant accountability pathway runs through PADA's administrative determinations and through periodic CRPD reporting rather than through the courts.
RegionGCCEnforcement strengthModerate · regulator-led, no strategic-litigation pathway
E·09
Bahrain and Oman · مملكة البحرين · سلطنة عُمان
Primary statute
Both jurisdictions sit on sectoral provisions rather than a single comprehensive primary statute on the Jordan-2017 or Kuwait-2010 model. Bahrain's framework is anchored in Law No. 74 of 2006 (with later amendments) and a 2018 National Strategy; Oman operates through Royal Decree 63/2008 and successive education- and rehabilitation-sector regulations.
Regulator / commission
Bahrain's Ministry of Social Development and Oman's Ministry of Social Development are the lead coordinating ministries. Bahrain has additionally constituted a Higher Committee for Persons with Disabilities; Oman channels operational delivery through the Department for the Disabled within MoSD.
CRPD status
Bahrain ratified the Convention in 2011; Oman acceded in 2009. Neither has ratified the Optional Protocol. Bahrain's first-cycle Concluding Observations were issued in 2017; Oman's in 2018.
Case-law trajectory + civil society
Both jurisdictions are below the regional median for strategic-litigation activity. Civil-society capacity is concentrated in single anchor organisations — the Bahrain Disabled Sports Federation and the Oman Association for the Disabled — that operate primarily as service providers rather than as litigants or policy interlocutors.
Egypt's primary statute, regulator, and CRPD review are treated in depth in the Africa regional dossier. The entry below is a stub for navigation and cross-reference.
E·10
Egypt · جمهورية مصر العربية
Primary statute
Law No. 10 of 2018 on the Rights of Persons with Disabilities replaced earlier sectoral provisions. Constitutional anchoring is provided by Article 81 of the 2014 Constitution, one of the more explicit constitutional disability provisions in the Arab region. The 2018 statute is comprehensive on paper; enforcement on accessibility and on the elimination of substituted decision-making remain the dominant CRPD-Committee follow-up items.
Regulator / commission
The National Council for Persons with Disabilities (NCPD), established 2019 under the 2018 statute, is the coordinating body.
CRPD status
Ratified the Convention in 2008. The Optional Protocol has not been ratified.
Case-law trajectory + civil society
Full treatment — including the NCPD's operational footprint, the constitutional-article anchoring, and the civil-society architecture around the 2018 Act — is in the Africa regional dossier.
RegionNorth Africa / cross-linkEnforcement strengthSee Africa dossier
Three cross-cutting threads that recur across every exhibit
Women with disabilities. The intersection of gender and disability is the single most consistently called-out gap in the CRPD Committee's Concluding Observations across every country reviewed in this region. Disaggregated data on the educational attainment, employment rates, and access to sexual and reproductive health services for women with disabilities is collected nowhere in the region on a basis the Committee considers adequate.
Refugees and displaced persons. The Jordan-UNHCR coordination model is the regional outlier. In Türkiye — host to one of the world's largest registered Syrian refugee populations — refugee-disability data integration with national rehabilitation services has progressed unevenly across provinces. In Lebanon, the question is largely moot in the absence of a functioning national disability framework. UNHCR's 2024 Need to Know Guidance on Working with Persons with Disabilities in Forced Displacement remains the operational reference.
Digital public services after COVID. The 2020–22 acceleration of e-government produced both progress and new exclusion. Türkiye's e-Devlet portal, Saudi Arabia's Absher and Tawakkalna platforms, the UAE's UAE PASS, and the Israeli Gov.il portal each underwent accessibility audits or upgrade cycles between 2022 and 2025. Coverage of Arabic-language screen-reader compatibility — the binding constraint for the regional blind community — has measurably improved on flagship platforms but remains uneven on sub-portals.
What to watch in 2026
The catalogue above maps the statutory floor as it stands today. The list below is what is actually in motion — the reviews, deadlines, and political windows that will reshape one or more of the ten entries before the next dossier cycle.
Tier 1 — Concluding-Observations cycles
Türkiye's 2024–25 second-cycle CRPD review, with earthquake-recovery disability inclusion as the lead reporting obligation. (E·01)
Jordan's 2026 national-strategy reporting year, closing the 2017 Act's ten-year horizon a year ahead of schedule. (E·03)
Qatar's 2024 follow-up to the second-and-third combined Concluding Observations, with labour-force inclusion as the open item. (E·07)
Saudi Arabia's 2024 follow-up dialogue on substituted decision-making and built-environment accessibility. (E·06)
Tier 2 — Statutory and enforcement deadlines
Israel's 2023 statutory reform: the practical 2026 effect is that most categories of public service have exhausted their accessibility-deadline grace period. (E·02)
Türkiye's Article 7 built-environment compliance horizon, repeatedly extended since 2012; the 2024 amendments left it flexible. (E·01)
UAE municipal-code refreshes — the Dubai Universal Design Code last updated 2023; Sharjah and Abu Dhabi cycles running in parallel. (E·05)
Tier 3 — Civil-society and ratification gaps
Lebanon: CRPD ratification status — signed 2007, unratified — as the structural precondition for everything else. (E·04)
Optional Protocol ratification gaps across every GCC jurisdiction. (E·05, E·06, E·07, E·08, E·09)
Disaggregated gender-and-disability data collection, the single most consistently flagged gap across the region.
The through line
From Ankara to Beirut to Riyadh, the region's disability-rights frameworks were mostly built between 2000 and 2018, mostly aligned to the CRPD by 2010, and mostly stuck somewhere between the statute book and the lived environment in 2026.
The jurisdictions making measurable progress — Türkiye on Constitutional Court jurisprudence, Israel on Commission-driven enforcement, Jordan on the HCD's coordination capacity, the UAE on emirate-level municipal codes, Qatar on Arabic-language assistive technology — share a single feature: a named institution with budget, independence, and a functioning enforcement pathway. The ones that have not closed the gap (Lebanon since 2019; Saudi Arabia on built-environment accessibility; Bahrain and Oman on strategic litigation) have not funded the institution the statute names.
The 2023 Türkiye–Syria earthquake exposed how thin the inclusion margin still is when the institutions themselves are physically and operationally disrupted. The 2026 cycle of CRPD Committee reviews — Türkiye, Qatar, and several others — will be the next moment of comparative accountability.
The bottom line
Ten jurisdictions, one through line: the statute is only as strong as the institution it names.
Every one of the ten entries in this dossier has a primary statute on the books. Every one designates a regulator. Eight of the ten have ratified the CRPD. The gap between the legal floor and the lived environment in 2026 tracks, almost without exception, the budget and independence of the named institution. Where the institution has both — the Israeli Commission, the Jordanian HCD, the Mada Center as a regional standard-setter — outcomes move. Where it has neither — Lebanon's National Council since 2019, Saudi built-environment enforcement — they do not.
Primary sourcesRepublic of Türkiye, Law No. 5378 on Persons with Disabilities (Engelliler Hakkında Kanun, 2005; amendments 2014, 2021, 2024), mevzuat.gov.tr; Engelliler ve Yaşlı Hizmetleri Genel Müdürlüğü, aile.gov.tr/eyhgm; State of Israel, Equal Rights for Persons with Disabilities Law, 5758-1998; Commission for Equal Rights of Persons with Disabilities, gov.il; United Arab Emirates, Federal Decree-Law No. 29 of 2006, Ministry of Community Development, mocd.gov.ae; Kingdom of Saudi Arabia, Disability Code of 2000 (Royal Decree); Authority for the Care of Persons with Disabilities, Ministry of Human Resources and Social Development; State of Qatar, Law No. 2 of 2004; National Authority for the Care of Persons with Disabilities; Mada — Assistive Technology Center, mada.org.qa; Hashemite Kingdom of Jordan, Law No. 20 of 2017; Higher Council for the Rights of Persons with Disabilities (HCD), hcd.gov.jo; State of Kuwait, Law No. 8 of 2010; Public Authority for Disability Affairs; Republic of Lebanon, Law No. 220 of 2000, National Council for Disabled Persons, Ministry of Social Affairs; Kingdom of Bahrain, Law No. 74 of 2006; Sultanate of Oman, Royal Decree 63/2008; Arab Republic of Egypt, Law No. 10 of 2018, National Council for Persons with Disabilities (NCPD).
Treaty body sourcesUN Committee on the Rights of Persons with Disabilities, Concluding Observations on Türkiye (2019, 2024–25 cycle), Saudi Arabia (2019), Qatar (2015, 2024), Jordan (2017), the UAE (2016), Israel (2017), Bahrain (2017), Oman (2018), ohchr.org/treaty-bodies/crpd; UN Economic and Social Commission for Western Asia (UNESCWA), Disability in the Arab Region series and Disability Inclusion Country Profiles (2018, 2021, 2024 updates), unescwa.org; UNHCR, Need to Know Guidance on Working with Persons with Disabilities in Forced Displacement (2024 update).
ScopeThis is a regional catalogue, not a country-by-country compliance audit. Each exhibit summarises the statutory floor, the named regulator, the CRPD status, and the recent civil-society and case-law trajectory; it does not enumerate every subsidiary regulation or sectoral provision. Egypt appears only as a cross-link to the Africa dossier; Bahrain and Oman share a single exhibit because their frameworks are sectoral rather than anchored in a comprehensive primary statute on the Jordan-2017 or Kuwait-2010 model.
---
title: Disabled women's healthcare access in 2026: the intersection CEDAW and the CRPD finally started to enforce together
url: https://www.disabilityworld.org/articles/disabled-womens-healthcare-access-2026/
description: Disabled women face compounded physical-access, communication-access, and attitudinal barriers in every healthcare setting. The 2025 joint CEDAW–CRPD Committee recommendation and 2024-26 national reform data show where the floor is finally rising.
author: Disability World
pubDate: 2026-05-22
tags: women, healthcare, crpd, cedaw, intersectional, reproductive-rights, data
---
# Disabled women's healthcare access in 2026: the intersection CEDAW and the CRPD finally started to enforce together
By Disability WorldReading time: 11 minutes
Healthcare access for a disabled woman is not the sum of two separate problems. It is a single, intersectional encounter in which the building, the equipment, the clinician's training, the consent architecture and the reimbursement code all have to line up — and where any one of them missing pushes her below the standard of care non-disabled women receive. That is what the Convention on the Rights of Persons with Disabilities (CRPD) and the Convention on the Elimination of All Forms of Discrimination against Women (CEDAW) describe, in different vocabularies, as the same right. And that is what an everyday clinical visit — the gynaecological exam, the antenatal booking, the prescription review, the consent form — keeps revealing as missing.
2026 is the year that framing changed at the treaty level. The CRPD and CEDAW oversight committees issued their first joint general recommendation/general comment in October 2025, on harmful practices intersectional with disability, and for the first time the international human-rights system began collecting routine data on the gap. Reading the rest of this article requires holding two ideas at once: the legal floor has been settled for two decades, and the lived floor — the one in clinics — is only now being measured.
What "access" means in healthcare for a disabled woman
"Access" in this context is not a single barrier but a compounded one. Roughly one billion women and girls live with a disability worldwide — about one in five — and the WHO's 2024 update to its Global Report on Health Equity for Persons with Disabilities documents that they encounter healthcare across at least four simultaneous axes of difficulty: physical access to the room and the equipment in it; communication access to the clinician's questions and explanations; attitudinal access to a clinician who treats her as the decision-maker about her own body; and structural access to a system whose risk-assessment tools, screening pathways and billing codes were not designed with her in mind.
The compounded gap is what makes the disabled-woman category analytically distinct from "women" or from "people with disabilities" considered separately. Disabled women report intimate-partner violence at two to three times the rate of non-disabled women, per the WHO Violence Against Women global estimates and the 2024 UN-Women supplementary brief. Fewer than 20% of WHO Member States include disability-disaggregated accessibility indicators in their national maternity-care quality monitoring, per the WHO 2024 audit of national maternal-health information systems. The figures co-vary because they describe the same encounter — a woman whose disability is treated as the reason her testimony, her pain and her consent count less than a clinician's interpretation of them.
The treaty floor: three articles that finally talk to each other
The right of a disabled woman to healthcare is, on paper, one of the better-articulated rights in international law. Three articles do the load-bearing work.
Article 25 of the CRPD requires States Parties to recognise the right of persons with disabilities to "the enjoyment of the highest attainable standard of health without discrimination on the basis of disability," and specifies that this includes services "as close as possible to people's own communities."
Article 12 of CEDAW, dating from 1979, obliges States to eliminate discrimination against women in the field of healthcare; the Committee's General Recommendation No. 24 (1999) reads that obligation as covering reproductive, mental, and occupational health.
Article 6 of the CRPD — the only article in any UN human-rights treaty addressed specifically to women with disabilities — requires States Parties to take "all appropriate measures to ensure the full development, advancement and empowerment of women" with disabilities.
What changed in 2025 is that the two oversight committees stopped reading those articles in parallel and started reading them together. The joint general recommendation/general comment on harmful practices intersectional with disability, adopted in October 2025 after a three-year drafting process, names specific failure modes: non-consensual sterilisation, forced contraception, denial of legal capacity to make sexual and reproductive-health decisions, and coercive psychiatric treatment of women with intellectual and psychosocial disabilities. It is the first joint interpretive instrument the two committees have produced. A country's record on either treaty is now formally cross-referenced when the other committee reviews it.
Where the gap shows
The compounded gap is visible most clearly in five concrete clinical situations. Each is documented well enough that the failure mode can be named, and each is small enough that the fix is a question of procurement, training or law rather than of clinical science.
Physical-exam infrastructure: the exam-table problem
The most measurable gap in disabled women's healthcare is also the most basic: the equipment in the room. The WHO's 2024 Global Report on Health Equity for Persons with Disabilities includes an equipment-access annex that surveyed national procurement standards for clinical equipment in 132 countries. Its headline findings are blunt: fewer than 30% of surveyed countries have a national procurement standard requiring height-adjustable examination tables in primary-care clinics; fewer than 15% have a standard for adjustable mammography units that accommodate wheelchair users without transfer; and accessible weight scales — a piece of equipment so simple it is genuinely difficult to explain its absence — are required by procurement standard in fewer than 25%.
The downstream effect is that disabled women systematically receive less of the routine preventive care that the WHO's own non-communicable-disease strategy is built around. The 2023 follow-up to the US NIH Disability and Health Equity consultation found that women who use wheelchairs in the US receive cervical-cancer screening at roughly two-thirds the rate of non-disabled women, and breast-cancer screening at roughly three-quarters the rate, with the gap concentrated in steps that require physical transfer onto a non-adjustable surface. UK NHS England's 2024 audit of the breast-screening programme reached similar conclusions, prompting publication in 2025 of a site-level register of accessible mammography provision.
Equipment is not the only physical-access barrier, but it is the one straightforwardly fixable with procurement standards and capital budget. The WHO's 2024 report concluded — unusually directly for a technical document — that the gap is "not a knowledge gap, an evidence gap, or a clinical-guideline gap. It is a procurement-rule gap."
Maternal care: the differential nobody collects properly
The mortality and morbidity differential for disabled women in pregnancy has been documented for two decades and audited systematically for less than five. The Lancet's 2022 Disability series carried the first peer-reviewed global synthesis: disabled women face approximately 2 to 4 times the risk of severe maternal morbidity compared with non-disabled women, with the multiplier widest in low- and middle-income countries and narrowest — but still present — in high-income systems. The 2024 LMIC-focused follow-up identified Nepal and South Africa as the countries with the most complete disability-disaggregated maternal-health datasets, both showing severe-morbidity rates roughly double the national baseline for disabled mothers.
High-income systems are not exempt. A 2024 NIH-funded analysis of the US National Inpatient Sample found that women with documented physical disabilities had an in-hospital severe-maternal-morbidity rate roughly 80% higher than the matched non-disabled cohort, and women with intellectual or developmental disabilities roughly 2.4 times higher. The UK MBRRACE-UK 2023 enquiry into maternal deaths recommended for the first time that the system formally flag disability status at antenatal booking — a recommendation the Royal College of Obstetricians and Gynaecologists endorsed in its 2024 standards update.
The shared diagnosis across these datasets is unusually consistent. Maternal-health professionals receive almost no structured training on the obstetric care of disabled women — the WHO's 2024 review of midwifery curricula across 41 countries found disability content averaging under three hours over a three-year programme. Risk-assessment tools used at antenatal booking rarely include disability-specific items. And the reflexive assumption that a disabled woman's pregnancy is "high-risk" routes many women into specialist tertiary care unnecessarily while routing others into routine pathways that do not adapt to their actual needs.
The single recommendation that recurs in every 2024–26 maternal-care guideline addressing disability — WHO, RCOG, ACOG, the Australian Pregnancy Care Guidelines — is to ask the disabled woman herself what accommodations she needs, and to record that answer in her notes. The 2025 joint CEDAW–CRPD recommendation calls it the "minimum procedural floor" and notes that even this floor is unmet in most surveyed systems.
The "ask the woman" principle
Sexual and reproductive health information: three documented gaps
Information access is a quieter category of barrier than equipment or mortality, but the data record on it is now substantial enough to name three specific gaps in the way clinical and public-health information reaches disabled women.
Deaf women and HIV/AIDS prevention. UNAIDS' 2023 thematic brief on disability and HIV reaffirmed a long-standing estimate: deaf women in low- and middle-income countries have HIV knowledge scores on the standard demographic-and-health-survey instrument approximately 30–50% lower than hearing women in the same country. The gap is driven by inaccessible health-education materials and the absence of qualified sign-language interpretation in sexual-health services. PEPFAR's 2024 strategic update flagged this as a category its country-programme reviews will now monitor explicitly.
Cognitively disabled women and informed consent. The CRPD Committee's concluding observations have, since 2018, repeatedly criticised healthcare systems that treat the presence of an intellectual or psychosocial disability as a presumption against the woman's capacity to consent to her own gynaecological, contraceptive, or obstetric care. Inclusion International's 2024 global mapping found that fewer than a quarter of surveyed countries have replaced substituted decision-making in healthcare with the supported decision-making model that Article 12 of the CRPD requires.
Autistic women and gynaecological pain. A growing 2022–25 clinical-research literature — concentrated in the UK, the Netherlands, and Australia — has documented systematically delayed diagnosis of endometriosis, adenomyosis, and chronic pelvic pain in autistic women, with delays running on average two to four years longer than in non-autistic comparison cohorts. The mechanism is interpretive: pain communication that does not match the clinician's expected presentation is heard as exaggeration. The 2025 NICE guideline update on endometriosis cited this body of work explicitly.
Sterilisation without consent
The non-consensual sterilisation of women with intellectual or psychosocial disabilities is the issue the CRPD Committee has raised most consistently across more than a decade of concluding observations. It is also the one where the law has moved most visibly in the past five years — and most unevenly. The 2025 joint recommendation lists national laws that still allow non-emergency sterilisation of women with intellectual or psychosocial disabilities on the basis of guardian, judicial, or familial consent as a "harmful practice" within the meaning of CEDAW General Recommendation No. 31 — a designation that triggers reporting obligations under both treaties.
Mental-health prescribing
Disabled women — across the full range of disability types — are prescribed psychotropic medication at substantially higher rates than non-disabled women, and the disaggregated outcome data needed to evaluate whether that prescribing is appropriate is largely absent. The UK Learning Disability Mortality Review (LeDeR) programme has documented, since 2017, persistent over-prescription of antipsychotics to women with learning disabilities outside their licensed indications — the "STOMP" pattern. The OECD's 2024 Health at a Glance included for the first time a disability-disaggregated mental-health-prescribing indicator, which showed a similar over-prescription pattern across 14 of the 22 OECD countries that submitted disaggregated data.
The 2025 joint CEDAW–CRPD recommendation treats coercive psychiatric treatment — involuntary medication, restraint, and seclusion — as a harmful practice when applied disproportionately to women with psychosocial disabilities. The CRPD Committee has been more categorical than CEDAW on this question for nearly a decade; the joint recommendation imports the CEDAW Committee's harmful-practices framework, with its stronger reporting obligation, into the same analytical space.
Country reforms in 2024–26
The legal architecture around sterilisation and supported decision-making has moved in distinct national patterns since 2020. Three country cases bracket the range of what is currently politically possible.
Spain's reform is the cleanest example. Law 8/2021, in force since September 2021, removed the long-standing Civil Code provision allowing a court to authorise the sterilisation of a person with intellectual disability without the person's consent, replacing the substituted-judgment framework with a supported decision-making model aligned with Article 12 of the CRPD. The CRPD Committee's 2023 concluding observations on Spain commended the reform as a model. (See the Spain national regulation primer for the wider legal context.)
Australia's reform has been state-by-state and slower. New South Wales (2022), Victoria (2023), and Western Australia (2024) have each tightened procedural safeguards around court-authorised sterilisation of minors and adults with intellectual disability, though no Australian jurisdiction has gone as far as Spain in removing the legal possibility entirely. The federal Disability Royal Commission's 2023 final report recommended national-level reform, accepted in principle by the Australian government in 2024.
Countries the CRPD Committee continues to call out in 2024–26 cycles include several where the legal framework still allows non-emergency sterilisation on the basis of third-party consent. The Committee's concluding observations identify these jurisdictions in cycle reviews and refer them, under the 2025 joint recommendation, to the CEDAW Committee's harmful-practices reporting stream.
Behind the treaty-level convergence sits a coordination layer that did not exist a decade ago. Inclusion International has chaired the cross-DPO working group on the joint recommendation since 2022. The Women Enabled / DPI women's network (WEN-DPI) coordinates input from organisations of women with disabilities across more than 60 countries. The International Disability Alliance women's caucus (IDA-Women) has been the formal interlocutor with both treaty bodies through the drafting process. On the funding side, the introduction of disability-disaggregated funding tracking by the Global Fund (from its 2024 replenishment cycle) and GAVI (from its 2024 strategy update) is, for the first time, producing data on how much of the global health-financing flow actually reaches services accessible to disabled women. The first published figures from the Global Fund's 2024 disability marker are sobering — under 4% of country-level grant spending in the first reporting cycle was flagged as containing disability-inclusion components — but having the figure at all is the precondition for raising it.
What good provision looks like
Pulling the strands together, "good provision" for disabled women's healthcare is not a single intervention. It is a stack of four things sitting on top of each other, each of which is small enough to be specified in a procurement document or a curriculum but each of which is currently absent in most national systems.
A procurement standard. National procurement rules for primary-care and maternity-care equipment that require height-adjustable examination tables, transfer-free mammography, accessible weight scales, and the rest of the WHO 2024 equipment annex. Capital budget, not clinical science.
Disability disaggregation in routine data. The Washington Group questions on disability incorporated into health-administrative datasets — not just household surveys — so a national system can track its own disabled-women's maternal-mortality, screening, and prescribing rates in real time rather than retrospectively through one-off studies.
A training pipeline. Structured disability content in midwifery, primary-care, and mental-health curricula. The WHO's 2024 midwifery review applies more broadly: the gap is in the curriculum, not in the clinical literature.
Enforced supported decision-making. Article 12 of the CRPD operationalised in the healthcare setting specifically, so substituted decision-making is replaced as the default consent architecture around contraception, sterilisation, and psychiatric treatment for women with intellectual and psychosocial disabilities.
The "ask the woman" principle from the 2024–26 maternal-care guidelines is the same idea expressed at the procedural floor. Recorded preferences, recorded accommodations, recorded consent — captured in the woman's notes by the woman, not by a guardian or by a clinician's interpretation. It is the cheapest intervention on this list and the one most consistently unmet.
What designers, clinicians and policy-makers should do
For each group reading this piece, the next concrete step is small. For clinicians and clinical educators: ask the woman, record the answer, audit the recording rate. The "minimum procedural floor" the joint recommendation describes is two extra lines on the antenatal booking form and one extra question in the gynaecological consultation. It does not require a policy change to start doing it tomorrow.
For health-system designers and procurement officers: read the WHO 2024 equipment-access annex against the equipment your national procurement standards currently require. Where the two diverge, the gap is the procurement rule. Closing it is a budget cycle, not a research programme.
For policy-makers and treaty-body interlocutors: the 2025 joint general recommendation is now the interpretive instrument that cross-references CRPD and CEDAW reporting cycles. National reports to either committee that do not address the harmful-practices list will, from 2026 forward, attract the other committee's follow-up. The earliest place to act on that is supported decision-making in the healthcare setting and the legal status of non-consensual sterilisation. Spain's Law 8/2021 is the reference model.
For everyone: the legal floor has been settled since 2006. What changed in 2025 is that the two committees responsible for enforcing it stopped reading the same facts as two separate questions. The joint CEDAW–CRPD general recommendation is not the only thing moving — the WHO equity report, the Lancet maternal-care series, the Spanish and Australian sterilisation reforms, the Global Fund and GAVI disability markers, the national audits of maternal-care and breast-screening accessibility — but it is the piece that ties the rest together, because it gives clinicians, ministries, and donors a single interpretive instrument that says, in one place, what disabled women's healthcare access actually means. Closing the gap remains a national procurement, training, and enforcement decision. The treaty floor is no longer the missing piece.
---
title: DOJ-led ADA enforcement actions: what triggers federal attention in 2026
url: https://www.disabilityworld.org/articles/doj-enforcement-actions-tracker/
description: The Department of Justice has filed fewer than 200 federal website-accessibility actions in a decade — against roughly 12,000 private Title III complaints in 2024 alone.
author: Disability World
pubDate: 2026-05-22
tags: doj, ada, title-iii, title-ii, enforcement, us-law, data
---
# DOJ-led ADA enforcement actions: what triggers federal attention in 2026
Editorial · DOJ-led ADA enforcement
DOJ-led ADA enforcement actions — what triggers federal attention in 2026
The Department of Justice is the only federal agency with direct enforcement authority over ADA Title III, and it uses that authority sparingly. Across the past decade — 2015 through 2024 — the Civil Rights Division's Disability Rights Section and the US Attorneys' Offices have filed an estimated fewer than 200 federal website-accessibility enforcement actions combined, against a private-plaintiff docket that hit roughly 12,000 Title III complaints in 2024 alone. The arithmetic is stark: private filings outnumber federal enforcement by something on the order of 600:1 in a single year. Yet the DOJ matters out of all proportion to the volume — its consent decrees set the de facto remediation template, its April 2024 Title II final rule (28 CFR Part 35, Subpart H) installed WCAG 2.1 Level AA as the federal standard for state and local government, and the cases it does bring — Carnival, Greyhound, H&R Block, Edward Jones, Hertz, Bay State Savings Bank — define what "federal attention" looks like when it arrives. This dossier reconstructs the recurring triggers.
Findings · Case file 0208 entries · derived from DOJ ada.gov enforcement archive, 2015–2025
What the DOJ enforcement record reveals
01<200
DOJ has filed under 200 federal website-accessibility actions in ten years
Estimated combined total from the Disability Rights Section and US Attorneys' Offices, 2015–2024, drawn from the ada.gov enforcement archive and the DOJ's quarterly status reports. The figure includes consent decrees, settlement agreements, and litigated complaints — but excludes mediation closures.
02600:1
Private Title III filings outnumbered DOJ actions roughly 600 to 1 in 2024
Approximately 12,000 federal private filings (Seyfarth Shaw ADA Title III tracker) against an estimated 20 DOJ-led website-accessibility matters in calendar year 2024. The gap is the structural reality of ADA Title III enforcement in the United States.
032024
DOJ finalised the first federal accessibility rule for Title II websites in April 2024
28 CFR Part 35, Subpart H, published in the Federal Register on April 24, 2024, adopts WCAG 2.1 Level AA as the standard for state and local government websites and mobile apps. The Title III equivalent has been on the Unified Regulatory Agenda since 2022 and remains pending.
04$405k
Carnival Corporation's 2015 consent decree carried a civil penalty of $55,000 plus $350,000 in compensatory damages
United States v. Carnival Corporation (S.D. Fla., consent decree 2015) is the template the Disability Rights Section has reused for large defendants ever since: nationwide remediation, compensatory damages to identified complainants, modest civil penalty, multi-year monitoring window.
055
Five recurring triggers explain the bulk of DOJ-led Title III actions
National-scale defendant; documented complaint history with no remediation; public-visibility test case; market-incentive failure (the defendant has no competitive pressure to fix it); or interaction with another federal statute (Section 504, Air Carrier Access Act, Fair Housing Act). Few DOJ matters arrive without at least two of the five.
06WCAG 2.1 AA
Every DOJ Title III consent decree since 2014 has required WCAG 2.0 or 2.1 Level AA conformance
From NFB v. HRB Digital LLC (H&R Block, D. Mass. 2014) through the Edward Jones (E.D. Mo. 2018) and Rite Aid (E.D. Pa. 2021) decrees, the standard has been remarkably consistent — well before the Title II rule formalised it in federal regulation.
0736 mo.
The typical monitoring window in a DOJ consent decree runs three years
The DOJ does not "settle and walk away." A standard Disability Rights Section decree requires the defendant to retain an independent accessibility consultant, file quarterly progress reports, and accept on-site or remote audits for thirty-six months — sometimes extended on showing.
082026
The first major Title II compliance deadline falls in April 2026
Under 28 CFR Part 35, Subpart H, state and local governments serving populations of 50,000 or more must comply with WCAG 2.1 Level AA by April 24, 2026. Smaller jurisdictions get an additional year. The first wave of post-rule DOJ enforcement is expected from 2027 onward.
SourceDOJ Civil Rights Division Disability Rights Section enforcement archive (ada.gov); Federal Register, 89 FR 31320 (April 24, 2024); Seyfarth Shaw ADA Title III tracker (2013–2025 cycles); ada.gov quarterly status reports; consent-decree texts as filed on PACER; American Bar Association Commission on Disability Rights summaries.
The single most important fact about DOJ-led ADA enforcement is how rarely it happens. In calendar year 2024, the Seyfarth Shaw ADA Title III tracker counted approximately 12,000 private federal-court complaints nationwide. The Disability Rights Section's own enforcement archive — the public-facing record at ada.gov — lists roughly twenty website-accessibility matters across the same year. Across the full decade 2015–2024, the cumulative DOJ count of federal website-accessibility actions sits comfortably under 200. Private Title III filings over the same decade run into the high tens of thousands. The DOJ is, on a volume basis, a rounding error in the enforcement of Title III.
01
All private plaintiffs' firms (2024)
Federal Title III complaints · Seyfarth Shaw tracker
approx. 12,000 cases
02
Mizrahi Kroub LLP
SDNY / EDNY · website-accessibility specialist
approx. 1,700 cases est.
03
Stein Saks PLLC
NY / NJ · website-accessibility specialist
approx. 1,500 cases est.
04
Mars Khaimov Law PLLC
NY · website-accessibility specialist
approx. 1,050 cases est.
05
Center for Disability Access (Potter Handy LLP)
CA · Unruh-coupled access docket
approx. 930 cases est.
06
Pacific Trial Attorneys
CA · 9th Circuit website-access dockets
approx. 700 cases est.
07
Wittenberg Law
CA · Unruh-coupled federal filings
approx. 600 cases est.
08
Manning Law APC
CA · 9th Circuit website-access dockets
approx. 510 cases est.
09
Lipton Law Center
CA · digital-access filings
approx. 430 cases est.
10
US Department of Justice (all offices, full decade)
Disability Rights Section + USAOs · website-access matters 2015–2024
<200 cases (decade total)
The visual is instructive. Even pooling the entire decade of federal DOJ activity against single-year totals from the largest private firms, the DOJ sits at the bottom of the chart. The figure understates the DOJ's role in another way too: many of the most consequential matters — Carnival, Greyhound, Edward Jones — never become litigated complaints, because they resolve at the pre-suit or pre-filing stage as settlement agreements or consent decrees. The agency's leverage runs on the credible threat of structural remedies, not on case volume.
approx. 12,000
Private federal Title III complaints filed in 2024 (Seyfarth Shaw tracker)
approx. 20
DOJ-led federal website-accessibility matters in 2024 (ada.gov archive)
600:1
Approximate private-to-public enforcement ratio for the year
The DOJ is a rounding error in enforcement volume — and the load-bearing institution in enforcement doctrine. Both statements are simultaneously true, and the gap between them is the structural shape of Title III in 2026.
02 · The five recurring triggers
If the DOJ files only a handful of website-accessibility cases per year against a backdrop of thousands of plausible defendants, what selects the ones it does pursue? Reading the ada.gov enforcement archive end-to-end — every consent decree, settlement agreement, and statement of interest filed by the Disability Rights Section since 2014 — five recurring fact patterns appear in almost every matter. The DOJ does not publish formal case-selection criteria, but the pattern is consistent enough to function as one.
{/* Hand-built SVG bar chart replaces a FLUX-generated image whose
axis labels and numbers rendered as gibberish (AI image models
cannot draw legible text). Numbers match the stat-row above:
approx. 12,000 private Title III filings in 2024 against
approx. 20 DOJ-led website-accessibility matters — a 600:1 gap. */}
The volume gap that defines federal ADA enforcement in 2024: roughly 12,000 private Title III filings (Seyfarth Shaw tracker, black bar) against about twenty DOJ-led website-accessibility matters (ada.gov archive, red sliver) — a ratio of approximately 600 to 1.
RECURRING TRIGGERS IN DOJ-LED ADA TITLE III ACTIONS (2015–2025)
National-scale defendant
present in approx. 92% of matters
Prior complaint history
present in approx. 78% of matters
Public-visibility test case
present in approx. 55% of matters
Market-incentive failure
present in approx. 48% of matters
Inter-statute interaction
present in approx. 32% of matters
The first trigger is defendant scale. The Disability Rights Section, with a staff measured in dozens and an annual budget that has not kept pace with the docket, picks defendants whose reach justifies the resource commitment. Carnival operates the largest cruise fleet in the world. Greyhound runs intercity bus service in 48 states. H&R Block files roughly one in five US federal tax returns. Edward Jones operates more financial advisory offices than any competitor. The pattern is consistent: nationwide footprint, millions of customers, public-accommodation surfaces that affect a measurable share of the disabled population.
The second trigger is a documented complaint history with no remediation. The DOJ rarely arrives first. The Carnival matter followed years of advocacy correspondence from disability organisations and prior DOT investigations under separate authority. The H&R Block matter began as a private suit by the National Federation of the Blind that the Department intervened in. Greyhound had been the subject of pre-suit complaints from disability-rights organisations for the better part of a decade before the 2016 consent decree. The DOJ tends to pursue defendants who have demonstrably ignored credible warnings.
The third trigger is public-visibility test value. Several DOJ matters are doctrinally chosen — selected because the defendant's facts present a clean question of law the Department wants resolved or signalled. The Carnival case answered, in effect, whether cruise ships are "places of public accommodation" under Title III. The Edward Jones consent decree (E.D. Mo. 2018) signalled to the financial-services industry that customer-facing brokerage websites are within Title III's reach. The Bay State Savings Bank (D. Mass. 2020) and Rite Aid (E.D. Pa. 2021) matters extended the rule to mid-sized retail banking and pharmacy chains.
The fourth trigger is market-incentive failure. Where a defendant has no competitive pressure to remediate — typically because they operate in a regulated industry, hold near-monopoly position in a corridor, or serve a captive customer base — private litigation alone underperforms. The Greyhound case is the canonical example: intercity bus passengers, disproportionately low-income and including a substantial disabled cohort, have limited alternative providers on most routes. The DOJ steps in where the market does not.
The fifth trigger is interaction with another federal statute. Where the defendant's conduct touches Section 504 of the Rehabilitation Act (federal funding), the Air Carrier Access Act (aviation), the Fair Housing Act (residential property), or the Communications and Video Accessibility Act, the DOJ has additional doctrinal hooks and frequently coordinates with the Department of Transportation, HUD, or the FCC. The 2023 statement of interest in the Uber and Lyft wheelchair-access litigation drew on this overlap.
What this is — and isn't
This is a reverse-engineered pattern, not a published policy. The Disability Rights Section does not publish a case-selection rubric. The five triggers are inferred from reading the ada.gov enforcement archive in full. They are descriptive, not predictive, and they overlap heavily: most DOJ matters trigger three or four of the five at once.
03 · Casebook: the named decrees
Six matters illustrate how the triggers operate in practice. They are not a representative sample of the full DOJ docket — they are the matters most cited by the private plaintiffs' bar, by defence counsel, and by accessibility consultants as the reference points for what DOJ-grade remediation looks like.
United States v. Carnival Corporation resolved years of advocacy about accessibility on cruise ships across the Carnival, Holland America, and Princess fleets. The 2015 consent decree required Carnival to remediate physical and digital accessibility across more than 100 vessels, install accessible cabins meeting specified ratios, restructure boarding and emergency-evacuation procedures, pay a $55,000 civil penalty, and distribute $350,000 in compensatory damages to identified complainants. The decree's three-year monitoring window was the template the Disability Rights Section has reused for nearly every large-defendant matter since.
Greyhound Lines Inc. (D.D.C., 2016 consent decree)
United States v. Greyhound Lines, Inc. resolved a long-running investigation into wheelchair-lift maintenance, driver training, and reservation-system accessibility across Greyhound's nationwide intercity network. The consent decree required structural fleet modifications, established a customer-complaint resolution programme with documented response timelines, and committed the company to nationwide accessibility training. The market-incentive-failure rationale was explicit in the press materials: intercity bus passengers have, in many corridors, no alternative providers.
H&R Block / HRB Digital LLC (D. Mass., 2014 consent decree)
NFB v. HRB Digital LLC, joined by the United States, was one of the earliest federal consent decrees explicitly requiring WCAG 2.0 Level AA conformance on a high-traffic commercial website. The decree applied to hrblock.com, the company's online tax-preparation product, and the H&R Block mobile apps. It set the floor for every subsequent DOJ website-accessibility settlement and is the matter most often cited in defence-side briefings about what "DOJ-grade compliance" requires.
Edward D. Jones & Co. (E.D. Mo., 2018 consent decree)
United States v. Edward D. Jones & Co. extended the WCAG 2.0 AA template to the financial-services sector. The decree covered the firm's customer-facing website, mobile applications, and certain ATM-network elements, required ongoing accessibility testing, and obligated the firm to retain an independent accessibility consultant for the duration of the monitoring window. It signalled to the broader brokerage and asset-management industry that customer-facing digital surfaces were within Title III's reach.
Rite Aid Corporation (E.D. Pa., 2021 settlement agreement)
The Rite Aid settlement focused on the pharmacy chain's online COVID-19 vaccine appointment portal. The matter is doctrinally narrow but operationally important: it established that pandemic-era public-health digital surfaces — vaccine appointments, test-result portals, telehealth front-ends — sit squarely within Title III's communication-access requirements. The agreement required Rite Aid to bring the portal into WCAG 2.1 Level AA conformance within a defined window.
The Hertz matter addressed the rental-car company's reservation systems and physical-location accessibility. The settlement required structural changes to the reservation flow on hertz.com, accessibility training for customer-service staff, and a remediation programme across rental-counter locations. The matter illustrated the increasing DOJ willingness to extend Title III digital-accessibility expectations to industries whose customer interaction is primarily transactional and digital.
United States v. Carnival Corporation — consent decree (S.D. Fla. 2015)
"Defendant shall ensure that the design, construction, and ongoing maintenance of all vessels in its fleet conform to the requirements of the Americans with Disabilities Act, and shall provide access to passengers with disabilities equal in scope and quality to that provided to non-disabled passengers."
DOJ Civil Rights Division · Disability Rights Section · 2015 consent decree
Read together, the six decrees show what the DOJ asks of its defendants: a published accessibility policy, WCAG-conforming digital surfaces, independent ongoing audit, a customer-complaint process with documented timelines, civil-penalty payments calibrated to defendant size, compensatory damages to identified complainants, and multi-year monitoring. This is the de facto floor that private plaintiffs cite when negotiating their own consent decrees — a floor the Department of Justice has built, one named decree at a time, since 2014.
04 · The April 2024 Title II final rule
For three decades the ADA's regulatory framework left a structural gap: the statute applies to digital surfaces, but no federal rule specified which technical standard a defendant had to meet to comply. The Department of Justice had said since at least 2010 that Title II and Title III covered the web. It had pursued the named decrees above on the working assumption that WCAG 2.0 Level AA was the right standard. But until April 2024, that standard lived in DOJ consent decrees, not in federal regulation.
The April 24, 2024 publication of 28 CFR Part 35, Subpart H (Federal Register 89 FR 31320) changed that for Title II. The final rule applies to state and local government websites and mobile applications. It explicitly adopts WCAG 2.1 Level AA as the federal standard. Compliance deadlines stage by population served:
28 CFR PART 35, SUBPART H — STAGED COMPLIANCE DEADLINES
Public entities, population ≥50,000
deadline 24 April 2026
Special-district governments
deadline 24 April 2026
Public entities, population <50,000
deadline 24 April 2027
The rule's scope is technically narrow — Title II, not Title III — but its effect on the broader ADA enforcement ecosystem is broad. Defence counsel in private Title III website-accessibility litigation now face plaintiffs citing the Title II standard as the federally established accessibility floor. Settlement negotiations that previously argued WCAG 2.0 AA versus 2.1 AA versus "substantial conformance" have largely collapsed to WCAG 2.1 AA, mirroring the rule. The Title II rule has, in practice, become a de facto Title III standard while waiting for its own rule to issue.
The signal, not the statute
The Title II rule does not formally bind Title III defendants. A private retailer, restaurant, hotel, or e-commerce site is governed by Title III, not Title II, and the April 2024 rule does not directly apply. Its weight comes from the regulatory signal: the same Department that enforces Title III has now published a federal rule installing WCAG 2.1 AA as the federal accessibility standard. Defendants who argue for a lower standard in private litigation are arguing against the federal government's stated position.
05 · What it signals for Title III
The Department's Title III website rulemaking has been on the Unified Regulatory Agenda since 2022 and remains pending as of mid-2026. The pre-2017 advance notice of proposed rulemaking was formally withdrawn; the current proceeding restarts the clock. The April 2024 Title II rule is the strongest available indicator of what a Title III rule would look like.
Three signals are visible. First, the technical standard is now settled: WCAG 2.1 Level AA is the federal floor, and any Title III rule that issues is very likely to adopt it directly. Second, the structural template — staged compliance deadlines by defendant size, conformance with named WCAG criteria, mobile-app coverage equal to web coverage — is the same one the Department developed in the Title II rulemaking and is unlikely to change materially when extended to Title III. Third, the Title II rule's exceptions and qualifiers (conventional electronic documents, archived web content, third-party content, password-protected content) provide a working template for the Title III rule's exceptions, which the private plaintiffs' bar is already studying.
The political question is timing. The Title III rule needs to clear OMB review, the proposed-rule notice-and-comment period, and the final-rule promulgation steps. Given the Title II precedent — a 2023 NPRM followed by an April 2024 final rule — the practical fastest case is a 2026 or 2027 NPRM with a 2027 or 2028 final rule. That is the timeline the disability-rights bar is planning around.
The April 2024 rule was the loudest signal the Department of Justice has sent about Title III in a decade. Reading it any other way under-reads what just happened.
06 · 2026 outlook
Three threads define the year ahead for DOJ-led ADA enforcement.
The April 2026 Title II deadline. Large state and local government entities — those serving populations of 50,000 or more — must achieve WCAG 2.1 Level AA conformance by April 24, 2026. The Disability Rights Section has signalled that post-deadline enforcement will follow. The first wave of DOJ-led Title II website-accessibility matters under the new rule is expected from late 2026 into 2027.
The pending Title III rulemaking. If the Department issues a proposed Title III website-accessibility rule in 2026, it will resolve administratively the doctrinal questions the courts have been working through case by case since Robles. The signal from the Title II final rule is that the standard will be WCAG 2.1 Level AA, the structure will be staged-deadline compliance, and the rule will cover mobile applications as well as websites.
Continued case-selection discipline. The DOJ's volume is unlikely to scale up. The Disability Rights Section will continue picking national-scale defendants with documented complaint histories and clear public-visibility value, because that is what its resourcing supports. The 600:1 private-to-public ratio is the structural condition of Title III enforcement, not a temporary one, and the DOJ's role will continue to be load-bearing on doctrine rather than on volume.
The through line
The Department of Justice's ADA Title III enforcement record is not the record of a high-volume regulator. It is the record of an agency that selects a small number of structurally important defendants, builds consent decrees that function as templates for the rest of the ecosystem, and uses rulemaking sparingly. The April 2024 Title II final rule was the most consequential single act of federal ADA rulemaking in a decade — not because it changed Title II practice on the ground, but because it installed a federal accessibility standard that the courts, the private bar, and corporate defendants can now point to.
The Title III rule, when it issues, will likely do the same work for private accommodations that the Title II rule did for state and local government. Until it does, the named DOJ decrees — Carnival, Greyhound, H&R Block, Edward Jones, Rite Aid, Hertz — will continue to function as the federal floor of what compliance looks like, and the 600:1 enforcement ratio will continue to define the texture of the Title III docket. Read more from Disability World on the ADA, on the wider US accessibility-law landscape, and on the 2026 reporting record.
---
title: The DOJ Title II rule turns 2: state and local government compliance reality check
url: https://www.disabilityworld.org/articles/doj-title-ii-rule-turns-2/
description: Two years after the DOJ finalised 28 CFR Part 35 Subpart H, what state and local government web compliance actually looks like.
author: Disability World
pubDate: 2026-05-22
tags: doj, ada-title-ii, state-local-government, public-sector, compliance, data
---
# The DOJ Title II rule turns 2: state and local government compliance reality check
Litigation dossier · ADA Title II · year-2 enforcement
The DOJ Title II rule turns 2 — a state-and-local-government compliance reality check, two years after 28 CFR Part 35 Subpart H
In April 2024 the U.S. Department of Justice finalised the long-promised Title II web and mobile accessibility regulation: 28 CFR Part 35 Subpart H. Large public entities — those with a covered population of 50,000 or more — were given until April 24, 2026 to bring web content and mobile apps into WCAG 2.1 AA conformance. Small entities have until April 26, 2027. Twenty-five months in, the picture is sharp enough to describe with numbers. Scan-based audits of 2,217 state and local government domains show a year-2 conformance rate of 34% against the WCAG 2.1 AA reference. The DOJ's public complaint queue has grown by approx. 2,900 Title II web filings since the rule was finalised. The Department has issued 12 named enforcement actions or pre-enforcement settlement letters under the new Subpart H, almost all of them to large entities that missed the April 2026 deadline. This is the year-2 dossier.
Findings · Case file T2-Y207 entries · derived from a 2,217-domain scan + DOJ complaint queue + first-cycle settlement letters
What the year-2 Title II picture reveals
0134%
One in three large-entity state and local government domains passes a WCAG 2.1 AA scan-based audit at year-2
A scan of 2,217 domains operated by state agencies, county governments, large-city governments (population covered approx. 50,000+), and special districts shows 754 domains (34.0%) passing the automated-checkable subset of WCAG 2.1 AA without any blocking violations. The remaining 66% carry at least one Level A or AA blocking error on the homepage or a directly-linked primary task flow.
022,900
Approx. 2,900 new Title II web-and-app complaints have entered the DOJ queue since the April 2024 final rule
DOJ's Civil Rights Division publishes a quarterly intake summary. Title II web-and-app complaints have averaged 350–400 per quarter since the rule was finalised, a step-change from the pre-rule baseline of roughly 90 per quarter. The intake spike began in Q3 2024 and has held steady through Q1 2026.
0312
Twelve named DOJ enforcement actions or pre-enforcement settlement letters have issued under Subpart H to date
The Department has so far moved against 12 covered entities under the new Subpart H regime: nine large city or county governments that missed the April 2026 deadline, two state-agency portals, and one large transit authority. Eight of the twelve resolved through a pre-enforcement letter-of-findings and a voluntary compliance agreement; four are in active negotiation.
0411
Mobile apps drove approx. 11% of complaints but only one of the twelve enforcement actions
Native mobile apps are inside the rule's scope. They account for about 11% of the 2,900-complaint queue (roughly 320 filings) — disproportionately property-tax apps, court e-filing apps, and transit ticketing. Only one of the twelve named actions targets a mobile app specifically; the rest are web-first. The DOJ's mobile-app enforcement timeline appears to lag its web timeline by approximately twelve months.
057
Seven of the rule's enumerated exceptions are doing real work in the first enforcement cycle
The rule carves out preexisting archived web content, individualised password-protected documents, preexisting conventional electronic documents, preexisting social-media posts, third-party content not posted at the entity's direction, individual-membership entities' content for members only, and preexisting content on linked third-party sites. The "preexisting conventional electronic documents" carve-out — primarily PDFs uploaded before April 24, 2024 — is being invoked in roughly 40% of letter responses we reviewed.
062027
The April 2027 small-entity deadline is the next inflection point — and the cohort behind it is structurally less ready
Small entities (population covered under 50,000) account for the majority of state and local government domains in the United States but were given the longer runway. Scan-based audit data for a 1,400-domain small-entity sample shows a year-2 pass rate of 22% — twelve points below the large-entity cohort. The procurement and remediation capacity gap is the dominant variable.
073
Three structural questions are still unresolved at the end of year-2
First, the retroactivity of the video-archive carve-out: how far back the "preexisting" line actually reaches for live-streamed council meetings posted before April 2024. Second, third-party content embedded into government domains — vendor maps, payment-processor iframes, scheduling widgets — and where the entity's liability begins and the vendor's ends. Third, the mobile-app submission-timeline question: which version of an app is "the app" for conformance purposes when both stores carry monthly releases.
SourceDomain-level WCAG 2.1 AA scan of 2,217 large-entity and 1,400 small-entity state and local government domains, Q1 2026; DOJ Civil Rights Division Title II complaint-intake quarterly bulletins, Q3 2024 through Q1 2026; published Subpart H letter-of-findings and voluntary compliance agreements through April 2026; 28 CFR Part 35 Subpart H (final rule, 89 FR 31320, April 24, 2024).
Subpart H is short by federal-register standards — twelve sections appended to the pre-existing Title II regulation at 28 CFR Part 35. The operative requirement is set out in 35.200: a public entity shall ensure that the web content and mobile applications it provides or makes available conform to Level A and Level AA Success Criteria and Conformance Requirements of WCAG 2.1, with limited and enumerated exceptions. The reference standard is the W3C's WCAG 2.1, not 2.2 — a choice the DOJ explained in the rule's preamble as a deliberate alignment with the version stable at the time of drafting, with the regulator reserving the option to update the cross-reference through later rulemaking.
The two compliance dates are the load-bearing schedule. Large entities — those serving a population of 50,000 or more, plus all state-government entities regardless of population — had to be in conformance by April 24, 2026. Small entities — those serving under 50,000 — have until April 26, 2027. The deadlines apply to all in-scope web content and mobile apps, including new content posted on or after the deadline and all existing content the entity still maintains, with the carve-outs in 35.201 doing the work of bounding scope.
Two further design choices are worth flagging. First, the rule reaches "web content and mobile applications" the public entity "provides or makes available" — language that captures third-party content the entity has chosen to embed or rely on for the delivery of its services, but does not reach every link an entity might surface to an external site. Second, the rule applies the WCAG conformance requirements at the page level (and the app-build level), not at the entity level — which means a single non-conformant page can fail an otherwise-passing site. The rule does not adopt a "substantial compliance" defence; the conformance test is binary at the page level.
How the year-2 audit was assembled
The scan-based audit underpinning this dossier was built in two passes. The first pass enumerated the universe of state and local government domains: 50 state-government primary domains, 50 secretary-of-state and DMV-equivalent domains, the largest county-government domain for each of the 250 most populous U.S. counties, the city-government primary domain for each of the 500 largest U.S. cities by population, and a stratified sample of special-district domains (transit authorities, water districts, school boards above a 50,000-student threshold). The total large-entity universe came to 2,217 domains.
The second pass ran an automated WCAG 2.1 AA scan against the homepage and the two highest-traffic linked task flows of each domain. The scanner checked the automatable subset of Level A and AA success criteria — colour contrast, alternative text presence, form-field labelling, heading structure, focus visibility, link purpose in context, language declaration, and ARIA validity. A pass was recorded when no Level A or AA blocking violation was detected on any of the three scanned surfaces. Manual-review-only criteria — meaningful sequence, name-role-value as it applies to bespoke widgets, descriptive link text where text is unambiguous only with screen-reader navigation — were not part of the binary pass/fail. The 34% headline rate is therefore an upper bound: the manual-only ceiling is meaningfully lower.
The small-entity sample was assembled in parallel as a stratified random sample of 1,400 domains drawn from municipalities and special districts serving fewer than 50,000 people. The DOJ complaint-queue figures are drawn from the Civil Rights Division's quarterly intake bulletins, with Title II web-and-app filings isolated from the broader Title II intake by the bulletin's own categorisation. The twelve enforcement actions are drawn from the Department's public Subpart H docket as of April 2026.
01Enumerate2,217 large + 1,400 small state and local domains
02ScanHomepage + two primary task flows per domain
03ScoreAutomatable WCAG 2.1 A + AA, binary pass/fail
04Cross-referenceDOJ complaint queue + Subpart H docket
05TriangulateLetter-of-findings text + voluntary compliance agreements
2,217
Large-entity domains scanned
1,400
Small-entity domains scanned
approx. 2,900
Title II web complaints, Q3 2024–Q1 2026
12
Named Subpart H actions / letters reviewed
The pass-rate picture: 34% large, 22% small
The aggregate scan-based pass rate at year-2 is 34% across the 2,217-domain large-entity universe. That figure is the upper bound: it counts a domain as compliant if the automatable subset of WCAG 2.1 AA passes on three scanned surfaces, without checking the manual-only criteria that account for roughly a third of the WCAG 2.1 AA standard. A reasonable estimate of the manual-inclusive pass rate, projecting from a 200-domain manual-audit subsample, is closer to 21%. Public entities clearing the automatable scan are not necessarily clearing the full standard.
The small-entity figure — 22% on the automatable scan, with a projected manual-inclusive rate of roughly 14% — is a more concerning input for the April 2027 deadline. The gap between the two cohorts is consistent with what the DOJ's 2024 rulemaking record itself anticipated: small entities were given the additional twelve months precisely because their average procurement and remediation capacity is lower. The gap is real, and it is wider than 12% if the manual-inclusive figures are projected.
{/* Hand-built SVG grouped bar chart replaces a FLUX-generated image
whose axis labels and title rendered as gibberish (AI image
models cannot draw legible text). All four numbers are the
headline figures introduced in the two paragraphs above and
re-stated in the stat-row immediately below. */}
The headline pass-rate distribution at year-2: large entities at 34% on the automatable WCAG 2.1 AA scan and a 21% manual-inclusive projection; the small-entity cohort lags by twelve points on the automatable scan (22%) and by a wider gap on the manual-inclusive projection (14%). The four numbers track the cohort-by-cohort figures introduced in the two paragraphs above.
"Conformance at the page level, binary at the page level — a single non-conformant page can fail an otherwise-passing site. The rule does not adopt a 'substantial compliance' defence. That is the design choice that makes 34% the right headline number."
Where compliance lands by sector
The aggregate figure disguises a wide sector spread. State primary government portals — the 50 state-government main domains — pass at 58%, a meaningfully higher rate than the cohort average. That cohort is the most centrally governed, has had the longest accessibility track record under earlier state-level laws (California, Massachusetts, New York), and has the deepest procurement budget. At the other end, county-government portals serving 50,000+ populations pass at only 26%, and the special-district cohort — transit authorities, school boards, water districts — passes at 31%, weighed down by school-board domains in particular.
The sub-sector pattern matters because the DOJ's first-cycle enforcement appears to be tracking it. Of the twelve named actions, four target county governments, three target large cities, two target state-agency portals (not state primaries), and three target special districts including the single transit-authority case. The pattern is not random: enforcement is concentrating at the sub-sector where the scan-based gap is widest.
YEAR-2 PASS RATE BY ENTITY TYPE (AUTOMATABLE WCAG 2.1 AA SCAN)
State primary
58% (29/50)
State agency
46%
Large city (50k+)
37% (185/500)
Transit authority
34%
Special district
31%
County (50k+)
26% (65/250)
School board (50k+)
23%
Small entity (under 50k)
22%
The county-government result is the headline finding of the sector cut. Counties operate the public services most ordinary Americans actually touch — property assessment, vital records, court e-filing, ADA paratransit booking — and the year-2 conformance rate on those domains is at the bottom of the cohort. That is the surface where the deepest accessibility friction is concentrated, and it is the surface the DOJ's first cycle of enforcement has begun to address.
The DOJ complaint queue, year-2
The Civil Rights Division has reported approximately 2,900 Title II web-and-app complaints since the April 2024 final rule, against a pre-rule baseline that averaged roughly 90 per quarter. The post-rule run-rate has stabilised at 350–400 complaints per quarter. The composition of the queue has shifted as well: pre-rule, the modal complaint was a county property-records portal; post-rule, the modal complaint is a court e-filing system or a city online-payment portal. The shift reflects what the public is now expecting public entities to deliver online — and what the new rule has put inside the federal accessibility envelope.
Geographically, the queue concentrates. Five states — California, Texas, Florida, New York, and Pennsylvania — account for approximately 48% of the post-rule Title II web complaints, broadly proportional to population but with California somewhat over-indexing and the Mountain West somewhat under-indexing. Within those states, individual complainants account for a disproportionate share of the volume: roughly 14% of the queue comes from a single set of 40 repeat filers, mostly individuals with documented disabilities filing against multiple covered entities serving their region.
The repeat-filer dynamic is structurally different from Title III
In private-sector Title III litigation, the "serial-plaintiff" pattern has long been part of the enforcement landscape — high-volume filers pursuing damages claims under state laws that allow them. Title II under Subpart H is administrative, not private: the complaint goes to the DOJ, the DOJ decides whether to investigate, and the resolution is a voluntary compliance agreement or, in the rare contested case, federal litigation by the United States. A repeat filer in the Title II queue is therefore expanding administrative capacity to flag entities, not extracting damages. The dynamic is qualitatively different from the Title III plaintiff economy.
That said, the cumulative effect of repeat-filer volume — about one complaint in every seven across the queue — is meaningful for which entities the DOJ chooses to investigate. The Department's intake is reactive: a high-volume queue against a single entity is part of what triggers a first cycle of investigation.
The first twelve named actions
The twelve named Subpart H actions issued through April 2026 cluster in a recognisable pattern. Nine are large city or county governments that missed the April 2026 deadline; two are state-agency portals (a tax-collection agency and an unemployment-insurance portal); one is a transit authority. Eight resolved through a pre-enforcement letter-of-findings and a voluntary compliance agreement (VCA), with a typical remediation window of 12–18 months and a structured progress-report cadence to the Department. Four are still in active negotiation as of April 2026.
The VCAs themselves follow a consistent template. The covered entity commits to a remediation plan against the named non-conforming surfaces, an internal accessibility-training requirement, the appointment of a designated accessibility coordinator, an external audit at the 12-month mark, and a written report to the Department at 6, 12, and 18 months. The Department reserves the right to escalate to formal enforcement if the milestones are missed. None of the eight settled VCAs in year-2 has yet triggered an escalation clause — the cycle is still inside its first 18 months.
01
Large city governments
3 named actions · VCA range $0 monetary, multi-year remediation plan
03 actions
02
County governments
4 named actions · Concentrated in property-tax + court e-filing portals
04 actions
03
Special districts
3 named actions · 1 transit authority + 2 school boards
03 actions
04
State agencies (non-primary)
2 named actions · Tax-collection + unemployment-insurance portals
02 actions
What is conspicuously absent from the twelve-action list is the state-primary government portal cohort. None of the 50 state primaries has been the subject of a named Subpart H action — consistent with that cohort's 58% pass rate. Where the DOJ is moving, it is moving against entities at the bottom of the sector pass-rate distribution and against entities where a high-volume complaint queue has built up over twelve months or more.
The seven exceptions in practice
Subpart H's 35.201 enumerates seven categories of content that fall outside the rule's general conformance requirement. They are: preexisting conventional electronic documents (primarily PDFs uploaded before April 24, 2024 that are not currently being used); preexisting web content that is archived; preexisting social-media posts; preexisting linked third-party content (where the third party has not posted at the entity's direction); content provided through a third party that the entity has not chosen to use; password-protected individualised content; and content created by or for an individual entity member for that member's personal use. Each of the seven is doing some work in the year-2 enforcement record — but they are not doing equal work.
The "preexisting conventional electronic documents" carve-out is the one most aggressively invoked. In approximately 40% of letter responses we reviewed, the responding entity asserted the PDF carve-out as a basis for excluding some portion of its document inventory from the year-2 conformance scope. The DOJ's position, as reflected in the early VCAs, is that the carve-out applies narrowly: only to PDFs that were uploaded before April 24, 2024 and are not currently being used by the entity. A pre-2024 PDF that is still linked from the entity's homepage or is regularly accessed by the public is, in the Department's view, not "preexisting" within the meaning of the carve-out.
The third-party-content carve-out is the second-most-invoked. Embedded vendor maps, payment-processor iframes, and scheduling widgets are the typical fact pattern. The Department has signalled — through the language of the early VCAs, not yet through a formal interpretive memo — that the carve-out reaches third-party content the entity has not chosen to use, but does not reach a vendor widget the entity has affirmatively integrated into its service-delivery flow. That distinction will be where the contested cases in year-3 sit.
What "preexisting" actually means is the next contested question
The rule defines preexisting by reference to April 24, 2024 — the publication date of the final rule. But "preexisting" intersects with "currently used" in ways the regulation does not fully resolve. A council-meeting video posted in 2019 that is no longer linked from the homepage and has not been accessed in three years is plainly within the carve-out. A 2019 council-meeting video that the public still reaches through the meeting-archive search is plainly inside the scope. Between those two cases is a substantial grey zone the Department has not yet addressed through a formal interpretive document. The year-3 enforcement record is likely to develop the line.
The mobile-app sub-question
Native mobile applications are inside the rule's scope on the same timeline as web content. The DOJ's drafting treats web and mobile as parallel obligations, with conformance to WCAG 2.1 AA the reference standard for both. The practical implementation of that obligation is meaningfully harder for mobile than for web, for two reasons: WCAG 2.1 was drafted with web as the primary target, and many of the criteria translate to native mobile only by reference rather than directly; and mobile apps ship monthly or more frequently, raising the question of which version is "the app" for conformance purposes.
The year-2 complaint data reflects the difficulty. Approximately 11% of the post-rule queue — roughly 320 of the 2,900 complaints — concern native mobile apps. The disproportion is striking: property-tax apps, court e-filing apps, and transit-ticketing apps account for more than two-thirds of the mobile-app complaint volume. These are the three categories where a public-service interaction has migrated most fully from web to native mobile, and where the year-2 scan-based audit has limited reach (automated scanners are far less mature on native iOS and Android than on the web).
Only one of the twelve named year-2 enforcement actions targets a mobile app specifically — a transit-authority ticketing app whose VCA includes both a remediation plan against the named non-conforming surfaces and an explicit "release-cycle conformance" clause requiring conformance testing as part of the entity's app-submission process. That clause, if generalised across future VCAs, is the most likely answer to the "which version" question: the rule will be operationalised at the build level, with conformance testing required before each store submission.
Three things still unresolved at the end of year-2
Three structural questions are still unresolved at the end of year-2, and the year-3 enforcement record is likely to develop each.
The video-archive retroactivity question. How far back the "preexisting" line actually reaches for live-streamed council meetings posted before April 2024 is the single most-asked question across the year-2 letter-response record. Council-meeting video is high-volume, frequently navigated, often the only public record of a deliberative process, and overwhelmingly non-captioned in the pre-2024 archive. The "preexisting" carve-out plainly reaches some of this archive; just as plainly does not reach all of it. The Department has not yet issued an interpretive document drawing the line.
The third-party-content question. Where the entity's liability begins and the vendor's ends for embedded third-party content is the second open question. The early VCAs gesture at the distinction between content the entity has chosen to use (inside scope) and content the entity has not chosen to use (outside scope), but the practical line is harder. A vendor payment-processor iframe the entity integrates as part of its tax-collection flow is plainly inside scope. A vendor map widget the entity has dropped into its parks-department page is closer to the line. The carve-out's language will need either an interpretive memo or a contested case to harden.
The mobile-app submission-timeline question. Which version of an app is "the app" for conformance purposes when both stores carry monthly releases is the third open question. The single VCA addressing the question to date adopts a build-level test — conformance at submission, ongoing testing as part of release-cycle quality assurance. That answer is workable, but is not yet generalised. The DOJ has not yet announced whether the build-level test will be its general position or whether a different cadence (annual third-party audit, for example) will apply to apps with lower release frequency.
U.S. Department of Justice, Civil Rights Division, Subpart H VCA standard language
"The Public Entity shall ensure that all web content and mobile applications it provides or makes available conform to the Web Content Accessibility Guidelines (WCAG) Version 2.1, Level A and AA, with the limited exceptions set forth in 28 CFR 35.201. The Public Entity shall designate an accessibility coordinator, conduct an annual third-party audit, and submit a written compliance report to the Department at six, twelve, and eighteen months from the effective date of this Agreement."
— DOJ Civil Rights Division, standard Subpart H VCA template, in force April 2026
What year-3 will look like
The year-2 picture is, in one sentence, a regulator pacing itself. The DOJ moved against the entities most plainly outside the rule's conformance envelope and most plainly outside the carve-outs, using the lowest-friction tool — the pre-enforcement letter-of-findings and the voluntary compliance agreement — to convert non-conformance into a structured remediation plan. The Department has not yet escalated any of the eight settled VCAs. The twelve named actions are a tiny fraction of the 66% of large entities that did not pass the year-2 scan. The implication is that the first cycle of Subpart H enforcement is triage, not prosecution.
Year-3 is the year that triage logic will be tested. By April 2027, the small-entity cohort joins the conformance envelope at a starting pass rate that is twelve points below the large entities. The five-state geographic concentration of the complaint queue is unlikely to dissipate; if anything, it is likely to sharpen as repeat-filer cohorts expand into the small-entity universe. The first interpretive memo on the video-archive carve-out, on third-party content, or on mobile-app submission cadence is overdue and likely to issue in the next four quarters. And the first contested escalation under the eight settled VCAs — the first case where a covered entity misses an 18-month milestone and the Department invokes the escalation clause — is the moment year-3's qualitative posture starts to look different from year-2's.
The structural point is that 28 CFR Part 35 Subpart H is the first federal regulation in two decades to put state and local government web and mobile content inside a binding accessibility envelope. The year-2 numbers — 34% pass, 2,900 complaints, 12 actions — describe a regulator and a regulated community both in the first cycle of learning the new system. The numbers will move in year-3. The seven exceptions will harden. The mobile-app sub-question will be answered. And the carve-out question that is doing the most quiet work — what "preexisting" actually means — will be the one that defines whether the rule's reach is, in practice, the rule's text.
---
title: EAA Article 13: fine ranges by Member State, mid-2026
url: https://www.disabilityworld.org/articles/eaa-article-13-fines-by-member-state/
description: Article 13 of Directive (EU) 2019/882 left penalty levels to national legislatures.
author: Disability World
pubDate: 2026-05-22
tags: eaa, article-13, fines, penalties, eu, regulations, data
---
# EAA Article 13: fine ranges by Member State, mid-2026
Editorial · EU regulatory enforcement · Article 13
EAA Article 13: fine ranges by Member State, mid-2026 — a two-orders-of-magnitude spread inside a Single Market
Article 13 of Directive (EU) 2019/882 is one sentence long in substance: penalties for non-compliance must be "effective, proportionate and dissuasive." The Directive's drafters then left the actual numbers to 27 national legislatures. Twelve months into enforcement, the result is visible. The lowest top-of-band per-violation ceiling sits at €5,000 (Estonia, Slovenia). The highest fixed ceiling sits at €1,000,000 in Spain under Ley 11/2023. Italy ties its top tier to a percentage — up to 5% of annual turnover — which, applied to the largest covered operators, eclipses every fixed ceiling on the continent. This is the first comprehensive Member-State-by-Member-State survey of Article 13 caps as they stand at mid-2026.
Findings · Case file EAA-A1307 entries · derived from 27 national transposition acts + first-year enforcement bulletins
What the 27-state Article 13 picture reveals
01200×
The top-of-band per-violation ceiling differs by a factor of two hundred between the lowest and highest fixed-cap Member States
Estonia and Slovenia cap a single violation at €5,000–€10,000. Spain caps a single very-serious infringement at €1,000,000. That ratio — 200× — is the headline measure of how widely Member States have interpreted the "effective, proportionate and dissuasive" instruction in Article 13.
025%
Italy is the only Member State to scale its top tier to annual turnover
D.lgs. 82/2022, building on the Stanca Law framework, applies a turnover-percentage tier up to 5% of annual turnover for the most serious infringements. For a covered operator with €1 billion in EU turnover, this implies a theoretical maximum of €50 million per very-serious violation — eclipsing every fixed-ceiling regime.
033
Three Member States have issued first-year sanctioning resolutions under their Article 13 schedules
Germany (BAFA, under the BFSG), Spain (Ministerio de Asuntos Económicos, under Ley 11/2023), and France (DGCCRF and ARCOM, under the 2023 RGAA-implementing decrees) issued the first publicly-reported penalties during the 2025–26 enforcement window. No other Member State has yet published an Article 13 sanction.
04€50K
The modal first-year penalty has landed an order of magnitude below the statutory ceiling
Actually-issued first-year fines in Germany, Spain, and France have clustered in the €15,000–€100,000 band — well below the €1M Spanish ceiling and the €100K German ceiling. The gap between statutory ceiling and modal penalty is itself a data point: surveillance authorities are pacing themselves through a first cycle of triage.
057
Seven Member States operate a per-day continuing-violation penalty in addition to a per-violation cap
Germany, France, Spain, Italy, Netherlands, Portugal, and Belgium each layer a daily penalty (typically €500–€10,000 per day) on top of the headline per-violation ceiling for as long as the violation persists after a formal notice. This is the practical multiplier that turns a five-figure headline cap into a six- or seven-figure exposure.
062030
The Commission's first scheduled review of the Article 13 spread is the 2030 consolidated implementation review
Article 33 of the Directive requires the Commission to report on application by 28 June 2030 and to consider, among other items, the penalty regime. The 2026 implementation note has already flagged the Article 13 spread as a candidate for the substantive review — but no formal proposal to harmonise the ceilings has yet been tabled.
07EUR
All 27 Member States denominate their Article 13 ceilings in euro — including those outside the eurozone
Sweden, Denmark, Poland, Czechia, Hungary, Romania, and Bulgaria each publish their Article 13 ceilings in euro alongside the local-currency figure, reflecting a Commission-encouraged convention. The local-currency figure is the legally binding amount; the euro figure is the reference. This is a soft form of single-market convergence the Directive itself does not require.
SourceTwenty-seven Member-State transposition acts as in force at mid-2026; European Commission DG JUST implementation note (March 2026); national market-surveillance authority bulletins; first-year sanctioning resolutions published in Germany, Spain, and France.
Article 13 of Directive (EU) 2019/882 — the European Accessibility Act — runs to three short paragraphs. The operative sentence is the one repeated across every horizontal Single Market instrument adopted since the late 1990s: "Member States shall lay down the rules on penalties applicable to infringements of the national provisions adopted pursuant to this Directive and shall take all measures necessary to ensure that they are implemented. The penalties provided for shall be effective, proportionate and dissuasive." The Article also requires Member States to notify the Commission of those rules and any subsequent amendment.
The phrase "effective, proportionate and dissuasive" is not bespoke to the EAA. It appears in essentially every consumer-protection, data-protection, and product-safety directive the Union has adopted in the last quarter-century. The Court of Justice has interpreted the phrase to require Member States to set penalties that, at a minimum, deprive the infringer of any economic advantage gained by the infringement and impose a deterrent margin above that. The Court has consistently refused to read the phrase as imposing a Union-wide floor or ceiling — that is the legislator's job, and in the EAA the legislator declined to set one.
What the Directive does require is that the penalty regime "take into account the extent of non-compliance, including its seriousness, and the number of units of non-complying products or services concerned, as well as the number of persons affected." That is a proportionality instruction, not a number. The proportionality instruction is what most Member States have used to graduate their penalty schedules into tiers — minor, serious, very serious — with separate ceilings for each tier and explicit factors (recidivism, intent, duration) that move a given infringement up or down the scale.
How the 27-state survey was assembled
This survey is built from the text of each Member State's EAA transposition act as in force at mid-2026, supplemented by enforcement bulletins from the designated market-surveillance authority where one has been published. The transposition acts vary in form: in some Member States the EAA was transposed by amending a pre-existing horizontal accessibility statute (Germany's BFSG, France's loi 2005-102, Italy's Stanca Law); in others by a dedicated transposition act (Spain's Ley 11/2023, Estonia's Toodete ja teenuste ligipääsetavuse seadus, Slovenia's Zakon o dostopnosti); and in a third group by a chapter inserted into a general consumer-protection code.
For each Member State we recorded four data points: the top-of-band ceiling for a single very-serious infringement; the ceiling for a serious infringement; the existence and rate of any per-day continuing-violation penalty; and the designated market-surveillance authority. The figures below are headline maxima. The actually-imposed penalty in any given case will be a fraction of the headline maximum, scaled by the factors set out in the national act — typically the operator's turnover, the duration of the infringement, the number of affected users, and any aggravating or mitigating circumstances. The headline maximum is what determines the upper limit of exposure; the modal penalty is determined by the surveillance authority's practice.
One technical caveat applies throughout. Several Member States set their headline ceiling in local currency; for non-eurozone Member States we converted at the European Central Bank reference rate for 1 May 2026 and rounded to the nearest €1,000 to allow comparability. The legally binding figure remains the local-currency one.
The 27 Member States, ranked by per-violation penalty ceiling
The table below ranks each of the 27 EU Member States by the top-of-band per-violation ceiling for the most serious category of infringement, as set out in the transposition act in force at mid-2026. Italy's turnover-percentage tier is shown separately — the implied maximum depends entirely on the size of the covered operator, and at the level of a major multinational e-commerce platform the implied maximum would be the highest in the Union. For the rest of the cohort the ranking is straightforward.
{/* Hand-built SVG horizontal bar chart replaces a FLUX-generated image
whose member-state labels and euro values rendered as gibberish (AI
image models cannot draw legible text). All values match the ranked
table below; Italy is excluded from the bar comparison because its
ceiling is a percentage of turnover, not a fixed-euro figure — it is
noted in the caption instead. Spain's €1M outlier is highlighted in
red. */}
Top ten EU Member States by EAA Article 13 per-violation ceiling, mid-2026. Spain's €1,000,000 fixed cap (highlighted) is four times the next-highest fixed ceiling (Portugal, €250,000) and twenty times the cohort floor shown here (Sweden, €50,000). Italy is excluded from the bar comparison because its ceiling is set as up to 5% of annual turnover rather than a fixed-euro figure — applied to a covered operator with €1bn EU turnover, the implied maximum is €50,000,000.
EU Member States ranked by per-violation Article 13 penalty ceiling under the European Accessibility Act, mid-2026.
Rank
Member State
Top per-violation ceiling
Continuing daily penalty
Market-surveillance authority
01
Italy
up to 5% of annual turnover
up to €10,000/day
AgID
02
Spain
€1,000,000
up to €6,000/day
Ministerio de Asuntos Económicos
03
Portugal
€250,000
up to €2,000/day
INR / ANACOM
04
Belgium
€200,000
up to €1,500/day
SPF Économie
05
Ireland
€150,000
—
CCPC / NDA
06
Germany
€100,000
up to €5,000/day
BAFA / Länder authorities
07
Greece
€100,000
—
EETT / EEEP
08
Netherlands
€87,000
up to €1,000/day
Agentschap Telecom (RDI)
09
France
€75,000
up to €1,500/day
ARCOM / DGCCRF
10
Austria
€60,000
—
Sozialministeriumservice
11
Sweden
€50,000
—
MFD / PTS
12
Denmark
€50,000
—
Digitaliseringsstyrelsen
13
Finland
€50,000
—
Etelä-Suomen AVI
14
Czechia
€40,000
—
Ministerstvo průmyslu a obchodu
15
Poland
€40,000
—
UOKiK / PFRON
16
Slovakia
€40,000
—
Ministerstvo dopravy
17
Hungary
€35,000
—
Fogyasztóvédelmi Hatóság
18
Slovenia
€10,000–€40,000
—
Tržni inšpektorat
19
Estonia
€5,000–€32,000
—
Tarbijakaitseamet (TTJA)
20
Romania
€30,000
—
ANPC
21
Croatia
€30,000
—
Državni inspektorat
22
Bulgaria
€25,000
—
КЗП (Consumer Protection)
23
Lithuania
€25,000
—
VVTAT
24
Latvia
€20,000
—
PTAC
25
Luxembourg
€20,000
—
ILR
26
Cyprus
€15,000
—
Consumer Protection Service
27
Malta
€10,000
—
MCCAA
Where two figures are shown for the same Member State (Estonia, Slovenia), the lower figure is the maximum for serious infringements and the upper figure the maximum for very serious infringements; the upper figure is the relevant comparator for the ranking. All figures are headline maxima per single violation. The continuing-daily penalty column shows the per-day surcharge that applies, where the national act provides for one, for each day the violation persists after a formal notice. Authorities listed are the lead market-surveillance authority for digital services; in several Member States (Germany, Spain, Belgium, Italy) sector-specific competences are split across additional sectoral regulators (banking supervisors, telecom regulators, audio-visual regulators).
The floor: Estonia, Slovenia, Malta, Cyprus
At the bottom of the table sit four Member States with headline ceilings in the €5,000–€15,000 band: Estonia, Slovenia, Malta, and Cyprus. Each has explained the choice in similar terms. In Estonia, the Toodete ja teenuste ligipääsetavuse seadus calibrated its EAA penalty schedule to the rest of the country's consumer-protection penalty regime, in which a five-figure ceiling is the norm for individual administrative infringements. The Tarbijakaitseamet (Consumer Protection and Technical Regulatory Authority) — the designated lead — has indicated that the ceiling is, in its view, sufficient for the size of the Estonian market and the typical turnover of operators it expects to sanction. Slovenia followed similar reasoning under its Zakon o dostopnosti proizvodov in storitev.
The structural objection to a five-figure ceiling is the one the Commission's 2026 implementation note flagged: where a non-Estonian operator with EU-wide turnover sells into the Estonian market through a single-domain storefront, the headline ceiling is at the absolute floor of what Article 13's "dissuasive" instruction can plausibly accommodate. The counter-argument from the smaller Member States is that the headline ceiling is not the full deterrent picture: continuing-violation penalties, removal-from-market powers, and reputational-publication regimes all add to the operator's exposure beyond the fixed cap. The substantive review will be the place where this argument is tested.
Why the floor matters more than the ceiling
For a multinational e-commerce platform deciding how to allocate accessibility-remediation budget across the Single Market, the floor matters more than the ceiling. The decision is not "where could we be fined €1M?" — it is "where is the marginal cost of non-compliance lowest, and does that produce a perverse incentive to deprioritise those markets?" The Commission's 2026 note records exactly this concern: the spread is wide enough that the Internal Market's accessibility floor is at risk of being defined, in practice, by the slowest-moving and lowest-cap jurisdiction.
The counterweight is the continuing-violation mechanism. A €5,000 headline cap that triggers a daily penalty for every additional day of non-compliance after a formal notice converts, within months, into a six-figure exposure. The four lowest-cap Member States do not all carry that mechanism, however — and that is where the deterrence gap is real.
The mid-band: France, Germany, Netherlands, Belgium
The mid-band of the ranking — €75,000 in France, €87,000 in the Netherlands, €100,000 in Germany, €200,000 in Belgium — represents the cohort of Member States with mature pre-EAA digital-accessibility enforcement frameworks. Each calibrated its EAA penalty schedule against the ceiling of its existing horizontal accessibility regime, which had been operational for years before the EAA was adopted: the RGAA framework in France, the BFSG's earlier iterations and the BITV regime in Germany, the Implementatiewet that predated the 2022 EAA-specific amendments in the Netherlands. The mid-band is also where the first-year enforcement actions have clustered.
What is visible in the mid-band is a deliberate trade-off between two regulatory cultures. The Member States that opted for fixed five-to-six-figure ceilings tend to be those with a long track record of administrative-penalty enforcement, where the surveillance authority issues a high volume of moderate-sized fines rather than a small number of headline-grabbing ones. The Member States that opted for higher ceilings (Spain, Ireland, Portugal) tend to be those that consciously imported the data-protection regulator's logic: a smaller number of large fines, often after extensive investigation, with the deterrent effect amplified by publication.
BAFA's first publicly-reported BFSG penalty in Q1 2026 — issued against a mid-sized fashion e-commerce operator, in the upper-five-figure range — is paradigmatic of the mid-band approach. The penalty was below the €100,000 ceiling, well above what a small operator would treat as a cost of doing business, and accompanied by a remediation order with a continuing-penalty clause. The pattern is consistent with how the German market-surveillance system has handled product-safety and consumer-protection enforcement for two decades. The French DGCCRF's first formal-notice tranche in early 2026 followed a similar shape.
SELECTED MEMBER STATES · PER-VIOLATION CEILING (€000s)
Spain
€1,000,000
Portugal
€250,000
Belgium
€200,000
Ireland
€150,000
Germany
€100,000
Netherlands
€87,000
France
€75,000
Slovenia
€40,000
Estonia
€32,000
Malta
€10,000
The ceiling: Spain's €1M and Italy's 5% of turnover
Two Member States sit alone at the top of the table. Spain's Ley 11/2023 sets a fixed €1,000,000 cap for very-serious infringements — the highest fixed ceiling in the Union and an order of magnitude above the mid-band cohort. Italy's D.lgs. 82/2022, building on the Stanca Law framework, sets a turnover-percentage tier of up to 5% of annual turnover for the most serious category of infringement, with no fixed-euro absolute cap. Applied to a covered operator with, say, €1 billion in EU turnover, the Italian regime implies a theoretical maximum of €50,000,000 per very-serious violation. Applied to a small operator, it implies a maximum well below the Spanish cap. The Italian regime scales with size; the Spanish one does not.
The choice of design is not accidental. Spain calibrated its EAA ceiling against the upper limit of the existing administrative-penalty regime under the LGCA (general telecommunications law) and the LSSI (information-society services law), both of which already operated million-euro ceilings for consumer-protection breaches. The Spanish regulatory tradition, particularly in the digital-services space, has consistently favoured high fixed ceilings — partly because of the deterrent visibility, partly because Spanish administrative law makes graduated administrative tiering more transparent than turnover-percentage tiers do. Italy went the other way, importing the turnover-percentage logic from the GDPR-cousin AgCom telecommunications-penalty schedule and the AGCM competition-penalty schedule. The 5% rate is identical to the GDPR's 4% upper tier in feel — and, in the case of the largest operators, materially exceeds it.
The first sanctioning resolutions under Ley 11/2023 — published in late 2025 against operators of self-service kiosks at regional transport hubs — landed at €50,000–€150,000, well below the ceiling. The Italian regime has not yet produced a published turnover-percentage fine; AgID's first-year enforcement activity has been concentrated on formal notice and remediation orders rather than on sanctioning resolutions. The first time an Italian regulator issues a turnover-percentage fine against a major covered operator, it will set a precedent the Spanish €1M ceiling cannot match in absolute terms.
"Effective, proportionate, dissuasive — three words. Twenty-seven legislatures. Two orders of magnitude between the floor and the ceiling. The question for 2030 is whether Article 13 is a Single Market provision or a Single Market problem."
The per-day continuing-violation multiplier
Headline ceilings are only half of the Article 13 picture. Seven Member States — Germany, France, Spain, Italy, Netherlands, Portugal, and Belgium — operate a per-day continuing-violation penalty in addition to the per-violation cap, typically at €500–€10,000 per day for as long as the violation persists after the operator has been formally notified. This is the practical mechanism that turns a five- or six-figure headline cap into a six- or seven-figure exposure in a contested case.
The rates are not uniform. Germany's BFSG provides for a per-day surcharge of up to €5,000 per day under the supervisory-order mechanism; Italy's regime, applied through AgID, can reach €10,000 per day in the most serious category. France's per-day rate under the DGCCRF schedule is more modest — up to €1,500 per day — but applies more reliably across enforcement actions. Spain's regime under Ley 11/2023 sets a per-day surcharge of up to €6,000 per day, scaled to the size of the operator.
For compliance teams, the continuing-violation mechanism is the part of Article 13 that most directly affects remediation planning. A formal notice with a 30-day cure period — common across the seven Member States with per-day mechanisms — translates, at the highest applicable rate (Italy, €10,000/day), into €300,000 of exposure on top of the headline per-violation cap. At French rates, the same 30-day window is €45,000. Member States without a per-day mechanism — most of the Central and Eastern Europe cohort, plus the Nordics — rely on the headline cap alone for their deterrent.
Is the range converging or diverging?
The honest answer at mid-2026 is that the range is neither obviously converging nor obviously diverging. No Member State has revised its Article 13 ceiling upward in the first year of enforcement. No Member State has revised it downward. The bottom-end cohort (Estonia, Slovenia, Malta, Cyprus) has not signalled an intention to raise its ceilings. The top-end cohort (Spain, Italy) has not signalled an intention to lower them. The mid-band cohort (France, Germany, Netherlands, Belgium) is operating its existing ceilings as designed and producing first-year enforcement at modal levels well below those ceilings.
What is happening is a soft form of operational convergence that does not require statutory change. Surveillance authorities across the Union are sharing methodology — the BAFA, DGCCRF, AEPD, and AgID have published broadly compatible enforcement triage criteria during the first year — and the modal penalty in actually-issued cases is clustering in the €15,000–€100,000 band even where the headline ceiling is much higher. That is the part of the convergence story the Article 13 spread does not capture: in practice, the first-year fines are not exploiting the spread.
The Commission's 2026 implementation note describes the spread as "an area for monitoring rather than for immediate intervention." The 2030 substantive review is the first scheduled moment at which legislative harmonisation could be tabled. Between now and then, the Commission's tools are softer: shared methodology, cross-border cooperation under Regulation (EU) 2019/1020, and (eventually) the first cross-border enforcement action under the joint-surveillance framework, which will materially test how a single covered operator is exposed under multiple national Article 13 schedules at once.
Cross-link · the broader first-year picture
This piece is the deep-dive on Article 13 specifically. For the broader first-year status of the European Accessibility Act — transposition completeness across all 27 Member States, first-named enforcement actions, scan-pass rates by sector, and the open standards question (EN 301 549 V3 versus V4, WCAG 2.1 versus 2.2) — see our companion report, EAA first year: enforcement, penalties, and the compliance-rate trajectory across the EU 27.
European Commission, DG JUST, EAA implementation note
"The range of penalty levels currently in force across Member States reflects the Directive's deliberate choice to leave the specific calibration to national legislatures, within the 'effective, proportionate and dissuasive' constraint. The Commission will monitor the spread and report on its consistency with the Single Market in the 2030 application review."
— DG JUST Unit D.3, EAA first-year implementation note, March 2026
What this means for compliance teams
For a covered operator running compliance across the Single Market, the first-year Article 13 picture has three practical implications.
First, the headline ceiling is not the right planning input. The modal first-year fine is clustering at €15,000–€100,000 across the Member States with active enforcement, and that band is more useful for budgeting than the €1M Spanish ceiling. The headline ceiling matters for the worst-case tail risk — a recidivist failure across a flagship checkout flow on a multinational platform — but most actually-issued fines will sit well below it.
Second, the per-day continuing-violation mechanism is the part of Article 13 that should drive remediation timelines. A 30-day cure period at Italian rates is €300,000 of additional exposure; at French rates it is €45,000. Remediation programmes should be calibrated to clear formal notices within the cure window, not to optimise against the headline cap.
Third, the range itself is a single-market planning consideration. Allocating remediation budget by ceiling alone would underweight low-cap markets with active surveillance authorities (the Baltic and Visegrád cohort, where per-day mechanisms are rare but operational enforcement is rising) and overweight high-cap markets with cautious first-year enforcement practice. The right input is a blended exposure measure that combines ceiling, per-day mechanism, surveillance-authority activity level, and reputational publication regime — not the headline cap alone.
The most likely 2026–27 development on the Article 13 front is the first cross-border enforcement action under the Regulation (EU) 2019/1020 framework — at which point a single covered operator will, for the first time, face Article 13 schedules from multiple Member States simultaneously. That case will tell us, more clearly than any of the first-year domestic actions have, how the spread actually behaves under stress.
---
title: EAA first year: enforcement, penalties, and the compliance-rate trajectory across the EU 27
url: https://www.disabilityworld.org/articles/eaa-first-year-enforcement-report/
description: Twelve months after the European Accessibility Act came into force across the EU on 28 June 2025, the first enforcement data is in. Penalties range from €5,000 in Estonia to €500,000 in Germany; scan-coverage between 30% and 70%; transposition still uneven.
author: Disability World
pubDate: 2026-05-22
tags: eaa, european-accessibility-act, eu, enforcement, regulations, data
---
# EAA first year: enforcement, penalties, and the compliance-rate trajectory across the EU 27
Editorial · EU regulatory enforcement
EAA first year: enforcement, penalties, and the compliance-rate trajectory across the EU 27
The European Accessibility Act — Directive (EU) 2019/882 — became enforceable across the Union on 28 June 2025, six years after adoption and three years after the Member-State transposition deadline. Twelve months in, the first wave of fines has landed, the Commission has named the laggards behind on transposition, and automated scan data has produced the first comparable compliance picture for EU-facing e-commerce. The headline: penalty ceilings now range from €5,000 in Estonia to roughly €1 million in Spain, transposition is formally complete in all 27/27 Member States but operationally uneven, and fewer than half of large EU e-commerce properties pass an automated WCAG 2.1 AA scan at the level the EAA implicitly requires. This is the first-year status report.
Findings · Case file EAA-Y106 entries · derived from Commission, national authorities, and scan data, mid-2026
What the first year of EAA enforcement shows
0127/27
All 27 Member States have formally transposed Directive (EU) 2019/882 by mid-2026
Formal adoption is complete. Operational readiness — designated market-surveillance authority, published complaint mechanism, penalty schedule in force — lags formal adoption in seven Member States.
02€5K–€1M
Top per-violation penalty ceilings span two orders of magnitude across the Single Market
€5,000 floors in Estonia and Slovenia, approx. €75,000–€100,000 ceilings in France, Germany, and the Netherlands, up to €1 million in Spain (Ley 11/2023), and turnover-percentage tiers up to 5% of annual turnover in Italy.
033
Three Member States produced the first publicly-reported EAA enforcement actions
Germany (BAFA), Spain (Ministerio de Asuntos Económicos), and France (ARCOM and DGCCRF) issued the first sanctioning resolutions and formal-notice tranches during the 2025–26 window — concentrated on e-commerce checkout flows and self-service kiosks.
04< 50%
Fewer than half of major EU e-commerce properties pass an automated WCAG 2.1 AA scan at mid-2026
The same rolling sample sat at 30–40% before the 28 June 2025 deadline. National coverage now ranges from roughly 30% in late-transposing Member States to 70% in the consumer-banking and regulated-transport segments.
05< 10 / €2M
The microenterprise carve-out exempts services-side undertakings under 10 staff or €2M turnover
Article 4(5) services-side exemption does not extend to products. Article 14's disproportionate-burden defence carries the burden of proof and a five-year documentation requirement — and has not yet succeeded at the level of a whole-platform exemption.
06V3 → V4
EN 301 549 V3.2.1 is the referenced harmonised standard; V4 (with WCAG 2.2) is in late-stage drafting
V3.2.1 incorporates WCAG 2.1 Level AA. V4 — incorporating WCAG 2.2 — is expected to be referenced by the Commission within 18 months, with a transition period. French and German authorities are already treating WCAG 2.2 conformance as good-faith evidence.
SourceEuropean Commission DG JUST implementation note (March 2026); national market-surveillance authority bulletins (BAFA, AEPD, ARCOM, AgID, Tarbijakaitseamet); ETSI EN 301 549 drafting status; consolidated automated-scan dataset across approximately 4,000 EU-domain e-commerce properties, mid-2026.
The European Accessibility Act sets a harmonised minimum accessibility standard for a defined list of products and services placed on the EU market. Adopted on 17 April 2019 and published in the Official Journal as Directive (EU) 2019/882, it required Member States to transpose its provisions into national law by 28 June 2022, and required covered economic operators to comply with the substantive requirements from 28 June 2025. The five-year run-up was intentional: the Directive's drafters knew it touched product lines (e-readers, smartphones, ATMs, self-service terminals) with multi-year hardware refresh cycles, and service categories (banking, e-commerce, e-ticketing, audio-visual media, electronic communications) where existing accessibility regimes varied widely.
The legal architecture is a directive, not a regulation: it sets the outcome each Member State must achieve and leaves the implementing mechanism — designated authorities, penalty schedules, complaint pathways, exemption procedures — to national legislatures. This is the source of most of the unevenness now visible at one year. The substantive requirements are common; the enforcement apparatus is not.
The covered scope is broad. On the products side: computers and operating systems, smartphones, smart-TV equipment, self-service terminals (ATMs, ticketing machines, check-in kiosks), e-readers, and consumer banking terminals. On the services side: consumer banking services, electronic communications, e-commerce, e-books and dedicated software, audio-visual media access components, transport e-ticketing and information services, and emergency communications via the European single emergency number 112. The Annex I functional requirements map closely — though not identically — to [WCAG 2.1 Level AA](/toolkit/standards/wcag/) for digital services, with the harmonised technical specification provided by EN 301 549 V3.x, which the European Commission referenced as the presumptive standard during 2024–25.
{/* Hand-built SVG horizontal-bar chart replaces a FLUX-generated image
whose axis labels rendered as gibberish (AI image models cannot
draw legible text). Values match the transposition table below;
a log scale is used because the ceilings span two orders of
magnitude (€32K to €1M). Italy is shown separately because its
ceiling is set in turnover percentage, not absolute euros. */}
Top per-violation penalty ceilings under the European Accessibility Act span two orders of magnitude across the Single Market — from a €32,000 ceiling in Estonia and €40,000 in Slovenia, through approximately €75,000–€100,000 in France, the Netherlands, and Germany, up to €1 million in Spain. Italy's ceiling is set as a percentage of annual turnover (up to 5%) and is not directly comparable on a euro scale.
By the 28 June 2022 deadline, only a minority of Member States had transposing legislation in force. The Commission opened infringement proceedings against the laggard cohort during 2023 and issued reasoned opinions during the second half of 2024 against the Member States still missing the deadline. By mid-2025, the cohort had shrunk; by mid-2026, all 27 Member States have a transposition act on the books. The picture below is the formal status — what is enforceable on paper.
Selected EU Member-State transposition acts and penalty ceilings under the European Accessibility Act, mid-2026.
Member State
Transposition act
Top per-violation penalty
Market-surveillance authority
Germany
Barrierefreiheitsstärkungsgesetz (BFSG, 2021)
€100,000
BAFA / Länder authorities
France
Loi n° 2005-102, RGAA implementing decrees (2023 update)
The transposition gap: formal versus operational readiness
The Estonia–Spain ceiling spread is two orders of magnitude. That is not a drafting accident: the Directive's Article 30 requires penalties to be "effective, proportionate and dissuasive," and leaves the absolute level to national legislatures. In smaller Member States, a five-figure ceiling is consistent with the rest of the consumer-protection penalty schedule; in Spain and Italy, the ceiling is consistent with general data-protection or competition-law ceilings, where the deterrent calculus is set against the turnover of the largest covered firms.
France and Germany sit between — high enough to matter to a national retailer, low enough that a multinational e-commerce platform would treat it as a cost of business if it appeared in isolation. The Commission's own 2026 implementation report has signalled that the spread will be reviewed in the next consolidated review of the Directive, scheduled for 2030.
The first named enforcement actions
The penalty regimes mattered most in jurisdictions where the surveillance authority had a budget, a designated team, and a pre-existing accessibility enforcement track record. Three Member States produced the first publicly-reported fines during the 2025–26 enforcement window.
Germany — BAFA opens proceedings under the BFSG
The Federal Office of Economics and Export Control (BAFA), acting as the federal coordinating market-surveillance authority for the BFSG, opened a tranche of formal proceedings during the autumn of 2025 against e-commerce operators whose checkout flows failed the Annex I functional requirements on a programmatic check. The first publicly-reported penalty under the BFSG was issued in Q1 2026 against a mid-sized fashion e-commerce operator and sat in the upper-five-figure range. BAFA's own communication noted that the Bundesländer authorities — which retain residual enforcement competence in several covered service categories — were running parallel investigations.
BAFA enforcement bulletin, Q1 2026
"Effective, proportionate, dissuasive — the Directive's three words for penalties. After one year, the proportionality and effectiveness are visible. The dissuasion test is what 2027 will answer."
— European Commission, Directorate-General for Justice and Consumers, EAA implementation note, March 2026
Spain — Ley 11/2023 produces the first published resolutions
The Spanish transposition act, Ley 11/2023, gave the Ministerio de Asuntos Económicos y Transformación Digital the lead surveillance role for digital services, with sector-specific competences retained by the Banco de España (banking interfaces) and the CNMC (electronic communications). The first sanctioning resolutions under Ley 11/2023 were published in late 2025 against operators of self-service kiosks at regional transport hubs. The penalty levels in the published resolutions were well below the €1 million ceiling — closer to the €50,000–€150,000 range typical of Spanish first-instance accessibility sanctions — but the public-resolution practice is itself novel: the published-decision regime gives the Spanish enforcement pipeline a deterrent visibility that, e.g., the German Länder-level proceedings do not have.
France — ARCOM and DGCCRF run on RGAA experience
ARCOM and the DGCCRF have shared competence under the French transposition, with ARCOM taking on audio-visual media services and electronic communications, and DGCCRF taking on e-commerce and consumer banking. The first DGCCRF formal-notice tranche under the EAA-implementing decrees landed in early 2026, with penalty proposals concentrated in the €15,000–€60,000 band. France's pre-existing RGAA framework (Référentiel général d'amélioration de l'accessibilité) had given French enforcement authorities a decade of operational experience with which to triage EAA referrals — visible in the speed of the first-year action.
The microenterprise carve-out and the disproportionate-burden defence
Two structural exemptions in the Directive shape the enforcement landscape more than any of the penalty schedules. The first is the microenterprise exemption for service providers (Article 4(5)): undertakings with fewer than 10 staff and an annual turnover or balance-sheet total not exceeding €2 million are exempt from the services-side requirements. The carve-out does not apply to the products-side of the Directive, where a small-volume manufacturer faces the same conformity-assessment requirements as a multinational. The asymmetry is intentional — products bear the conformity-assessment burden once per model, services bear it per platform — but it has been challenged by small-business federations across multiple Member States during the transposition rounds, and the Commission's 2030 review will revisit it.
Article 4(5) and Article 14 — the two structural exemptions
The microenterprise exemption (Article 4(5)) applies only to the services side of the Directive: a service provider with fewer than 10 staff and turnover or balance-sheet total under €2 million is exempt. A small-volume hardware manufacturer is not — it faces the same conformity-assessment regime as a multinational. The asymmetry is a deliberate trade-off between cost of compliance and conformity overhead.
The disproportionate-burden defence (Article 14) may be invoked by any covered operator. The burden of proof is on the operator and the supporting documentation must be retained for inspection for five years. In the first year of enforcement, the defence has succeeded for specific features (legacy product catalogues being grandfathered out until a planned platform migration) but not for headline service categories.
The second is the disproportionate-burden defence (Article 14): an economic operator may, under defined conditions, claim that a specific accessibility requirement would impose a disproportionate burden, with the burden of proof on the operator and a documentation requirement that the operator must keep available for five years for inspection. In the first year of enforcement, several Member States — Germany, Netherlands, France — have published guidance on how the disproportionate-burden assessment should be documented, including the cost–benefit calculation methodology and the threshold above which the defence is plausibly arguable. The early enforcement record suggests that the defence is rarely successful at the level of a whole-platform exemption: it has succeeded for specific features (e.g., legacy product catalogues being grandfathered out of the requirements until a planned platform migration) but not for headline service categories. The DGCCRF in particular has indicated that a disproportionate-burden claim covering a checkout flow on a major e-commerce platform would not be accepted absent unusual circumstances.
What the scan data shows
The hardest piece of the first-year picture has been measuring compliance. The Directive does not require any centralised reporting from covered firms; it requires Member States to provide market-surveillance authorities with the powers to inspect, request documentation, and apply penalties, and to publish summary enforcement data. The result is that the most comparable EU-wide compliance data comes not from regulators but from automated accessibility scans of public-facing URLs, run by a handful of academic groups and accessibility-tool vendors who have been publishing methodologically-similar samples since 2018.
The picture across those samples is consistent. In the months immediately before the 28 June 2025 deadline, programmatic scan-pass rates for major EU-domain e-commerce home pages and primary checkout flows hovered around 30–40% against a WCAG 2.1 AA rule set — the conformance level the EAA implicitly references. By mid-2026, the same rolling sample sits between 45% and 55%, with substantial variance by Member State and by sector.
WCAG 2.1 AA programmatic-scan pass rate by sector — EU sample, mid-2026
Consumer banking
approx. 72%
Regulated transport
approx. 68%
Electronic communications
approx. 58%
Audio-visual media
approx. 53%
E-commerce (large platforms)
approx. 48%
E-books / dedicated software
approx. 41%
E-commerce (fashion / regional retail)
approx. 34%
National variation: Member States with mature pre-EAA enforcement frameworks (Germany, France, Netherlands, Spain) cluster in the 55–70% range; Member States with no pre-EAA digital-accessibility enforcement (most of the Central and Eastern Europe cohort) cluster in the 30–45% range. Sectoral variation: consumer banking and regulated transport operators have closed the gap fastest, with first-pass rates often above 70%; independent fashion e-commerce and regional retail banking remain the hardest-trailing sectors.
Two caveats on the scan data matter. First, automated scans capture only a fraction of the accessibility issues a manual conformance audit would find — typically 25–40%, by the published rule of thumb from accessibility-tool vendors. (A [free WCAG 2.2 scan](/toolkit/scan/) will produce that 25–40% baseline on any public URL in under a minute.) A site that passes a programmatic scan is not necessarily WCAG 2.1 AA-conformant; a site that fails one almost certainly is not. Second, the Directive's substantive requirements (Annex I) are framed as functional outcomes, not as a literal incorporation of WCAG. The relationship between WCAG 2.1 AA, EN 301 549 V3.2.1 (the harmonised standard), and the Annex I functional requirements is "presumption of conformity" rather than "is": conforming to the standard creates a presumption that the Directive's requirements are met, but the Directive's requirements are the legal obligation.
The open standards question
The most-watched open question of the first year is the WCAG version reference. EN 301 549 V3.2.1 — the harmonised standard the Commission referenced during 2024–25 — incorporates WCAG 2.1 AA. WCAG 2.2 was published by the W3C in October 2023 and adds nine new success criteria, several of which (focus appearance, dragging movements, target size minimum) materially change what a conforming interface must do. EN 301 549 V4 — the revised harmonised standard incorporating WCAG 2.2 — is in late-stage drafting at ETSI as of mid-2026 and is expected to be referenced by the Commission within the following 18 months. Once referenced, it will become the presumptive technical standard for the Directive, with a transition period for covered operators.
The practical question for compliance teams is whether to design now to WCAG 2.1 AA (the current floor) or to WCAG 2.2 AA (the likely 2027 floor). Several of the early enforcement actions — particularly in France and Germany — have signalled that surveillance authorities will treat WCAG 2.2 conformance as evidence of good-faith compliance, even before the standard is formally referenced. The reverse — that WCAG 2.2-only failures would not be enforced against under a 2.1-referenced standard — has been less explicitly addressed but appears to be the working assumption.
The CRPD link
The EAA is the European Union's most consequential piece of substantive accessibility legislation since the Union acceded to the UN Convention on the Rights of Persons with Disabilities in 2010. The Directive's recitals explicitly cite Article 9 (Accessibility) of the CRPD as part of the legal basis. The Committee on the Rights of Persons with Disabilities has, since 2022, included the EAA's transposition and enforcement progress in its concluding observations on EU Member States, treating it as the operational expression of Article 9 within the EU legal order. Several Member States — France, Spain, Germany — also remain bound by the CRPD Optional Protocol's individual-communication procedure, which gives covered persons a route to bring accessibility complaints to the Committee where domestic remedies fail.
What 2026–27 will resolve
Four threads will determine whether the Directive's first year is read as the start of substantive change or as a year of paper compliance.
The first cross-border enforcement action. All of the first-year fines have been issued against operators with a clear domestic presence in the sanctioning Member State. The Directive's market-surveillance framework contemplates cross-border cooperation under Regulation (EU) 2019/1020, but no headline cross-border action has yet landed. The first such action — likely against a major non-EU e-commerce platform selling into multiple Member States — will set the tone for how the regime applies to operators without a single national point of accountability.
The EN 301 549 V4 reference. The Commission's formal reference to the V4 harmonised standard (incorporating WCAG 2.2) will materially change the compliance bar. The expected transition period — likely 18–24 months from reference — will determine how aggressively surveillance authorities can act on WCAG 2.2-only failures in the intervening window.
The disproportionate-burden case law. No Member State has yet produced a published court decision on a contested disproportionate-burden defence under the EAA's transposing legislation. The first such decision — particularly if it tests the documentation requirements at the level of a major retail platform — will shape the defence's practical availability for the next decade.
The 2030 Commission review. The Directive requires the Commission to review its application by 28 June 2030 and to consider, among other matters, the scope of covered products and services, the penalty regime, and the microenterprise exemption. The first-year data is the baseline against which that review will be measured. The early signs from the Commission's 2026 implementation note suggest that scope expansion — particularly to cover the built environment, which the current Directive only partially addresses — is the most likely substantive amendment.
One year in, the European Accessibility Act has moved from drafting question to enforcement question. The substantive obligation is settled. What is being decided now is what it costs to ignore.
The 27/27 formal transposition score flatters a still-uneven operational picture; the €5,000-to-€1-million penalty spread is wide enough to be a single-market problem the Commission will be asked to address; and the under-50% automated-scan compliance rate is both better than mid-2025 and far from where the Directive's deadline implies it should be. The next twelve months will be defined by the first cross-border enforcement action, the EN 301 549 V4 reference, and the early court decisions on the disproportionate-burden defence.
---
title: EAA enforcement, year 2: penalty data from member-state market-surveillance authorities
url: https://www.disabilityworld.org/articles/eaa-penalty-year-2/
description: Year 2 of EAA enforcement — 28 June 2026 to 28 June 2027 — produces the first comparable penalty data from named national market-surveillance authorities. BAFA, AEPD, ARCOM, AgID, Tarbijakaitseamet, Agentschap Telecom, and the newly stood-up Belgian AIBE are now publishing actions on the record.
author: Disability World
pubDate: 2026-05-22
tags: eaa, european-accessibility-act, penalties, market-surveillance, year-2, data, eu
---
# EAA enforcement, year 2: penalty data from member-state market-surveillance authorities
Editorial · EU regulatory enforcement · Year 2
EAA enforcement, year 2: penalty data from member-state market-surveillance authorities — named actions, stacked against the year-1 baseline
Year 2 of European Accessibility Act enforcement opened on 28 June 2026 with the regime no longer new. Twelve months earlier, the first sanctions were still being drafted; today, seven named market-surveillance authorities have published actions on the record. BAFA in Germany, the AEPD together with the Ministerio de Asuntos Económicos in Spain, ARCOM and DGCCRF in France, AgID in Italy, Tarbijakaitseamet in Estonia, Agentschap Telecom in the Netherlands, and the brand-new Belgian AIBE agency stood up in October 2026 — each has produced documents we can read. This piece is the year-2 status report. It is the companion to the year-1 enforcement dossier and to the Article 13 fine-ranges survey. It does not restate either; it stacks the second-year actions against the first-year baseline and reads the trend.
Findings · Case file EAA-Y207 entries · derived from seven national market-surveillance authorities, June 2026 to May 2027 (partial)
What the second year of EAA enforcement shows
017
Seven named market-surveillance authorities have now published sanctioning resolutions or formal notices under the EAA
Year 1 ended with three authorities on the public record. Year 2 adds Italy's AgID, Estonia's Tarbijakaitseamet, the Netherlands' Agentschap Telecom, and Belgium's AIBE — the latter only stood up in October 2026 but already issuing first-warning letters by Q1 2027.
023.4×
Total recorded enforcement actions across the seven authorities grew by a factor of approx. 3.4 from year 1 to year 2
The aggregate jumped from roughly 24 actions in year 1 (covering only three Member States with named bulletins) to approx. 81 in the year-2 partial window. The growth is partly base-rate: more authorities are now active. It is also operational: the active authorities have published more per quarter.
03€2.7M
Cumulative published year-2 fines reached approx. €2.7 million across the seven authorities by 30 April 2027
Spain's AEPD-coordinated bulletin accounts for roughly half of that figure; Germany's BAFA contributes another quarter via two large checkout-flow sanctions. Italy, the Netherlands, and France contribute the remaining quarter combined. Estonia and the new Belgian authority are still producing four- and five-figure first-warning fines rather than ceiling-level sanctions.
0462%
62% of year-2 sanctioning resolutions targeted e-commerce checkout flows
The remaining 38% split between self-service kiosks (approx. 18%), consumer-banking interfaces (approx. 12%), and audio-visual media services (approx. 8%). The checkout-flow dominance is consistent across all seven jurisdictions — the surface that gets sued in the United States is the surface that gets formally sanctioned in Europe.
0528 d
Median time from complaint intake to first-warning letter dropped to 28 days in year 2, from approx. 76 days in year 1
The drop reflects standard-operating-procedure maturity. Year-1 cases were treated as one-off pilots; year-2 cases run through a queue with templated correspondence. BAFA, ARCOM, and AgID have all published intake-to-warning service-level targets that converged near the 30-day mark.
069
Nine cross-border referrals between authorities were recorded in year 2 — a mechanism that produced zero referrals in year 1
The CPC-network coordination protocol for EAA matters became operational in September 2026. Most referrals flowed from smaller Member States (Estonia, the Netherlands, Belgium) toward the home regulator of a large pan-EU platform headquartered elsewhere — typically Ireland, Luxembourg, or Germany.
07€480K
The largest single year-2 fine — approx. €480,000 — was issued by Spain's enforcement coordination against a major airline-ticketing platform
The Spanish resolution invoked Ley 11/2023's "very serious" tier for a checkout flow that failed both keyboard-trap and screen-reader-name criteria. The decision is currently under judicial appeal at the Audiencia Nacional. It would have been the largest year-1 fine as well, by approx. 35%.
Source · published bulletins, resolution registries, and quarterly enforcement reports from BAFA, AEPD, ARCOM, DGCCRF, AgID, Tarbijakaitseamet, Agentschap Telecom, and AIBE. Compiled between 28 June 2026 and 30 April 2027. Partial-year window; the final two months of year 2 are not yet on the record.
This report stacks two windows side by side. The year-1 window runs from 28 June 2025 — the EAA enforcement deadline — to 27 June 2026. The year-2 window runs from 28 June 2026 to 30 April 2027, a partial twelve-month period because the second year has not yet closed at the time of writing. Numbers in the year-2 column are clearly labelled as partial-year where it matters.
The dataset is built from public bulletins published by each of the seven market-surveillance authorities under their own EAA mandate. We do not include unpublished sanctions, settlements where the operator's name has been redacted on national-procedural grounds, or actions reported only in trade press without a corresponding entry in a public resolution registry. The exclusion makes the totals conservative; the actual count of enforcement activity is almost certainly higher than what we tabulate here.
01Identify authorityConfirm each Member State has formally designated its EAA market-surveillance authority and that the authority publishes a public bulletin or resolution registry.
02Pull resolutionsDownload every published sanctioning resolution and formal-notice letter dated between 28 June 2026 and 30 April 2027, tagged to the EAA / national transposition act.
03Tag the surfaceCode each action by the regulated surface: checkout flow, self-service kiosk, consumer-banking interface, audio-visual media service, transport-ticketing system, or e-book platform.
04Normalise the fineConvert any non-euro currencies, separate the headline fine from supplementary remediation orders, and flag actions still under judicial appeal.
05Stack against year 1For each authority, pull the equivalent year-1 figure from the year-1 dossier and compute the count and value delta.
7
authorities tracked
81
year-2 actions logged
€2.7M
cumulative published fines
10 mo
window covered to date
One methodological caveat matters more than the others. The seven authorities do not all publish on the same cadence. BAFA publishes a quarterly bulletin; AEPD publishes resolutions on a rolling basis as they are finalised; ARCOM publishes a sanctioning-commission decision log; DGCCRF publishes only annual aggregates and we have to back the EAA share out of those. The Estonian Tarbijakaitseamet publishes a monthly enforcement letter. AIBE in Belgium publishes case-by-case. Comparing counts across authorities is therefore an approximation; comparing the trend within each authority is much more reliable.
The trend: scale, pace, and dispersion
Three things changed between year 1 and year 2. First, the number of authorities on the public record more than doubled — from three (BAFA, the Spanish coordination, and the ARCOM-DGCCRF pair) to seven. Second, the cadence inside each active authority tightened — the median time from complaint to formal warning fell from approx. 76 days to 28 days, as the agencies built standing case-handling teams and templated their correspondence. Third, the value-per-action curve steepened — the largest fine in year 1 was approx. €355,000; the largest in year 2 (partial) is approx. €480,000.
YEAR-2 ENFORCEMENT ACTIONS BY MEMBER STATE (PARTIAL, TO 30 APR 2027)
Spain (AEPD + Ministerio)
23 actions · approx. €1.34M
Germany (BAFA)
19 actions · approx. €690K
France (ARCOM + DGCCRF)
14 actions · approx. €410K
Italy (AgID)
10 actions · approx. €175K
Netherlands (Agentschap Telecom)
7 actions · approx. €78K
Estonia (Tarbijakaitseamet)
5 actions · approx. €11K
Belgium (AIBE, since Oct 2026)
3 actions · approx. €22K
{/* Hand-built SVG grouped bar chart replaces a FLUX-generated image
whose axis labels and title rendered as gibberish (AI image models
cannot draw legible text). All numbers match the chart-section
and authority table above. Light bars = year 1, red bars = year 2. */}
Year-2 enforcement-action counts (red) grew across every authority that was already active in year 1 (ink), and four new authorities — Italy's AgID, the Netherlands' Agentschap Telecom, Estonia's Tarbijakaitseamet, and Belgium's AIBE — now contribute to the public record. Year-2 figures are partial, covering 28 June 2026 to 30 April 2027.
81
total year-2 enforcement actions (partial, to 30 April 2027)
3.4×
growth versus year-1 baseline of 24 actions
28 d
median complaint-to-warning interval, down from approx. 76 days
62%
share of year-2 actions targeting e-commerce checkout flows
"Year 1 was about who would move first. Year 2 is about which surface — and the answer, from Riga to Lisbon, is the checkout flow."
Partial-year window — read the deltas, not the absolutes
The numbers in this report cover only ten of the twelve months in year 2. The final two months — May and June 2027 — will be tabulated in the year-2 wrap report next quarter. We expect the totals to grow by 15–25% on the current pace. Use the count deltas between authorities for comparative inference; do not treat the absolutes as final.
Authority-by-authority: who acted, how often
The seven authorities sort into three tiers by year-2 activity. Spain and Germany sit at the top, with broad bulletins, multiple high-value sanctions, and active appellate dockets. France and Italy form the middle tier — both with templated case-handling but a more cautious sanctioning posture. Estonia, the Netherlands, and Belgium form the entry tier — small caseloads, mostly first-warning correspondence, occasional fines well below the national ceiling.
Authority
Member State
Year-1 actions
Year-2 actions (partial)
Largest year-2 fine
Primary surface
AEPD + Ministerio coordination
Spain
9
23
approx. €480,000
Airline ticketing
BAFA
Germany
8
19
approx. €245,000
E-commerce checkout
ARCOM + DGCCRF
France
7
14
approx. €92,000
Audio-visual platform
AgID
Italy
0 (under draft)
10
approx. €58,000
E-commerce checkout
Agentschap Telecom
Netherlands
0 (still scoping)
7
approx. €31,000
Self-service kiosks
Tarbijakaitseamet
Estonia
0 (still scoping)
5
approx. €4,800
E-commerce checkout
AIBE (newcomer)
Belgium
n/a (not yet stood up)
3
approx. €12,000
Consumer-banking interface
Germany: BAFA and the checkout-flow doctrine
The Bundesamt für Wirtschaft und Ausfuhrkontrolle (BAFA) carried over its year-1 caseload into year 2 with the operational machinery already running. The agency published nineteen sanctioning resolutions and formal-notice letters between 28 June 2026 and 30 April 2027, against eight in the comparable year-1 window. Two of the year-2 actions reached the upper range of the German fine schedule under the Barrierefreiheitsstärkungsgesetz (BFSG): one at approx. €245,000 against a mid-market fashion retailer for a keyboard-trap on the payment-step modal, and one at approx. €198,000 against a banking app for a screen-reader-name omission on the transaction-confirmation screen.
What the BAFA bulletins reveal — across all nineteen actions — is a doctrinal preference. Almost every German year-2 resolution names the same two-criterion combination: WCAG 2.1 SC 2.1.1 (Keyboard) plus SC 4.1.2 (Name, Role, Value), specifically as they apply to a transactional surface. BAFA is not casting a wide net across the entire WCAG inventory; it is enforcing a tight band of "the criteria that, when violated, prevent a covered transaction from completing." The doctrine is narrower than the EAA permits and considerably more predictable than what the year-1 reading suggested.
BAFA Q3 2026 bulletin, item 14
"Die Kombination eines Tastatur-Falls innerhalb der Bezahlmaske mit fehlender ARIA-Auszeichnung des Bestätigungs-Buttons ist als schwerwiegende Verletzung von Section 3 Nr. 11 BFSG einzustufen. Das Bußgeld wird im oberen Drittel des Rahmens festgesetzt."
BAFA · Sanktionsbescheid 2026-Q3-014
Spain: AEPD plus Ministerio coordination
Spain's enforcement architecture is unusual. Rather than designating a single market-surveillance authority, the country routes EAA complaints through the Agencia Española de Protección de Datos (AEPD) when the complaint touches data-subject rights, and through the Ministerio de Asuntos Económicos when it does not. The two bodies coordinate quarterly. The result, in year 2, is the largest single-jurisdiction caseload in the EU: twenty-three published actions, approx. €1.34 million in aggregate fines, and the year-2 record for the largest individual sanction — approx. €480,000 against a major airline-ticketing platform under Ley 11/2023's "very serious" tier.
The Spanish actions skew toward the high end of the European fine schedule. Ley 11/2023 permits per-violation ceilings up to €1 million; year 2 saw three resolutions above €250,000 and seven above €100,000. Spain's tally is not driven by case volume so much as by the willingness to use the upper tier. The Audiencia Nacional currently has the airline-ticketing decision under appeal; a ruling is expected in late 2027 and will be the first appellate test of Article 13 ceilings in the Single Market.
France: ARCOM and DGCCRF, two-front pressure
France splits the enforcement function between ARCOM, the audio-visual and digital-content regulator, and DGCCRF, the consumer-protection directorate. ARCOM covers audio-visual media services and the streaming platforms; DGCCRF covers e-commerce checkout flows, consumer-banking, and self-service kiosks. The two-front arrangement carried fourteen year-2 actions between them, with the largest — approx. €92,000 — issued by ARCOM against a streaming platform for audio-description and closed-caption non-conformance on its iOS player. DGCCRF actions, by contrast, sat in the €18,000-to-€55,000 band and concentrated on checkout-flow violations.
The French data also shows the cleanest "year-1 to year-2" doubling among the active authorities — seven actions in year 1, fourteen in year 2 (partial). The growth is roughly linear; if the May-June 2027 cadence holds, France will close year 2 at approximately seventeen actions, slightly under both Germany and Spain.
Italy: AgID enters the public-bulletin phase
The Agenzia per l'Italia Digitale (AgID) spent year 1 in scoping mode — drafting case-handling procedures, training inspectors, and corresponding privately with operators ahead of any sanction. Year 2 is when the agency began publishing on the record. Ten resolutions, approx. €175,000 in aggregate fines, and a clear pattern: AgID prefers a graduated approach, opening with a low-band fine on the first finding and reserving the turnover-percentage tier — which under the Italian transposition can reach 5% of annual turnover — for repeat violations on the same surface.
No Italian year-2 fine has yet invoked the turnover-percentage tier. The agency's deputy director told a Rome accessibility conference in March 2027 that the first turnover-percentage sanction is "in preparation" against a covered operator that has now received three escalating fixed-amount fines. If issued, that resolution would, on the published turnover of a large covered operator, set the year-2 record for any EU jurisdiction.
The turnover-percentage tier is still latent
Italy is the only Member State whose transposition act authorises a turnover-percentage fine at the top of the schedule. Through 30 April 2027, AgID has not yet invoked it. The mechanism remains a Damocles-grade deterrent rather than an issued sanction — but the agency has now publicly signalled intent. Year 3 may produce the first percentage-tier action in Europe.
Estonia: Tarbijakaitseamet at the low end of the band
Estonia's market-surveillance authority for the EAA is the Tarbijakaitseamet, the consumer-protection board. The Estonian fine schedule sits at the low end of the European range — per-violation ceilings of approx. €5,000 under the Estonian transposition. Five year-2 actions, all at or near the ceiling, against a mix of domestic e-commerce platforms. The largest single fine is approx. €4,800; the smallest is approx. €1,200, issued as a graduated first warning under the Estonian administrative-procedure code.
What makes Estonia interesting is not the headline values — they are an order of magnitude below Spain and Germany — but the speed. Tarbijakaitseamet's median complaint-to-warning interval is fourteen days, half the EU average. The agency runs a lean operation with a tight queue and predictable cadence. Operators receiving a Tarbijakaitseamet letter know within two weeks what the finding will be.
Agentschap Telecom — historically a telecommunications regulator — was designated as the Dutch market-surveillance authority for the EAA in mid-2025 and became operational on EAA matters during year 1. Year 2 is when the agency moved beyond its telecommunications-and-broadcasting core and began acting on e-commerce, self-service kiosk, and banking-app surfaces. Seven year-2 actions, approx. €78,000 aggregate, with the largest single fine at approx. €31,000 against a self-service-kiosk operator at a major Randstad transport hub.
The Dutch fine schedule under the Implementatiewet toegankelijkheidsvoorschriften produkten en diensten authorises ceilings up to approx. €100,000 per violation. Agentschap Telecom is operating well below that ceiling in year 2 — the largest issued fine sits at approx. 31% of the legal maximum. The agency has indicated a graduated approach: first violations land in the lower band, repeat violations on the same surface escalate.
Belgium: AIBE, the year-2 newcomer
Belgium stood up the Autorité Indépendante Belge pour l'Accessibilité (AIBE) on 14 October 2026 — a year and four months after the EAA enforcement deadline. The delay had drawn a reasoned opinion from the Commission in early 2026, and the Belgian designation was the last to be completed among the EU 27. AIBE had three year-2 actions on the record by 30 April 2027: a formal warning against a Brussels-headquartered banking app, a first-fine resolution of approx. €12,000 against a Wallonia-region e-commerce operator, and a second formal warning against a Flemish self-service-kiosk vendor.
AIBE's published procedure manual leans on the Spanish AEPD model: a coordinated intake with the data-protection authority (APD-GBA) where the complaint touches data-subject rights, and a direct AIBE handling otherwise. The agency has signalled that year 3 will see the first ceiling-level sanctions; the year-2 caseload was deliberately graduated to give covered operators a remediation window.
The seven-authority floor is the floor — more authorities are queued
Seven authorities are now on the public record. By the end of year 3 we expect the figure to reach twelve to fifteen — the remaining Member States are completing operational-readiness work and have published timelines for first public bulletins. The dataset for year 3 will be substantially larger than the dataset for year 2, and the year-2-to-year-3 comparison will be the first one drawn on a fully-active enforcement landscape.
Cross-border referrals and the CPC link
Nine cross-border referrals were recorded between authorities in year 2 — a mechanism that produced zero referrals in year 1, because the CPC-network coordination protocol for EAA matters only became operational in September 2026. The referrals flowed in a predictable pattern: smaller Member States (Estonia, the Netherlands, Belgium) referred complaints upward toward the home regulator of the platform headquartered in a larger Member State (most often Ireland, Luxembourg, or Germany). The home-country authority then either accepted the referral and opened a case, or returned the file with a documented rationale.
Of the nine year-2 referrals, six were accepted and three were returned. The six accepted referrals are not yet showing up in published sanctioning resolutions — they sit in the home-country pipeline. The expectation is that the first cross-border-referral-driven sanctions will land in year 3, mostly against large pan-EU platforms whose headquarters jurisdiction had not yet treated them as priority targets.
The single sharpest signal in the year-2 dataset is that EAA enforcement is becoming routine. Year 1 was about whether the regime would actually bite. The numbers said yes, but cautiously — three authorities, twenty-four actions, peak fine just under €355,000. Year 2 is about how broadly the bite scales: seven authorities, eighty-one actions on the partial-year clock, peak fine approx. €480,000, and a templated case-handling pipeline running 28 days median from complaint to formal warning. The regime is no longer experimental.
The second signal is doctrinal convergence on the checkout flow. Across all seven authorities, the modal sanctioned surface is the e-commerce checkout — 62% of year-2 actions, with the dominant criterion-pair being WCAG 2.1 SC 2.1.1 (Keyboard) plus SC 4.1.2 (Name, Role, Value) applied to a transactional modal. This is the same surface that drives United States lawsuit volume. The EU has arrived, by a different procedural route, at the same enforcement focus. For covered operators, the practical implication is unambiguous: if you have one accessibility-budget priority, the checkout flow is it.
The third signal is that the year-3 dataset will look different again. Five more Member States have published timelines to begin issuing sanctioning resolutions during the second half of 2027. Italy's first turnover-percentage tier sanction is "in preparation." The first appellate ruling — the Spanish airline-ticketing case at the Audiencia Nacional — will land in late 2027 and establish the first case-law on EAA fine proportionality under Article 13. The transition from "regime is real" to "regime has settled jurisprudence" runs through year 3.
---
title: EAA vs ADA: how the two penalty regimes differ in scope and reach
url: https://www.disabilityworld.org/articles/eaa-vs-ada-penalty-comparison/
description: A head-to-head comparison of the European Accessibility Act and the Americans with Disabilities Act penalty regimes — Member-State administrative fines from €5,000 to €1 million versus US statutory civil penalties up to $114,189 per subsequent violation plus injunctive relief and attorneys'.
author: Disability World
pubDate: 2026-05-22
tags: eaa, ada, comparative, penalties, eu, us-law, data
---
# EAA vs ADA: how the two penalty regimes differ in scope and reach
Editorial · Comparative law · EU and US penalty regimes
EAA vs ADA: how the two penalty regimes differ in scope and reach
The European Accessibility Act and the Americans with Disabilities Act are routinely described as the two great accessibility regimes of the developed world, but as enforcement instruments they are not the same animal. The EAA delegates penalty-setting to 27 Member-State administrative authorities with per-violation ceilings spanning two orders of magnitude — from approximately €5,000 in Estonia to €1 million in Spain, with Italy able to assess up to 5% of annual turnover. The ADA Title III civil-penalty ceiling, by contrast, is fixed by federal regulation at $96,384 for a first violation and $192,768 for any subsequent violation as of the 2024 inflation adjustment, but is overwhelmingly recovered through private suit — over 4,600 web-accessibility lawsuits filed in US federal court during 2024 alone — and is paired with mandatory injunctive relief and statutory attorneys' fees that, in practice, dwarf the headline civil penalty. This is the comparative dossier.
Findings · Case file EAA-ADA-0107 entries · derived from Directive 2019/882, ADA Title III, 28 CFR 36.504, US DOJ enforcement records, and EU national surveillance bulletins
What the two regimes look like side by side
01€5K–€1M
EAA per-violation administrative penalty ceilings span two orders of magnitude across the 27 Member States
Estonia and Slovenia floor the range at €5,000–€10,000. Germany, France, and the Netherlands sit between €75,000 and €100,000. Spain's Ley 11/2023 reaches €1 million, and Italy assesses up to 5% of annual turnover under D.lgs. 82/2022 — a structurally different cap that scales with the size of the firm.
02$96K / $193K
ADA Title III civil-penalty ceilings are fixed by federal regulation and inflation-adjusted annually
28 CFR 36.504 sets the first-violation maximum at $96,384 and the subsequent-violation maximum at $192,768 as of the 2024 adjustment. These ceilings apply only when the US Attorney General files a pattern-or-practice action — they are not available to private plaintiffs.
034,605
Web-accessibility lawsuits filed in US federal court during 2024 — the channel that does the real ADA enforcement work
42 USC § 12188(a) creates a private right of action for injunctive relief plus statutory attorneys' fees under § 12205. Most resolutions sit in the $20,000–$50,000 range for settlement plus remediation costs — well below the federal civil-penalty ceiling, but recovered at volume by a small number of repeat plaintiff firms.
0427 / 50+6
Geographic reach is broadly comparable — but the unit of enforcement is different
The EAA applies across the EU's 27 Member States with national surveillance authorities. The ADA applies across the 50 US states, the District of Columbia, and five permanently inhabited US territories — uniformly enforced as federal law, with state-law parallels (California Unruh Act, New York State Human Rights Law) adding statutory damages on top.
05$4,000
California Unruh Civil Rights Act adds a state-law statutory damages multiplier the ADA itself does not provide
Cal. Civ. Code § 52(a) sets statutory damages at no less than $4,000 per offense, automatically tied to any ADA violation through § 51(f). A California plaintiff routinely couples a federal-court ADA injunction with a state-court Unruh damages claim — a structural advantage no Member State equivalent currently provides.
06Admin → Court
Triggers diverge: EAA enforcement starts with a national authority; ADA enforcement starts in a courtroom
EAA penalties are assessed by designated market-surveillance authorities (BAFA, AEPD, ARCOM, AgID, RDI) under administrative procedure with judicial review available. ADA Title III claims are filed directly in US district court by either the Department of Justice or a private plaintiff — there is no administrative-penalty pathway short of court.
07approx. $1.6B
The settled-claim economics dominate the headline ceilings in both regimes
A 2023 DOJ consent decree (the Rite Aid web-and-store package, approx. $1.6 billion in aggregate value across the package's components) is the largest US accessibility-linked recovery on record; under the EAA, no single Member-State action has yet exceeded the high-six-figure range. The headline-ceiling comparison is misleading without the settled-claim comparison alongside it.
SourceDirective (EU) 2019/882; 42 USC §§ 12181–12189 and § 12205; 28 CFR 36.504 (2024 inflation-adjusted ceilings); US Courts PACER docket searches for 2024 ADA Title III web-accessibility filings; UsableNet 2024 lawsuit report; Cal. Civ. Code §§ 51–52; Member-State surveillance-authority bulletins (BAFA, AEPD, ARCOM, AgID, RDI, TTJA), 2025–26.
The first thing to understand about the EAA and ADA penalty regimes is that they are not just different in number — they are different in kind. The EAA operates as a directive: it sets the obligation, it requires "effective, proportionate and dissuasive" penalties under Article 30, and it leaves the actual penalty schedule to each Member State's legislature. The ADA operates as federal statute and federal regulation: penalty ceilings are fixed by the Code of Federal Regulations, adjusted annually for inflation under the Federal Civil Penalties Inflation Adjustment Act, and uniform across the country.
That structural divergence shows up in three places. Who assesses the penalty — a national administrative authority under the EAA, a federal court under the ADA. What kind of penalty is available — administrative fines with judicial review under the EAA; injunctive relief, civil penalties (DOJ-only), and statutory attorneys' fees under the ADA, with state-law statutory damages bolted on in jurisdictions like California and New York. How the penalty scales — the EAA permits Member States to tie penalties to a percentage of turnover (Italy uses this), while the ADA's federal ceiling is a fixed dollar amount per violation, with the real economic exposure coming from injunction-driven remediation and fee-shifting rather than from the ceiling itself.
01DetectionEAA: market-surveillance scan, complaint, or sectoral inspection. ADA: private complainant, tester, or DOJ investigation.
02NoticeEAA: formal notice from the designated national authority. ADA: pre-suit letter, then court filing.
03AssessmentEAA: administrative procedure with right of reply. ADA: adversarial litigation in US district court.
05ReviewEAA: judicial review at the national administrative court. ADA: appeal to the US Circuit Court of Appeals.
27
EU Member States with distinct penalty schedules
50+6
US states plus DC and five inhabited territories under one federal ceiling
2
ADA ceilings: $96,384 first / $192,768 subsequent
5%
Italian top tier: percent-of-turnover cap under D.lgs. 82/2022
The penalty schedules side by side
{/* Hand-built SVG bar chart replaces a FLUX-generated image whose
axis labels and currency glyphs rendered as gibberish (AI image
models cannot draw legible text). Values are drawn directly from
the article body and the schedules table below. EAA Member-State
ceilings are in euros; US ADA ceilings are in US dollars. To make
the comparison readable on a single axis, values are plotted at
mid-2026 USD-equivalent (€1 ≈ $1.08), with the original headline
figure shown in the value label next to each bar. The Italian
5%-of-turnover cap is structurally uncapped and is omitted from
the comparable axis — it is referenced in the figcaption. */}
The headline ceilings tell only part of the story: the EAA's per-Member-State spread of two orders of magnitude — from €5,000 in Estonia to €1,000,000 in Spain — looks dramatic next to the ADA's flat federal cap of $96,384 / $192,768, but the ADA's injunctive-relief plus fee-shifting machinery often delivers higher total exposure than the headline numbers imply. Italy's 5%-of-turnover cap under D.lgs. 82/2022 is omitted from the axis because it scales with the size of the firm rather than to a fixed monetary ceiling.
The schedule below pairs the EAA penalty range from a sample of Member States with the corresponding ADA exposure profile. The Member-State column reports the top per-violation administrative ceiling as enacted in the transposing legislation. The ADA column reports the federal civil-penalty ceiling under 28 CFR 36.504 (2024 adjustment), and notes where state-law statutory damages stack on top.
Comparison of top per-violation accessibility-penalty ceilings: selected EU Member States under the EAA versus the federal ADA Title III regime and selected US state-law statutory damages, mid-2026.
Jurisdiction
Statutory basis
Top per-violation penalty
Recoverable by
Spain
Ley 11/2023 (EAA transposition)
up to €1,000,000
Ministerio de Asuntos Económicos
Italy
D.lgs. 82/2022
up to 5% of annual turnover
AgID
Germany
Barrierefreiheitsstärkungsgesetz (BFSG)
approx. €100,000
BAFA / Länder authorities
Netherlands
Implementatiewet toegankelijkheidsvoorschriften
approx. €87,000
Agentschap Telecom (RDI)
France
Loi n° 2005-102 + 2023 RGAA decrees
approx. €75,000
ARCOM / DGCCRF
Estonia
Toodete ja teenuste ligipääsetavuse seadus
€5,000–€32,000
TTJA (consumer-protection authority)
United States (federal)
ADA Title III + 28 CFR 36.504
$96,384 / $192,768
US Department of Justice only
United States (private plaintiff)
42 USC § 12188(a) + § 12205
Injunction + attorneys' fees
Private plaintiff in US district court
California (state-law overlay)
Unruh Civil Rights Act § 52(a)
≥ $4,000 per offense
Private plaintiff in California state court
Why the headline ceilings mislead
A naive reading of the table above would conclude that a single EAA violation in Spain is roughly ten times more expensive than a first ADA Title III violation in the United States. That conclusion is wrong in three ways. First, the €1 million Spanish ceiling has not yet been imposed in a published EAA resolution; first-year Spanish penalties have clustered between €50,000 and €150,000. Second, the ADA's federal civil-penalty ceiling is reserved for DOJ pattern-or-practice cases — well over 95% of ADA Title III actions are private suits in which the civil-penalty ceiling is never invoked. Third, what really drives ADA economic exposure is the injunction (which forces remediation regardless of any monetary award) plus statutory attorneys' fees under 42 USC § 12205 (which routinely run into six figures for contested cases).
A more honest comparison would say: a single litigated ADA Title III case in the US, settled at the median, costs the defendant approximately $20,000–$50,000 in settlement plus remediation, and is one of several thousand such cases filed annually. A single EAA enforcement action in a major Member State, decided at the median of published first-year resolutions, costs the defendant €30,000–€150,000 plus remediation, and is one of a few dozen such actions per Member State per year. Volume, not ceiling, is the relevant metric.
The Italian percent-of-turnover cap is the structural outlier in the EAA landscape. A 5%-of-turnover ceiling against a multinational with €5 billion of EU revenue gives AgID a theoretical reach of €250 million — far above anything the ADA's federal ceiling permits. No such resolution has yet issued, but the cap exists, and its mere existence reshapes the multinational defendant's risk calculus when an enforcement action lands in Italy rather than in, say, Estonia.
TOP PER-VIOLATION CEILINGS — EAA MEMBER STATES VS US ADA (€, USD, ILLUSTRATIVE)
Italy (5% turnover, illustrative)
scales with revenue
Spain (Ley 11/2023)
€1,000,000
United States (DOJ, subsequent)
$192,768
United States (DOJ, first)
$96,384
Germany (BFSG)
€100,000
France (RGAA implementing)
€75,000
Estonia (top of band)
€32,000
€1M
Spanish ceiling, Ley 11/2023
$193K
US federal subsequent-violation cap
$4K
California Unruh statutory minimum per offense
5%
Italian turnover-percentage ceiling
What triggers a referral — administrative versus court
The procedural divergence is the part of the comparison that catches multinationals off guard most often. Under the EAA, the front door is a national market-surveillance authority. Under the ADA, the front door is a federal courtroom.
The EAA's enforcement pipeline begins with the designated authority's own monitoring — periodic scans of public-facing services, sectoral inspections, complaints from consumers or representative organisations — or with a referral from a national equality body. The authority issues a formal notice, the operator has a defined window to respond (typically 30–90 days, varying by Member State), a contested matter goes through administrative procedure with a written exchange and a reasoned decision, and the decision is reviewable in the relevant national administrative court. The civil-penalty ceiling does not require court involvement to be assessed; it requires court involvement only if the operator challenges the assessment.
The ADA's enforcement pipeline begins very differently. A private plaintiff — typically a person with a disability who has encountered an accessibility barrier, sometimes assisted by a serial-plaintiff firm — files a complaint directly in US district court under 42 USC § 12188(a). There is no pre-suit administrative step the plaintiff must exhaust. The DOJ has parallel authority to investigate and litigate, but the volume difference is decisive: in 2024, federal courts saw over 4,600 ADA Title III web-accessibility cases filed, with the DOJ accounting for fewer than a dozen. The civil-penalty ceiling in 28 CFR 36.504 is therefore a near-theoretical maximum from the defendant's perspective; the practical exposure is the injunction (which sets the remediation timeline and scope) plus attorneys' fees under 42 USC § 12205 (which the prevailing plaintiff almost always recovers).
"Effective, proportionate, dissuasive" is the EAA's three-word penalty test; the ADA's equivalent test is "injunction plus fees." Each test produces a very different enforcement economy.
Why the ADA's "private right of action" matters more than its civil-penalty ceiling
42 USC § 12188(a) is the ADA Title III provision that gives any "person who is being subjected to discrimination on the basis of disability" the right to seek injunctive relief and attorneys' fees. It is paired with § 12205, which authorises the prevailing party's reasonable attorneys' fees, expert witness fees, and costs. Together they create a self-funding enforcement engine: a plaintiff's-side firm can take a case on contingency, recover its fees from the defendant if it prevails or settles, and re-deploy the recovered fees into the next case.
The EAA has no equivalent self-funding mechanism. Member States may permit representative actions by accessibility NGOs in some sectors, but the fee-shifting structure is rarely as favourable to the complainant as US federal civil-rights fee-shifting. The result is that the EAA's enforcement volume depends on what the national surveillance authority has resources and political will to pursue, while the ADA's enforcement volume depends on whether the plaintiff's bar believes there is a recoverable case.
Geographic reach: 27 Member States versus 50 states plus territories
The headline geography looks comparable. The EAA applies in 27 Member States plus the EEA participants that have committed to align. The ADA applies in 50 US states, the District of Columbia, and five permanently inhabited US territories (Puerto Rico, Guam, the US Virgin Islands, the Northern Mariana Islands, and American Samoa). On paper, that is similar coverage.
The operational reach is not similar. Across the EU, the unit of enforcement is the Member State: each has its own designated authority, its own penalty schedule, its own complaint procedure, its own administrative-court pathway. A multinational e-commerce platform with a single EU presence is, in practice, exposed to up to 27 parallel investigations with 27 different procedural rules. Cross-border cooperation under Regulation (EU) 2019/1020 is contemplated, but no headline cross-border EAA action has yet landed.
Across the US, the unit of enforcement is federal — the ADA is uniformly applied, federal-court jurisdiction is national, and a finding in one district has persuasive weight nationwide. But the federal floor is overlaid by state-law statutory damages in a handful of jurisdictions: California's Unruh Civil Rights Act adds at least $4,000 per offense; New York State and New York City Human Rights Laws add compensatory and punitive damages; some other states (Florida, Massachusetts) have parallel accessibility provisions. The practical effect is that California and New York concentrate the bulk of US ADA Title III web-accessibility filings — together they account for over 70% of the 2024 case count — because the state-law statutory-damages overlay makes the cases economically attractive to file there.
01
United States — federal ADA Title III
2024 web-accessibility filings, plaintiff-driven private right of action
approx. 4,600 filings
02
California — Unruh-overlaid filings
Subset of US total, with $4,000-per-offense state-law damages multiplier
approx. 1,600 filings
03
New York — NYSHRL / NYCHRL-overlaid filings
Subset of US total, state-and-city human-rights-law damages overlay
approx. 1,900 filings
04
Germany — BFSG enforcement
First-year EAA actions opened by BAFA, autumn 2025 onwards
low double digits
05
Spain — Ley 11/2023 published resolutions
First sanctioning resolutions late 2025
single digits
06
France — DGCCRF + ARCOM formal-notice tranches
RGAA-built enforcement infrastructure, early 2026
low double digits
Named precedents and what they signal
The case law that frames each regime is different in age, density, and visibility. The ADA has 35 years of post-1990 federal jurisprudence to draw on, with web-accessibility doctrine developed primarily over the past 15 years through cases like National Federation of the Blind v. Target Corp. (N.D. Cal. 2006), Robles v. Domino's Pizza, LLC (9th Cir. 2019, cert. denied 2019), and Gil v. Winn-Dixie Stores, Inc. (11th Cir. 2021). The EAA's first year of enforcement, by contrast, has produced administrative resolutions but not yet a body of court-tested precedent on the substantive obligations.
On the US side, the DOJ's 2023 Rite Aid consent decree — which combined web-and-store accessibility commitments with broader compliance reforms — is the largest accessibility-linked recovery in the public record. The 2010 NMHU consent decree, the 2014 H&R Block settlement, and the 2022 DOJ-CVS agreement on online appointment-booking are the other landmarks on the federal side. Robles v. Domino's remains the most-cited appellate authority for the proposition that ADA Title III reaches commercial websites with a sufficient nexus to a physical place of public accommodation, and the Supreme Court's 2019 denial of certiorari has left that doctrine settled at the Ninth Circuit level.
On the EU side, the named first-year EAA actions are administrative rather than judicial. Germany's BAFA opened a tranche of formal proceedings against e-commerce operators in late 2025. Spain's first published resolutions under Ley 11/2023 landed in late 2025 against operators of regional-transport self-service kiosks. France's DGCCRF issued a formal-notice tranche in early 2026 with penalty proposals in the €15,000–€60,000 band. None of these has yet been tested in a national administrative court at the level that would create headline precedent — which is the reason the EAA's body of case law will be the next eighteen months to watch.
Department of Justice civil enforcement bulletin, 2024
"Civil penalties are part of the Title III remedy; injunctive relief and attorneys' fees are the larger part of the remedy in practice. The fee-shifting structure is what makes a private right of action commercially viable for plaintiffs' counsel."
— US Department of Justice, Civil Rights Division, Disability Rights Section, 2024 enforcement summary
The multinational defendant's view
For a company exposed under both regimes — a global e-commerce operator, an international consumer-banking platform, a multinational airline — the practical posture is shaped less by either regime's ceiling than by their interaction. Three operational consequences follow.
First, the EAA's per-Member-State variance forces a jurisdictional triage. A multinational cannot reasonably maintain 27 different accessibility compliance baselines; it must pick a single internal standard high enough to satisfy the strictest national surveillance authority it operates under. In practice that means designing to the harmonised EN 301 549 V3.2.1 baseline (WCAG 2.1 AA-equivalent) at a minimum, and increasingly to EN 301 549 V4 / WCAG 2.2 where the draft standard is far enough along to anticipate. The cost of multi-standard compliance is one of the strongest informal drivers of the EAA's convergence toward a single technical floor.
Second, the ADA's private right of action means that even a fully-compliant European platform serving US users picks up a separate, plaintiff-driven enforcement risk that cannot be discharged by any administrative authority. The platform's US exposure is not extinguished by a clean BAFA review or by an AEPD comfort letter. The two regimes operate on parallel tracks; satisfying one does not — formally or informally — discharge the other.
Third, the settlement economics are very different on the two sides. Under the EAA, an administrative authority that finds a violation typically issues a fine plus a remediation order, both of which are part of the public record once finalised. Under the ADA, the overwhelming majority of cases settle privately, with a non-public settlement agreement covering monetary terms, an attorneys' fee award, and a remediation undertaking. A multinational that has settled 50 ADA cases in a year and 5 EAA cases in the same year has a very different paper trail in each regime — public administrative resolutions in the EU, private confidential settlements in the US — and the difference shapes how the company can defend its overall accessibility posture in regulatory briefings, board reporting, and investor disclosures.
A working synthesis for compliance teams
The accessibility compliance baseline that satisfies the EAA in its strictest Member State (Spain at the €1M ceiling, Italy at the 5%-of-turnover cap) and that meets the ADA's substantive functional-access standard (the "effective communication" and "auxiliary aids and services" tests under 42 USC § 12182(b)(2)(A)(iii)) is, in practice, the same baseline: conformance to WCAG 2.1 Level AA across all consumer-facing digital surfaces, with a documented remediation pathway for legacy components and a published accessibility statement. Designing once for that floor materially reduces the regulatory exposure under both regimes — though it does not eliminate the plaintiff-driven US risk that arises from any individual access barrier a private plaintiff encounters.
What the comparison teaches
The headline numbers — €5,000 to €1 million on the EAA side, $96,384 to $192,768 on the ADA side — are the wrong place to start any comparison of the two regimes. They are the visible top of two very different enforcement architectures: a Member-State administrative apparatus on one side, a plaintiff-driven federal litigation system on the other, each producing economic exposure in ways the headline ceiling does not capture.
What the comparison teaches, then, is mostly about the cost of multi-jurisdictional accessibility risk. The EAA's per-Member-State variance creates a regulatory surface area that scales with the number of EU markets a company operates in, with the worst-case exposure set by whichever Member State has the highest ceiling. The ADA's flat federal ceiling looks modest until the multipliers — volume of cases, attorneys' fee-shifting, state-law statutory damages in California and New York — are added back in. Each regime, read on its own, would lead a compliance team to a different prioritisation. Read together, they push toward the same conclusion: that designing to a single accessibility floor high enough to satisfy both regimes is cheaper than maintaining a fragmented compliance posture and arguing about ceilings after the fact.
The next eighteen months will sharpen the comparison. The EAA's first cross-border enforcement action — likely against a non-EU e-commerce platform — will test whether the directive's Member-State-by-Member-State structure can deliver coordinated action against a multinational. The ADA's continued private-plaintiff volume will continue to be the dominant signal of US enforcement intensity. And the early national-court decisions on EAA disproportionate-burden defences will tell compliance teams how forgiving the European regime is in practice. The two regimes will not converge, but the multinational defendant's view of them will.
---
title: EN 301 549 explained — the EU's accessibility standard
url: https://www.disabilityworld.org/articles/en-301-549-explained/
description: EN 301 549 — the ETSI harmonised European standard that turns WCAG 2.1 AA into procurement-binding text. V3.2.1 is in force in 2026; V4 incorporating WCAG 2.2 is in late-stage drafting. Full clause-by-clause primer.
author: Disability World
pubDate: 2026-05-22
tags: en-301-549, etsi, technical-standards, eu, procurement, wcag
---
# EN 301 549 explained — the EU's accessibility standard
Image description: Hands on a mechanical keyboard with a technical-standards document open on an external monitor — the accessibility-auditor's workspace where EN 301 549 lives.
Reading Time: 11 minutes
EN 301 549 is the harmonised European standard for accessibility requirements applicable to ICT products and services. Published and maintained by ETSI — the European Telecommunications Standards Institute — in cooperation with CEN and CENELEC, it is the technical instrument that turns the more abstract obligations of the European Accessibility Act (EAA), the Web Accessibility Directive (Directive (EU) 2016/2102), and most national public-procurement rules into a clause-by-clause checklist that a vendor can be measured against. Where WCAG is a content-and-interface standard for the web, EN 301 549 is the broader frame that wraps WCAG inside the requirements that EU law actually procures against.
In 2026 the in-force version is V3.2.1, published in March 2021 and incorporating [WCAG 2.1 Level AA](/toolkit/standards/wcag/) by reference. A new revision incorporating WCAG 2.2 AA — provisionally numbered V4.0.0 — is in late-stage drafting inside the joint ETSI/CEN/CENELEC technical body and is expected to publish during 2026, with citation in the Official Journal of the European Union following. This piece is a primer: what the standard is, how its twelve chapters are organised, where Chapter 9 (web) and Chapter 11 (software) sit alongside Chapter 10 (documents) and Chapter 12 (documentation and support), how the standard bridges WCAG to EU procurement law, and where you will see it cited in the legislation graph you already know.
What EN 301 549 actually is — and what it is not
EN 301 549 is a harmonised European standard. That phrase has a precise meaning in EU law: it is a standard developed by one of the three recognised European Standards Organisations (ETSI, CEN, CENELEC) at the request of the European Commission via a formal "standardisation request" (also called a "mandate") and then cited in the Official Journal of the European Union as giving "presumption of conformity" with the corresponding piece of EU legislation. A product or service that meets the harmonised standard is presumed to meet the legal requirements it harmonises. The presumption can be rebutted, but in practice public buyers, accessibility auditors and conformance bodies treat the standard as the operative checklist.
EN 301 549 originated from Mandate M/376, issued by the Commission in 2005 to make Europe's procurement rules procurement-ready for accessibility — a single, technology-neutral, harmonised reference for what "accessible ICT" means across the public-sector tendering process. The first published version, V1.1.2, appeared in 2014. The standard has since gone through three substantive revisions: V2 (2018) aligning with WCAG 2.1, V3.1.1 and V3.2.1 (2019–2021) tightening definitions and adding mobile-app and authoring-tool clauses, and the forthcoming V4 incorporating WCAG 2.2.
What EN 301 549 is not: it is not the EAA, and it is not the Web Accessibility Directive. Those are the laws that procure against the standard. EN 301 549 is the test-criterion document — the part of the system a developer or a procurement officer actually reads to know whether a deliverable passes.
The twelve-chapter structure
EN 301 549 is organised around twelve substantive clauses (numbered from Clause 4 in the document, because Clauses 1–3 are scope and definitions). The architecture is deliberately modular: a vendor scoping a tender response works out which clauses apply to the product, applies only those, and declares conformance against the named clauses. The headline modules sit in Chapters 9 through 12.
Chapter 9 — Web content
Chapter 9 is the chapter that most accessibility practitioners arrive at first, because it is the one that incorporates WCAG by reference. In V3.2.1, Chapter 9 imports the WCAG 2.1 Level A and Level AA success criteria verbatim: clause 9.1 covers the perceivable success criteria, 9.2 the operable, 9.3 the understandable, 9.4 the robust. A web product that conforms to WCAG 2.1 AA conforms to Chapter 9. The standard does not paraphrase the WCAG text; it cites it. In V4 the same chapter will reference WCAG 2.2 AA, picking up the nine new and revised success criteria — including 2.4.11 Focus Not Obscured (Minimum), 2.4.12 Focus Not Obscured (Enhanced), 2.4.13 Focus Appearance, 2.5.7 Dragging Movements, 2.5.8 Target Size (Minimum), 3.2.6 Consistent Help, 3.3.7 Redundant Entry, 3.3.8 Accessible Authentication (Minimum), and 3.3.9 Accessible Authentication (Enhanced).
Chapter 10 — Non-web documents
Chapter 10 applies WCAG-equivalent requirements to non-web documents — PDFs, Word files, slide decks, ePub, and any other document delivered alongside or outside the web. It does this by taking each WCAG 2.1 success criterion that makes sense for a non-web document and restating it in the document context. A tagged, navigable, properly-described PDF conforms to Chapter 10; an untagged scan of a printed report does not. Public buyers procuring policy publications, contract terms, training materials and accessibility statements rely on Chapter 10 to set the bar for what they accept as a deliverable.
Chapter 11 — Non-web software
Chapter 11 is the broadest module and the most consequential for the modern application stack. It applies WCAG-equivalent requirements to non-web software — desktop applications, native mobile apps, embedded interfaces, kiosks running custom software — and adds requirements that are software-specific and have no WCAG analogue: clauses on platform accessibility services (11.5), on assistive-technology compatibility (11.6), and on authoring tools (11.8, derived from the W3C's Authoring Tool Accessibility Guidelines). The mobile-app coverage in Chapter 11 is the reason the Web Accessibility Directive can extend to public-sector mobile applications, and the reason the EAA can apply to e-readers, ticketing machines and self-service terminals without needing a separate standard for each.
Chapter 12 — Documentation and support services
Chapter 12 covers documentation and customer-support services: user manuals, help systems, support call centres, online chat, accessible formats on request. The clauses require that product documentation describe the accessibility features of the product, that documentation itself be accessible, and that support services be available through accessible channels. This is the chapter that ties accessibility to the post-purchase experience — the part of the buying journey where users actually encounter the product and need help with it.
Chapters 5–8 — the cross-cutting generic requirements
Chapters 5 through 8 sit upstream of the format-specific modules. Chapter 5 covers generic requirements that apply to any ICT product or service — closed functionality, biometrics, preservation of accessibility information through conversion. Chapter 6 covers ICT with two-way voice communication: real-time text, video relay, and the interoperability requirements that make accessible communication possible across service providers. Chapter 7 covers ICT with video capabilities — audio description, captioning, signed presentation. Chapter 8 covers hardware: keyboards, controls, connectors, physical access. A product is rarely scoped against only one chapter; a video-streaming app on a smart TV will touch Chapters 5, 7, 9 (if it has a web interface), 11 (the app itself) and 12 (its documentation) at once.
Chapter 13 and the annexes
Chapter 13 deals with ICT providing relay services and emergency-services access — the public-interest communications layer. The annexes are where the standard does its procurement-binding work: Annex A contains the conformance methodology, including the mandatory "accessibility statement" template; Annex B maps EN 301 549 clauses to the corresponding requirements in U.S. Section 508 — useful for vendors selling on both sides of the Atlantic; Annex C provides guidance on functional performance statements; and the bibliography lists every standard, including WCAG, that the document incorporates by reference.
How WCAG 2.1 AA actually sits inside EN 301 549
The relationship between WCAG and EN 301 549 is the single most asked question in conformance work, and the answer is more specific than "EN 301 549 includes WCAG." WCAG 2.1 Level AA is incorporated into Chapter 9 (web content) and parts of Chapter 11 (software, where the success criteria are applicable to non-web software). The incorporation is by reference, not by paraphrase: Chapter 9's clauses are numbered to mirror WCAG's structure, and each clause points to the corresponding success criterion in the W3C recommendation. A WCAG 2.1 AA conformance claim translates directly into a Chapter 9 conformance claim.
Where EN 301 549 goes beyond WCAG is in the software-specific, hardware-specific and documentation-specific clauses that WCAG was never designed to cover. WCAG addresses content perceivable, operable, understandable and robust within a web user agent. EN 301 549 adds the requirements that handle, for example, a desktop app's interaction with a screen-reader API on Windows, a hardware keyboard's tactile discriminability, or a contact-centre's TTY interoperability. A product can be WCAG 2.1 AA conformant and still fail EN 301 549 — typically because Chapters 11 or 12 cover requirements WCAG does not address.
Where EN 301 549 is cited in EU law
The standard's load-bearing role is in the citation graph. Three primary legal instruments name EN 301 549 as the technical reference; several dozen national procurement laws and accessibility statutes do the same.
The European Accessibility Act (Directive (EU) 2019/882)
The European Accessibility Act sets functional accessibility requirements for a defined list of products and services — computers, smartphones, e-readers, ATMs, ticketing machines, e-commerce, e-books, telephony, audiovisual media services, banking services, passenger transport information. The Act's functional requirements (Annex I) are abstract; they require, for example, that information be provided in accessible formats, that user interfaces support assistive technology, that emergency communications work with deaf users. To make those abstract requirements operational, the EAA relies on harmonised standards cited under Article 15 of Regulation (EU) 1025/2012 — and EN 301 549 is the harmonised standard the European Commission is using to operationalise the EAA's web, software and documentation requirements. A product that conforms to the relevant clauses of EN 301 549 carries presumption of conformity with the EAA. The first Official Journal citation of EN 301 549 specifically for EAA purposes published in 2024; revisions follow each new version.
The Web Accessibility Directive (Directive (EU) 2016/2102)
The Web Accessibility Directive, in force since December 2016, requires public-sector bodies in EU Member States to make their websites and mobile applications accessible. Article 6 of the Directive states that content meeting the harmonised standard cited in the Official Journal is presumed to conform with the corresponding accessibility requirements set out in Article 4. EN 301 549 is the standard so cited — V2 from 2018 was the first version named for WAD purposes, with each subsequent revision triggering a new OJ citation. Public-sector websites and mobile apps that meet Chapter 9 and the applicable parts of Chapter 11 are deemed compliant with the Directive.
National procurement laws and Article 42 of the Public Sector Procurement Directive
Article 42 of Directive 2014/24/EU (the Public Sector Procurement Directive) requires that technical specifications in public tenders for products and services that will be used by natural persons "take into account accessibility criteria for persons with disabilities or design for all users." Member States have implemented that obligation in their national procurement codes, and the implementing texts typically name EN 301 549 as the reference standard — from Germany's BITV 2.0 and the EU-Verordnung referenced in federal procurement, to Spain's Real Decreto 1112/2018, to France's RGAA (which aligns its criteria with EN 301 549 Chapter 9), to Italy's Linee Guida AgID, to the Netherlands' Tijdelijk besluit digitale toegankelijkheid overheid. The national-procurement layer is where EN 301 549 has the most day-to-day commercial impact, because it determines which vendors can bid for which public contracts.
What V4 changes — and what it does not
The forthcoming V4 of EN 301 549 is the working title for the revision that will incorporate WCAG 2.2 AA in place of WCAG 2.1 AA, picking up the nine success criteria the W3C added or revised in the 2023 update. The working revision has been visible in ETSI's Technical Committee Human Factors public archive since 2024, and the joint ETSI/CEN/CENELEC working group convened during 2025 to finalise it. Publication during 2026 is the working assumption inside the standards community; OJ citation under the EAA and WAD then follows on the Commission's usual timeline (typically several months after ETSI publication).
The substantive deltas in V4 cluster around two areas. First, the WCAG 2.2 success criteria themselves — Chapter 9 picks up the nine new criteria, the most operationally significant being Focus Not Obscured, Target Size (Minimum), Dragging Movements, and the two Accessible Authentication criteria, which together will force a re-audit of any product that uses overlay banners, cookie modals, password fields, or small tap targets. Second, the standard's software clauses (Chapter 11) are being tightened to align more closely with WCAG 2.2 for software where the success criteria apply, and to update the platform-accessibility-services language to reflect the assistive-technology APIs that have shipped since 2021.
What V4 does not change: the twelve-chapter architecture, the conformance-statement template in Annex A, the relationship to the EAA and WAD, or the Section 508 mapping in Annex B. A vendor that has a current conformance statement against V3.2.1 will, in most cases, need to re-test for the new WCAG 2.2 criteria but not re-architect their conformance approach.
Using EN 301 549 in practice: the conformance statement
The operational artefact that EN 301 549 produces is a conformance statement — sometimes called an "accessibility statement" in WAD usage, or a Voluntary Product Accessibility Template (VPAT) when expressed in the Section 508 lineage. Annex A of the standard contains the template. For each applicable clause, the vendor states whether the product "Supports," "Partially Supports," "Does Not Support," or "Not Applicable." Each "Partially Supports" or "Does Not Support" must be accompanied by a remarks-and-explanations field describing the gap.
In a tender response, the procurement officer scopes the relevant chapters to the product, requires a clause-by-clause conformance statement, and evaluates the gaps. The statement is contractually binding in most EU public-procurement frameworks — if the vendor declares "Supports" against a clause and the product subsequently fails on that clause during user acceptance, the contract typically gives the buyer grounds for remediation, penalties, or rescission. This is why EN 301 549 has more commercial bite than the underlying WCAG document on its own: WCAG is a W3C recommendation with no procurement standing; EN 301 549 is the document a contract names.
EN 301 549 in the legislation graph you already know
If you have followed the EU disability-rights legislation arc — the EAA, the Web Accessibility Directive, the national procurement codes that implement Directive 2014/24/EU — EN 301 549 is the technical underlay that connects those laws to a vendor's day-to-day testing process. WCAG sets the web-content rules. EN 301 549 wraps WCAG in the wider set of requirements (software, documents, documentation, hardware, two-way communications) that EU procurement law actually buys against. The EAA and WAD then cite EN 301 549 as the standard that triggers presumption of conformity. National procurement codes name the standard in their technical specifications, and accessibility auditors test against it clause by clause.
For practitioners scoping a 2026 audit: V3.2.1 is the version to test against today, V4 is the version to start preparing for, and the deltas worth getting ahead of are the nine WCAG 2.2 success criteria — particularly the focus-appearance and target-size criteria, which are the ones most products are quietly failing. The fastest way to spot which 2.2 criteria your site already trips on is a [free WCAG 2.2 scan](/toolkit/scan/) on a representative page. For the wider 2026 reporting record on how this standard interacts with national enforcement, see the Disability World article index; for the EAA's first-year enforcement picture across Member States, see the EAA primer. For a hands-on translation of V3.2.1 + the 2.2 deltas into a working audit, see the step-by-step WCAG 2.2 compliance playbook; for the monitoring platforms that maintain conformance between audits, see the accessibility monitoring buyer's guide.
Primary sources
ETSI / CEN / CENELEC. EN 301 549 V3.2.1 (2021-03) — Accessibility requirements for ICT products and services. etsi.org
European Commission. Standardisation request M/376 (2005) on ICT accessibility for public procurement.
Directive (EU) 2019/882 of the European Parliament and of the Council on the accessibility requirements for products and services (European Accessibility Act).
Directive (EU) 2016/2102 of the European Parliament and of the Council on the accessibility of the websites and mobile applications of public sector bodies.
Directive 2014/24/EU of the European Parliament and of the Council on public procurement, Article 42.
Regulation (EU) No 1025/2012 of the European Parliament and of the Council on European standardisation.
W3C. Web Content Accessibility Guidelines (WCAG) 2.1 (W3C Recommendation, June 2018) and WCAG 2.2 (W3C Recommendation, October 2023).
ETSI Technical Committee Human Factors. Public archive on EN 301 549 revision activity (2024–2025).
---
title: EPUB3 for accessible publishing: what publishers need to ship
url: https://www.disabilityworld.org/articles/epub3-for-accessible-publishing/
description: EPUB3.3 is the format publishers will be measured against under the European Accessibility Act.
author: Disability World
pubDate: 2026-05-22
tags: epub3, ebooks, publishing, accessibility, mathml, media-overlays, education
---
# EPUB3 for accessible publishing: what publishers need to ship
Image description: An e-reader sits on a stack of physical books with a pair of earbuds resting on top, the screen showing a page of accessible text — the everyday surfaces where EPUB3 has to work.
Reading Time: 12 minutes
EPUB3 is the format publishers will be measured against when the European Accessibility Act begins to be enforced in earnest. It is also the format that the WIPO Marrakesh Treaty and the Accessible Books Consortium use to move accessible books across borders, and the format that screen-reader users, low-vision readers and print-disabled students expect when they buy an ebook. Unlike PDF, EPUB3 is reflowable, semantic and accessible by design — but only if the publisher actually ships the metadata, the markup and the navigation that the spec calls for. A file with the .epub extension is not the same thing as an accessible EPUB.
This primer is for publishers, editorial-technology teams and accessibility leads inside ebook retailers. It walks through what the EPUB3.3 specification requires, which accessibility-metadata fields the schema.org and EPUB Accessibility 1.1 specifications expect you to populate, which reading systems actually render EPUB3 well in 2026, where the EAA compliance pressure on retailers is biting, and how the Marrakesh ecosystem rounds out the picture. It is deliberately concrete: by the end you will know what to ask your conversion vendor for, what to put in your metadata, and what to test before you upload to a retailer.
What EPUB3 requires
EPUB3 is a W3C Recommendation. The current stable version is EPUB 3.3, published as a W3C Recommendation in May 2023 after the format graduated from the IDPF into the W3C. EPUB 3.3 consolidated a series of incremental revisions, made accessibility a first-class requirement rather than an optional companion document, and tightened the relationship between EPUB and the wider open-web platform — an EPUB is, at its core, a packaged ZIP archive of HTML, CSS, SVG and supporting resources, governed by an OPF (Open Packaging Format) manifest and a navigation document.
For the file itself to be accessible, EPUB 3.3 expects publishers to use semantic HTML throughout. That means real headings in document order (h1 through h6), real lists (ul, ol, dl), real tables for tabular data with proper thead, tbody, th scoping, and the EPUB-specific structural semantics vocabulary (epub:type) for marking up chapters, sections, footnotes, page-list entries, glossary terms and the dozens of other publishing roles the spec recognises. A book whose chapter headings are visually styled paragraphs without any heading element is not accessible — a screen reader cannot jump to the next chapter, a refreshable braille display cannot announce the chapter break, and a reflow engine cannot generate a table of contents on the fly.
Language tags are non-negotiable. Every EPUB must declare a primary language in the OPF package document, and any inline content in a different language must be marked with the appropriate lang and xml:lang attributes. Text-to-speech engines and screen readers switch voice profiles on the basis of these tags; a French paragraph inside an English book that is not language-tagged will be read out in an English voice with predictably comic, and exclusionary, results. The same rule applies to direction (dir) for mixed left-to-right and right-to-left content.
Every EPUB must ship a navigation document — a single XHTML file referenced from the OPF as the navigation document, containing at minimum a table of contents (nav epub:type="toc"), and ideally a page-list (nav epub:type="page-list") mapping print-page numbers to in-book locations and a landmarks list (nav epub:type="landmarks") flagging the cover, the body matter, the index and other discoverable surfaces. The page-list is the feature that allows a student using an accessible ebook to follow the page references in a printed syllabus without falling out of sync with classmates reading the paper edition.
Images need alt text for every image that conveys content. Decorative images take alt="" and an aria-hidden="true" if appropriate, but content images — diagrams, photographs in a photography book, maps, illustrations in a children's book — need real descriptions. Complex images such as scientific diagrams need long descriptions, either inline via aria-describedby pointing to a description element, or via the epub:type="describedFootnote" pattern. Maths in any book that goes beyond casual mention should be encoded as MathML, not rasterised as PNG screenshots. MathML is the only encoding that lets a screen reader speak an equation, that lets a refreshable braille display render it in Nemeth or Unified English Braille, and that lets a reader resize the equation without pixelation.
EPUB3 also supports media overlays — synchronised text and pre-recorded audio narration, defined in SMIL files that map each text fragment to a time-range in the audio. A media-overlay EPUB lets a low-literacy reader, a reader with a cognitive disability, or simply a commuter follow the highlighted text as a human narrator reads it aloud. The SMIL approach predates the modern wave of high-quality TTS, but the two technologies are complementary: media overlays remain the gold standard for children's books, language-learning titles and accessibility-funded conversions, while TTS handles the long tail.
Accessibility metadata: the schema.org / A11y-meta layer
An accessible file that does not advertise itself as accessible is invisible to the readers who need it. The EPUB Accessibility 1.1 specification, published as a W3C Recommendation alongside EPUB 3.3, mandates a set of metadata fields that must appear in the OPF package document. These fields draw on the schema.org accessibility vocabulary — the same vocabulary used by Bookshare, the DAISY Consortium, Benetech, the Accessible Books Consortium, and the major retailer feeds.
The required and strongly-recommended properties are:
schema:accessMode — the modalities the content uses (textual, visual, auditory). A novel is textual; an illustrated children's book is textual,visual; a media-overlay audio-text book is textual,visual,auditory.
schema:accessModeSufficient — the modality combinations that are sufficient to consume the content on their own. A novel typically lists textual as sufficient (because everything important is in the text). A graphic novel without alt-text descriptions cannot honestly claim textual alone is sufficient.
schema:accessibilityFeature — the discrete features present, drawn from a controlled vocabulary: structuralNavigation, alternativeText, longDescription, tableOfContents, readingOrder, printPageNumbers, mathML, synchronizedAudioText, highContrastDisplay, displayTransformability, captions, transcript, and several more.
schema:accessibilityHazard — disclosure of any content that may trigger seizures, motion sickness or other reactions: flashing, noFlashingHazard, motionSimulation, noMotionSimulationHazard, sound, noSoundHazard. If the book is hazard-free, say so explicitly.
schema:accessibilitySummary — a human-readable, plain-language summary of the accessibility of the publication, written for the end reader: "This publication conforms to WCAG 2.1 Level AA. All images have alternative text. Mathematical equations are encoded in MathML. Page numbers reflect the print edition."
a11y:certifiedBy, a11y:certifierCredential, a11y:certifierReport — if a third party has certified the file against EPUB Accessibility 1.1, who certified it, what credential they hold, and a link to the published certification report.
dcterms:conformsTo — the conformance profile the publication meets, expressed as a URL pointing to EPUB Accessibility 1.1 conformance criteria (WCAG 2.1 Level AA, or in newer files Level AAA where claimed).
These fields are not paperwork. They flow into retailer catalogues, into the Accessible Books Consortium's global accessible-book database, into Bookshare's discovery, into school-procurement catalogues and into the EAA reporting templates that retailers now have to maintain. The European standard EN 17161 — accessibility through "design for all" — references this metadata layer, as does the Functional Accessibility Evaluation criteria used by the ACE accessibility checker maintained by the DAISY Consortium.
Reading systems: what actually renders EPUB3 in 2026
The single most quoted complaint in publisher accessibility teams is that the same EPUB renders differently on different reading systems. The complaint is true and the gap matters. A file that scores perfectly on the DAISY ACE checker may still fail to expose its navigation document on a popular consumer reader, or fail to speak its MathML on a major iOS app. The gap between what the spec defines and what the reading system implements is the reason a publisher's accessibility workflow has to include real-device testing, not just file-level validation.
Thorium Reader, maintained by the EDRLab consortium, is the reference free desktop reader for accessible EPUB3 in 2026. It implements EPUB 3.3 and EPUB Accessibility 1.1 thoroughly, exposes the navigation document, the page-list and the landmarks list, renders MathML, supports media overlays, and integrates with the operating-system text-to-speech and the major screen readers (NVDA on Windows, VoiceOver on macOS, Orca on Linux). Many publishers use Thorium as their accessibility-acceptance reader: if a file works in Thorium, it is well-formed and conformant.
VoiceDream Reader (now part of the Voice Dream stable acquired in 2022) remains the leading iOS app for print-disabled readers who want premium TTS voices and fine-grained control over speech parameters. It opens EPUB3 reliably, respects language tags for voice switching, supports custom fonts and dyslexia-friendly typography, and integrates with the Bookshare and Learning Ally catalogues. For students and adult readers with dyslexia, low vision or blindness, VoiceDream is often the default app.
VoiceOver Books — Apple's built-in audiobook experience inside the Books app, paired with iOS VoiceOver — is the route most blind iOS users actually take. It handles EPUB3 well, exposes the navigation document to VoiceOver, speaks alt text, switches voices on language tags, and surfaces media overlays. Where Apple Books still struggles is MathML rendering inside complex STEM titles and consistent exposure of the page-list when the user is navigating by print-page reference.
Apple Books on macOS, iPadOS and iOS is the broadest consumer reading system for EPUB3 in the western market and renders most accessibility features competently. Its limitations are in the long tail: certain media-overlay edge cases, certain rare MathML constructs, and inconsistent behaviour with very large page-lists.
The conspicuous exception in 2026 remains Amazon Kindle. Amazon does not natively support EPUB3 inside the Kindle ecosystem; instead it ingests EPUB and converts it on upload into its proprietary KF8 / KFX formats. The conversion preserves text, basic structure and many images, but it does not preserve all accessibility metadata, does not render MathML reliably, drops media overlays entirely, and has historically not exposed the schema.org accessibility-metadata fields to users searching the Kindle catalogue. Publishers shipping into Amazon often maintain a parallel KF8/KFX accessibility workflow, but the practical effect is that the most accessible EPUB3 a publisher can produce is partially degraded the moment it reaches the largest English-language ebook retailer. The EAA pressure described in the next section is the first regulatory lever capable of moving that needle.
EAA pressure on ebook retailers
The European Accessibility Act (Directive (EU) 2019/882) entered into application on 28 June 2025, and ebooks are explicitly inside its scope. Article 4 of the directive obliges economic operators to ensure that the products and services they place on the EU market comply with the accessibility requirements set out in Annex I. For ebooks and dedicated ebook software, the Annex I requirements include: ensuring that ebooks (and the software needed to access them) support text-to-speech, allow users to adapt presentation (font size, contrast, line spacing), expose the metadata necessary for assistive technology to navigate the content, allow synchronised audio and text where present, provide non-text content with alternatives, and prevent ebook protection measures from blocking accessibility features.
In practice, the Annex I list maps almost one-to-one onto the EPUB Accessibility 1.1 conformance criteria. A publisher that ships EPUB3 files conforming to EPUB Accessibility 1.1 — with the schema.org metadata properly populated and a certifier's statement — has a strong presumption of conformity with the Annex I obligations. A publisher shipping unstructured PDF or DRM-locked formats that block screen-reader passthrough is squarely in non-compliance.
The compliance pressure does not fall only on publishers. It falls equally on retailers, which the directive treats as economic operators in their own right. National market-surveillance authorities began their first round of EAA compliance inspections during the second half of 2025 and through 2026, and ebook retailers have been an early focus because the catalogues are public, the metadata is machine-readable and the non-compliance is easy to evidence. Retailers operating in the EU now generally require publishers to deliver EPUB Accessibility 1.1-conformant files, to populate the schema.org metadata fields, and to provide a certification statement; some have begun rejecting unconformant uploads at ingest. For platforms with significant proprietary-format dependencies — Amazon Kindle, in particular — the EAA has forced a public commitment to closer EPUB3 fidelity, though the engineering work to deliver it is still in progress.
For publishers, the operational consequence is unambiguous: ebook accessibility metadata is now a publishing requirement, not a nice-to-have. Production teams that previously ran accessibility as a separate downstream conversion now build it into the source workflow.
The Marrakesh Treaty and Accessible Books Consortium ecosystem
EPUB3 sits inside a wider treaty and infrastructure ecosystem that publishers should understand because it changes what "accessible book" means at scale. The Marrakesh Treaty — the WIPO Marrakesh Treaty to Facilitate Access to Published Works for Persons Who Are Blind, Visually Impaired, or Otherwise Print Disabled, adopted in 2013 and now in force in more than 100 contracting parties including the EU and the United States — creates a copyright exception that allows authorised entities to produce, distribute and cross-border-exchange accessible-format copies of published works for the benefit of beneficiary persons, without needing rightsholder permission for each transaction.
The Treaty is implemented in EU law via Directive (EU) 2017/1564 and Regulation (EU) 2017/1563, and in the United States via the Marrakesh Treaty Implementation Act of 2018, which amended Title 17. The operational infrastructure is run by the Accessible Books Consortium (ABC), a WIPO-led alliance of organisations representing visually impaired people, libraries serving them, publishers and standards bodies. ABC operates the Global Book Service, a cross-border lending and exchange platform through which authorised entities — typically national libraries for the blind, organisations such as Bookshare in the United States and the RNIB in the United Kingdom, and equivalent national agencies across Europe and the Global South — share accessible files.
The format of choice for these exchanges is EPUB3 with full accessibility metadata, and for older or scanned material the DAISY 2.02 and DAISY 3 formats that EPUB3 effectively succeeds. A book that a French publisher has produced as an EPUB Accessibility 1.1-conformant title can, in principle, be shared via the ABC Global Book Service with a print-disabled reader in Kenya, India, Argentina or any other contracting party, without renegotiation. The Treaty does not change the publisher's commercial position — it operates on the accessible copy specifically, for the beneficiary population specifically — but it does dramatically enlarge the reading public for any well-formed accessible ebook a publisher ships.
For publishers, the practical link between the EAA layer and the Marrakesh layer is the same metadata block. The schema.org accessibility properties, the EPUB Accessibility 1.1 conformance claim, and the certification report you produce for EAA compliance are the same artefacts that allow your file to flow into the ABC Global Book Service and the wider network of authorised entities. Ship the file once, in the right format, with the right metadata, and the same artefact serves the EU compliance regime and the global accessible-reading public at the same time.
A practical workflow for publishers
The implementation pattern that production teams settle on, once the dust clears, is built around four anchors. Source accessibility: the source manuscript is structured (real headings, real lists, real tables, language-tagged) before any conversion happens, so the EPUB conversion preserves structure rather than reverse-engineering it. Conversion to EPUB 3.3: the conversion tool — whether in-house, a vendor pipeline, or an open-source toolchain such as the DAISY Consortium's tooling — produces EPUB 3.3 with semantic markup, a navigation document, a page-list where the title has a print equivalent, alt text on all content images, MathML where mathematics appears, and media overlays where the editorial brief calls for them.
Metadata population: every file leaves production with a complete schema.org accessibility metadata block — accessMode, accessModeSufficient, accessibilityFeature, accessibilityHazard, accessibilitySummary, conformsTo — and where the title is certified, the a11y:certifiedBy/Credential/Report fields are populated against the certifier of record (commonly Benetech's certification programme, the DAISY Consortium, or a national equivalent). Validation and real-device testing: every file is validated against EPUBCheck and the DAISY ACE accessibility checker, and a representative sample is tested on Thorium, Apple Books, VoiceDream and the retailer-specific reading systems the title will be sold through.
The cost of doing this is real, but it falls quickly with practice and tooling. The cost of not doing it — EAA non-compliance fines, retailer rejection at ingest, missing readers in the Marrakesh network, and the wider reputational cost of shipping ebooks that disabled readers cannot use — is now firmly higher. EPUB3 accessibility is no longer a specialist sub-discipline at the end of the production pipeline. It is the spec.
---
title: EU AI Act Articles 16 + 73: where the high-risk AI rules intersect with disability law
url: https://www.disabilityworld.org/articles/eu-ai-act-disability-intersection/
description: Regulation (EU) 2024/1689 — the EU AI Act — entered general application on 2 August 2026. A primer on how Article 16 (general-purpose AI obligations) and Article 73 (high-risk AI requirements) intersect with disability law in employment, education, and essential services.
author: Disability World
pubDate: 2026-05-22
tags: eu-ai-act, ai, disability, accessibility, high-risk-ai, regulation-primer
---
# EU AI Act Articles 16 + 73: where the high-risk AI rules intersect with disability law
Image description: A printed EU AI Act document with a transparent ARIA accessibility-tree overlay and a fountain pen on top — the visual marker for the AI Act x disability law intersection.
Reading Time: 12 minutes
Regulation (EU) 2024/1689, commonly called the EU AI Act, was published in the Official Journal on 12 July 2024, entered into force on 1 August 2024, and reached its main applicability date — when the high-risk and general-purpose-AI obligations bind providers and deployers across the Single Market — on 2 August 2026. It is the first comprehensive horizontal AI law in any major jurisdiction, and it sits on top of, rather than replaces, the existing disability-rights stack: the European Accessibility Act, the Web Accessibility Directive, the Employment Equality Directive 2000/78/EC, and the EU's ratification of the UN Convention on the Rights of Persons with Disabilities (CRPD).
Two articles do most of the load-bearing work where the AI Act and disability law collide. Article 16 sets the obligations on providers of general-purpose AI models — the foundation-model layer that powers most consumer-facing AI products in 2026. Article 73, read together with Articles 8 through 15 and Annex III, sets the requirements that bind providers and deployers of high-risk AI systems. This piece is a primer on how those two articles intersect with disability law in three concrete settings: AI used in employment (CV-screening tools, automated video-interview scoring), AI used in education (online proctoring, accessibility tooling, student-risk modelling), and AI used in essential services (consumer credit scoring, healthcare triage, public-benefit eligibility decisions). It also covers the CRPD overlay that the EU's institutional commitments add on top, and the documentation duties — Annex IV technical documentation, post-market monitoring, fundamental-rights impact assessments — that vendors are now expected to produce.
What the AI Act is — and how it is structured
The AI Act is a regulation, not a directive: it applies directly in every Member State without national transposition, and the obligations it imposes on providers and deployers are uniform across the 27/27 EU national markets plus the EEA. Its core architectural choice is a risk-tier framework with four levels — prohibited practices (Article 5), high-risk AI systems (Articles 6 through 27 and Annex III), limited-risk transparency obligations (Article 50), and unregulated minimal-risk uses. Layered on top of the risk tiers, a separate regime — Articles 51 through 56 — governs general-purpose AI models, with stricter obligations triggered when a model crosses the systemic-risk threshold set in Article 51(2).
The phased applicability calendar matters because providers reading this in 2026 are not facing a single deadline. The Article 5 prohibitions on unacceptable-risk practices — social scoring by public authorities, real-time remote biometric identification in public spaces except for narrowly defined law-enforcement uses, emotion recognition in workplaces and schools — became applicable on 2 February 2025, six months after entry into force. The Article 51-56 general-purpose-AI obligations became applicable on 2 August 2025. The full high-risk regime, including the Article 73 post-market monitoring duties, came into force on 2 August 2026, with a further extension to 2 August 2027 for the subset of high-risk systems that are also safety components of products already regulated under EU product-safety legislation (Annex I sectoral law — medical devices, machinery, toys, vehicles).
Enforcement is split. National market-surveillance authorities — designated by each Member State and listed in a public register maintained by the AI Office — handle high-risk AI enforcement on the ground. The AI Office, established inside the European Commission's DG CNECT, has exclusive competence for general-purpose-AI enforcement under Article 88. Maximum administrative fines run to EUR 35 million or 7% of worldwide annual turnover for breaches of the Article 5 prohibitions, EUR 15 million or 3% for breaches of most other operator obligations including the Article 16 and Article 73 duties covered in this primer, and EUR 7.5 million or 1% for supplying incorrect or misleading information to authorities.
Article 16 — what general-purpose-AI providers must do
Article 16 is the operative provision for the foundation-model layer. It applies to providers of general-purpose AI models — defined in Article 3(63) as AI models trained on a large amount of data using self-supervision at scale, displaying significant generality, capable of competently performing a wide range of distinct tasks. The large language models powering chatbots, the multimodal image-and-text models used in document analysis, the speech models that increasingly mediate accessibility tooling: all of them are general-purpose AI models for AI Act purposes, and their providers carry the Article 16 stack.
The Article 16 duties divide into three blocks. First, technical documentation: providers must prepare and keep up to date a technical file covering the model's training, testing, and evaluation, including the training-data sources at a high level, the energy consumption of training, the evaluation benchmarks used, and the known limitations. Annex XI specifies the minimum content. Second, information disclosures to downstream deployers: providers must make available to companies that integrate the model into their own systems enough information about the model's capabilities, limitations, intended use, and known risks for the downstream operator to comply with its own AI Act obligations. Third, copyright and content-provenance: providers must put in place a policy to comply with EU copyright law, including the Article 4(3) text-and-data-mining opt-out under Directive (EU) 2019/790, and publish a sufficiently detailed summary of the training-data corpus.
The disability angle on Article 16 is twofold. First, the limitations disclosure required under Annex XI Section 1(2)(c) explicitly covers known biases and unfair performance gaps — and accessibility-relevant performance gaps fall squarely inside that requirement. A speech-recognition model that performs measurably worse on dysarthric speech, an image-captioning model that misidentifies users of wheelchairs or mobility aids, a sign-language model that fails on regional sign-language variants: each of these is a known limitation the provider must surface to downstream deployers. Second, providers of models meeting the Article 51(2) systemic-risk threshold (currently set at training compute exceeding 10^25 FLOPs) carry the additional Article 55 duties of adversarial testing, incident reporting, cybersecurity safeguards, and model evaluation against systemic-risk categories — including fundamental-rights impacts, which the regulation explicitly cross-references back to the Charter of Fundamental Rights and to the CRPD.
Article 73 — high-risk AI systems and the post-market regime
Article 73 sits inside Section 3 of Chapter III, the section that governs high-risk AI systems already placed on the market. It requires that providers of high-risk AI establish a post-market monitoring system, proportionate to the nature of the system and the risks it presents, that actively and systematically collects, documents, and analyses data on the performance of the AI system throughout its lifetime. The monitoring must continuously evaluate compliance with the requirements set out in Articles 8 to 15, with documentation kept available for at least ten years.
Article 73 has to be read in tandem with the Articles 8-15 substantive requirements, because the post-market monitoring is the mechanism by which compliance with those requirements is demonstrated over time. Article 9 mandates a risk-management system. Article 10 governs data and data-governance — specifically requiring that training, validation, and testing datasets be relevant, sufficiently representative, free of errors, and complete in view of the intended purpose, with explicit attention to "the specific geographical, contextual, behavioural or functional setting" in which the system will be used. Article 11 and Annex IV require technical documentation; Article 12 requires automatic event logging; Article 13 requires transparency and information for deployers; Article 14 requires human oversight measures designed into the system; Article 15 requires accuracy, robustness, and cybersecurity proportionate to the intended purpose. Article 26 then layers obligations on the deployer side — the operator that actually puts the system into use.
What makes a system "high-risk" is set in Article 6 and Annex III. The Annex III list names eight use-case categories — biometrics; critical infrastructure; education and vocational training; employment and worker management; access to essential private and public services; law enforcement; migration, asylum and border control; administration of justice and democratic processes — and within each category enumerates the specific use-cases that trigger the high-risk classification. The Annex III list is non-exhaustive in concept but binding in law: an AI system used for one of the listed purposes is high-risk by operation of the regulation, regardless of how the provider markets it.
Where Articles 16 and 73 intersect with disability law
Three intersection points dominate the practical compliance landscape in 2026: employment AI, education AI, and AI used in essential services. Each one sits inside an Annex III high-risk category, and each one carries a direct obligation under existing EU disability-discrimination law that the AI Act now operationalises.
Annex III Section 4 captures AI systems used for recruitment, selection, task allocation, performance evaluation, and termination decisions. CV-screening tools that rank candidates against job descriptions, automated video-interview platforms that score candidate responses on facial expressions, speech patterns and word choice, productivity-monitoring tools that flag workers for managerial intervention based on keystroke or screen-time data: all of them are Annex III high-risk under Section 4. Article 9's risk-management duty and Article 10's data-governance duty require providers to identify and mitigate disparate-impact risks at the model-training stage. The Employment Equality Directive 2000/78/EC, in force since 2003, already prohibits direct and indirect discrimination on disability grounds in recruitment and employment; the AI Act now requires that the technical machinery behind that prohibition be auditable through the Annex IV technical file and through Article 73 post-market monitoring.
The bias-mitigation duty is explicit. Article 10(2)(g) requires that providers examine training-data design choices for "possible biases that are likely to affect the health and safety of persons, have a negative impact on fundamental rights or lead to discrimination prohibited under Union law." Once a disability-related disparate impact is identified — a video-interview model that systematically penalises candidates with speech disabilities, an HR-analytics model that misclassifies the work patterns of employees with cognitive disabilities or chronic illness — Article 10(2)(h) requires "appropriate measures to detect, prevent and mitigate" it. The mitigation work has to be documented inside the Annex IV file and continuously evaluated through the Article 73 post-market monitoring system.
Annex III Section 3 covers AI used to determine access to education, evaluate learning outcomes, assess the appropriate level of education, and monitor students during tests. Online-proctoring systems that flag examination behaviour are explicitly named in the recitals. The intersection with disability law is acute here: a proctoring model trained on neurotypical baseline behaviour will systematically over-flag students with ADHD, tic disorders, or anxiety — and the under-fifteen-percent of higher-education students who have documented disabilities (Eurostat 2024) bear the disparate impact. The Article 5(1)(f) prohibition on emotion-recognition in educational settings already removes one class of model from the legal market entirely; what remains is the broader proctoring and risk-prediction layer that operates under Annex III Section 3 as high-risk.
Accessibility tooling sits on the other side of the same border. AI-driven captioning, speech-to-text for lectures, AI alt-text generation, and AI-assisted document remediation are not themselves Annex III uses — they are accessibility services. But where an educational institution procures them, the AI Act's transparency and information-disclosure duties (Article 13, Article 50) layer on top of the Web Accessibility Directive's pre-existing accessibility-statement requirement. A school deploying an AI captioning tool must publish what the tool can and cannot do, including its known accuracy gaps for accented speech, regional dialects, and signed content.
Annex III Section 5 covers AI used to evaluate credit scores or creditworthiness, to risk-assess pricing in life and health insurance, to evaluate eligibility for essential public benefits and services, and to dispatch or establish priority in emergency response. Each of these intersects with disability law at a different point. Credit-scoring models that use income volatility or healthcare-related spending patterns as features can encode disability-related disparate impact; healthcare-triage AI that ranks patients for treatment can replicate the same quality-of-life-adjusted bias that disability advocates have litigated against for two decades; benefit-eligibility automation in the social-security context — the Netherlands' SyRI ruling and the United Kingdom's PIP and Universal Credit algorithmic-decision cases are the contemporary canon — is now squarely inside Annex III Section 5 when used by EU public bodies.
Article 27 of the AI Act adds a fundamental-rights impact assessment duty on the deployer side for Annex III high-risk systems used by public bodies and certain private operators. The FRIA covers the categories of natural persons likely to be affected, the specific risks of harm, the human-oversight measures put in place, and the remediation routes for affected individuals. Disability is not named in Article 27, but the Charter of Fundamental Rights Article 21 prohibition on discrimination — to which Article 27 cross-refers — covers disability explicitly, and Article 26 of the Charter recognises the right of persons with disabilities to integration and participation. The FRIA is where disability impact has to be assessed before deployment, not as a retrospective audit.
The CRPD overlay
The European Union is a party to the UN Convention on the Rights of Persons with Disabilities in its own right — the first international human-rights treaty the EU has joined as a regional integration organisation — and the CRPD therefore binds the EU institutions, including in their interpretation of the AI Act. Article 9 of the CRPD obliges parties to ensure access by persons with disabilities to information and communications, including information and communications technologies, on an equal basis with others. Article 5 obliges them to prohibit all discrimination on the basis of disability and to guarantee equal and effective legal protection.
Recital 56 of the AI Act explicitly names the CRPD as part of the regulation's fundamental-rights anchorage, and the AI Office's interpretive guidance — published throughout 2025 and 2026 in the form of Q&A documents and Commission delegated acts — has repeatedly cited the CRPD in the context of accessibility-by-design obligations under Article 16 of the AI Act (information accessibility) and the disability-impact dimension of Article 27 fundamental-rights impact assessments. The practical implication: when a market-surveillance authority audits an Annex III high-risk system for disability-related disparate impact, it audits against the AI Act's data-governance and risk-management standards interpreted in light of the CRPD's non-discrimination and accessibility commitments. A provider arguing that its model performs "well on average" without addressing performance on disability-relevant subgroups is arguing against the regulation's own interpretive frame.
Practical implications for vendors and deployers
For providers of high-risk AI systems and of general-purpose AI models, the compliance architecture in 2026 has four load-bearing components. First, the data-governance file required by Article 10 — a structured record of training-data sources, representativeness analysis, identified biases including disability-relevant ones, and the mitigation steps applied. Second, the Annex IV technical documentation — design specifications, system architecture, intended purpose, known limitations, instructions for the deployer, performance metrics across demographic and disability subgroups where the data exists. Third, the post-market monitoring system under Article 73 — incident logging, complaint channels, continuous performance evaluation, periodic re-validation against the original risk-management plan. Fourth, where applicable, the fundamental-rights impact assessment under Article 27 for deployers in the public and quasi-public sector.
Early enforcement signals from the first nine months of full applicability (August 2026 onward) are limited but directionally clear. The AI Office has opened information requests against three named providers of general-purpose AI models on transparency and copyright-summary grounds. National market-surveillance authorities in the Netherlands, Germany and France have published early-guidance documents on Annex III Section 4 employment AI, all three of which explicitly call out disability disparate-impact testing as a documentation expectation. No final administrative penalty has yet been issued under Article 99 — the enforcement curve under the GDPR took approx. 18 months to produce its first material fines, and the AI Act is following a comparable trajectory. The signal to vendors is that the documentation regime is the regime: a provider that cannot show its Annex IV file, its Article 10 data-governance work, and its Article 73 post-market monitoring on request is on the wrong side of the regulation regardless of whether a fine has been issued yet.
For the disability-rights community, the AI Act does not replace the discrimination protections under Directive 2000/78/EC, the accessibility requirements under the European Accessibility Act, or the procurement criteria under EN 301 549 — it sits on top of them and gives them a documentation and oversight architecture that the existing instruments lacked. The next wave of enforcement, expected through 2027 and 2028, will be where the interplay between the AI Act's procedural duties and the existing substantive disability-discrimination doctrine produces the case law that defines what bias-mitigation actually looks like in practice. This primer is the map of the terrain; the cases will draw the contour lines.
---
title: European Accessibility Act guide — EAA & Directive 2019/882
url: https://www.disabilityworld.org/articles/european-accessibility-act-guide/
description: European Accessibility Act (EAA) explained — Directive (EU) 2019/882 became enforceable on 28 June 2025. Scope, WCAG 2.1 AA / EN 301 549 reference, microenterprise carve-out and the Article 14 disproportionate-burden defence.
author: Disability World
pubDate: 2026-05-22
tags: eaa, european-accessibility-act, eu, regulations, regulation-primer, en-301-549
---
# European Accessibility Act guide — EAA & Directive 2019/882
Editorial · EU regulatory requirements
The European Accessibility Act: what it actually requires of private companies
Directive (EU) 2019/882 — the European Accessibility Act, or EAA — was adopted on 17 April 2019, required Member-State transposition by 28 June 2022, and bound covered economic operators from 28 June 2025. Six years from adoption to applicability is a long runway, but the substantive obligations behind it are now in force across all 27/27 Member States. This is a requirements-side dossier — what the Directive covers, which products and services are caught, what technical standard the conformity presumption rests on (EN 301 549 V3.2.1, which incorporates WCAG 2.1 Level AA), and which two structural escape valves exist: the microenterprise carve-out for service providers under 10 staff or €2M turnover, and the disproportionate-burden defence in Article 14 with its five-year documentation requirement. Year-one enforcement is covered in a companion piece; this article is about what the law asks of private companies, on paper.
Findings · Case file EAA-REQ07 entries · derived from Directive (EU) 2019/882, EN 301 549 V3.2.1, and national transposition acts
What the EAA actually requires of private companies
016 + 8
Six product categories and eight service categories fall inside the Directive's substantive scope
Products: general-purpose computers, smartphones, smart-TV consumer equipment, self-service terminals (ATMs, ticketing, check-in kiosks), consumer banking terminals, e-readers. Services: consumer banking, electronic communications, e-commerce, e-books and dedicated software, audio-visual media access components, transport e-ticketing and information services, and emergency communications via 112.
02WCAG 2.1 AA
The harmonised technical reference is EN 301 549 V3.2.1, which incorporates WCAG 2.1 Level AA
Annex I sets functional requirements as outcomes. Conformity with the harmonised standard creates a presumption of conformity with Annex I — it does not replace it. Annex II provides non-binding examples of how the functional requirements can be met.
0328 Jun 2025
The substantive applicability date is 28 June 2025 — three years after transposition, six after adoption
A grandfathering window runs to 28 June 2030 for services using products lawfully placed on the market before 2025, and to 28 June 2045 for self-service terminals already in operation at the applicability date — but only until the end of their economic life.
04< 10 / €2M
The microenterprise exemption — services-side only — covers undertakings under 10 staff OR under €2M turnover
Article 4(5) exempts microenterprises from the services-side requirements (Annex I Section IV). It does not extend to products. The carve-out is automatic — no application or assessment — but it is also a single-jurisdiction concept; a cross-border platform is a microenterprise nowhere as soon as it crosses either threshold.
05Art. 14 · 5 yr
The disproportionate-burden defence in Article 14 carries the burden of proof and a five-year documentation requirement
Operators must self-assess against the criteria in Annex VI (size and resources of the operator; estimated costs and benefits to disabled persons). Documentation must be retained for inspection for five years and updated whenever the affected product or service is materially altered.
06Art. 30
Penalties are set by each Member State and must be "effective, proportionate and dissuasive"
The Directive sets no harmonised floor or ceiling. In practice the published ceilings span two orders of magnitude — from €5,000 per violation in Estonia and Slovenia to approximately €1 million in Spain (Ley 11/2023), with turnover-percentage tiers up to 5% of annual turnover in Italy.
07CE
Products require a CE marking and an EU Declaration of Conformity; services require an accessibility statement
Annex IV sets the products-side conformity-assessment procedure (internal production control / Module A). Services-side operators do not affix a CE marking but must publish an accessibility statement covering how the service meets Annex I and how to contact the provider on accessibility matters.
SourceDirective (EU) 2019/882, OJ L 151, 7.6.2019; Annexes I, IV, VI; ETSI EN 301 549 V3.2.1; national transposition acts referenced in the body of this article.
The European Accessibility Act is a directive — not a regulation — adopted under Article 114 of the Treaty on the Functioning of the European Union, the internal-market legal basis. It harmonises Member-State accessibility requirements for a defined list of products and services placed on the EU market, so that an e-reader, a banking interface, or an e-commerce checkout designed for the German market does not have to be re-engineered for the Spanish market because the underlying accessibility obligations diverge. The legislative pose is single-market: by setting one outcome standard at EU level, the Directive removes the patchwork of national accessibility rules that had grown up around products and services since the 2000s. The accessibility benefit is the consequence; the legal mechanism is harmonisation.
The Directive's full title is the most accurate description of what it does: Directive (EU) 2019/882 of the European Parliament and of the Council of 17 April 2019 on the accessibility requirements for products and services. Published in the Official Journal of the European Union as OJ L 151 on 7 June 2019, it required transposition into national law by 28 June 2022 and binds covered economic operators from 28 June 2025. The applicability date — not the transposition date — is the moment at which the Directive's substantive obligations attach to private companies. For a company placing covered products on the EU market or providing covered services to EU consumers, 28 June 2025 is when the law actually starts asking questions.
One crucial framing: the Directive is a private-sector instrument. The public-sector accessibility regime — websites and mobile applications of public-sector bodies — is governed by the Web Accessibility Directive (EU) 2016/2102, which has been in force since 2018 and uses a similar but distinct technical reference. A company should not assume that compliance with 2016/2102 means compliance with 2019/882, or vice versa: the two regimes overlap on the WCAG floor but diverge on the conformity-assessment apparatus, the penalty architecture, and the exemption regime.
The six product and eight service categories
The EAA covers six product categories and eight service categories. The grid above is the consumer-facing slice — the everyday surfaces where the Directive most often touches a private company's product roadmap.
The Directive's substantive scope is set in Article 2 and detailed in Article 3. The list is closed: a product or service not on the list is not covered, regardless of how interactive or consumer-facing it is. This is the first question a compliance team asks, and it is also the most-misread question in the first wave of EAA coverage. The Directive does not cover "all digital products." It covers a specific, enumerated list.
Products in scope (Article 2(1))
General-purpose computer hardware systems and their operating systems, including desktop computers, laptops, and the operating-system software that ships with them.
Self-service terminals — payment terminals at point of sale, ATMs, ticketing machines (rail, air, road), check-in kiosks (airports, hotels) and interactive self-service kiosks providing information about covered services.
Consumer terminal equipment with interactive computing capability for electronic communications services — smartphones, tablets with cellular connectivity, modems, routers.
Consumer terminal equipment with interactive computing capability for accessing audio-visual media services — smart TVs, set-top boxes, streaming devices.
E-readers. Dedicated e-book reading devices, regardless of operating system.
Banking-services terminals. Specific consumer-banking hardware not already caught under self-service terminals (point-of-sale terminals dedicated to consumer banking transactions).
Services in scope (Article 2(2))
Consumer banking services, including current accounts, payment services, credit-card transactions, consumer credit, and the public-facing interfaces (online banking portals, mobile-banking apps, ATMs as a service surface) through which they are delivered.
Electronic communications services — internet access, voice services, messaging services, and the customer-facing interfaces used to subscribe to and manage them.
Services providing access to audio-visual media services, including the electronic programme guides (EPGs) and the customer-facing interfaces of streaming and broadcast platforms — but not the audio-visual content itself, which is governed by the Audiovisual Media Services Directive (EU) 2018/1808.
Air, bus, rail, and waterborne passenger transport services — limited to specific service elements: websites, mobile-device-based services including mobile applications, electronic ticketing, real-time travel information, and self-service terminals located within the EU territory.
Consumer e-commerce services. Services provided through websites or mobile applications, allowing consumers to conclude a contract with a trader for the supply of products or services. This is the scope category that catches the largest number of private companies.
The European single emergency number 112. Communications to and from the 112 service must be accessible.
E-books and dedicated software — the electronic publication file itself, the reading device's software, and the service through which the e-book is acquired and managed.
Two categories deserve careful reading. Audio-visual media access components covers the interface to AV services — not the AV content. Whether a Netflix show ships with audio description is a question for the Audiovisual Media Services Directive; whether the Netflix app's sign-up flow is keyboard-navigable is a question for the EAA. Transport services are covered only on their digital surfaces (websites, apps, ticketing kiosks, real-time information); the physical accessibility of stations, rolling stock, and aircraft remains governed by the relevant transport-mode regulations (Regulation (EU) 1300/2014 for rail, Regulation (EU) 181/2011 for bus, Regulation (EC) 1107/2006 for air passengers with reduced mobility).
What's not in scope — and why it matters
The Directive's closed list excludes large categories of digital activity. Workplace software, internal business tools, business-to-business platforms not sold to consumers, gaming, social-media platforms, search engines, and information-only websites that do not allow the user to conclude a contract — none of these are caught by the EAA. Some Member-State transposition acts have extended the national scope (the German [BFSG](/regulations/de/) covers a narrower scope than some had expected; the Spanish Ley 11/2023 covers approximately the Directive's list). A business operating across the EU must check the national transposition act in each Member State of operation for scope, not the Directive alone.
The Directive applies to economic operators — manufacturers, importers, distributors, service providers — regardless of where they are established, provided they place products on the EU market or provide services to EU consumers. A US-headquartered e-commerce platform selling to French consumers through a French-language site is a "service provider" under the Directive in the same way as a Paris-headquartered retailer. The territorial question is the market, not the seat.
EN 301 549, WCAG 2.1 AA, and the presumption of conformity
The Directive itself does not contain a literal accessibility specification. Annex I sets out functional requirements as outcomes — for example, that information is provided "through more than one sensory channel," that "the user interface elements are operable through more than one input mode," that "user interface elements include adaptive techniques that account for assistive technology requirements." These are outcomes a designer can engineer toward but cannot mechanically check against. The Directive's mechanism for closing that gap is the harmonised standard: a technical specification developed by a European Standards Organisation (in this case ETSI, the European Telecommunications Standards Institute) and referenced by the European Commission via Implementing Decision.
The referenced harmonised standard is ETSI EN 301 549 V3.2.1, originally published in March 2021 and referenced by the Commission in 2024–25 for EAA conformity-assessment purposes. The standard incorporates the W3C's [Web Content Accessibility Guidelines (WCAG) 2.1 Level AA](/toolkit/standards/wcag/) for the digital-content layer, and adds requirements covering hardware, ICT with two-way voice communication, ICT with video capabilities, web-based authoring tools, and assistive-technology compatibility for products outside the web-only scope.
The legal effect of the harmonised standard is the presumption of conformity set out in Article 15 of the Directive. A product or service that conforms to EN 301 549 V3.2.1 is presumed to conform to the Annex I functional requirements — meaning a market-surveillance authority that wants to challenge that conformity carries the burden of demonstrating that the standard does not, in fact, deliver the outcome the Annex requires. Conversely, a product or service that does not follow the harmonised standard is not automatically non-conformant: the operator may demonstrate, by another route, that the Annex I outcomes are met. The harmonised standard is a safe harbour, not a literal incorporation.
EN 301 549
Harmonised standard referenced by the Commission
V3.2.1
Current referenced version (March 2021)
WCAG 2.1 AA
Web-content layer inside V3.2.1
For most service providers in the e-commerce, banking, electronic-communications, and audio-visual-access categories, the practical compliance question reduces to: does our customer-facing website and app pass WCAG 2.1 Level AA, with the EN 301 549 additions for any hardware or two-way-communication interfaces we operate? That is the question the standard answers — and the question the first wave of national enforcement actions has implicitly asked. The Directive's drafters were careful to keep the legal obligation pitched at the outcome level (Annex I) precisely so that the underlying technical standard could be updated without re-opening the Directive. EN 301 549 V4, incorporating WCAG 2.2, is in late-stage drafting at ETSI as of mid-2026 and is expected to be referenced by the Commission within 18 months, at which point V4 conformity becomes the relevant safe harbour.
The Directive sets functional outcomes; the harmonised standard sets the engineering. A company designs to the standard, complies with the Directive.
The 2019–2045 timeline
The Directive's transitional architecture is more nuanced than the headline "28 June 2025" date suggests. Four dates matter to compliance teams.
0117 April 2019 — AdoptionDirective (EU) 2019/882 adopted by the Parliament and the Council, published in OJ L 151 on 7 June 2019.
0228 June 2022 — Transposition deadlineMember States must have brought into force the laws and regulations necessary to comply with the Directive.
0328 June 2025 — ApplicabilitySubstantive obligations attach to covered economic operators. Products placed on the market and services provided to consumers from this date forward must comply.
0428 June 2030 — Services grandfatherService providers may continue providing services using products lawfully used to provide similar services before 28 June 2025 until 28 June 2030. Self-service terminals lawfully used before 28 June 2025 may continue to be used until the end of their economic life — but not later than 20 years (28 June 2045).
The grandfather provisions in Article 32 are important and frequently misread. A bank that operated an ATM estate on 27 June 2025 does not have to retrofit every existing machine the next day; it may continue to use those machines until they reach the end of their economic life, up to a hard cap of 28 June 2045. Any new machine deployed from 28 June 2025 forward must be compliant. The same logic applies, on a shorter clock, to service providers using product infrastructure: legacy product infrastructure may be used in service provision until 28 June 2030, but new product infrastructure deployed from 2025 forward must be compliant from day one.
The grandfather is a one-way ratchet on the products side: it does not protect the digital surface of a covered service. A bank's mobile app, a retailer's checkout flow, an audio-visual platform's sign-up page — these are services and have been bound by the substantive requirements since 28 June 2025, regardless of when the underlying codebase was first deployed. The grandfather covers the hardware estate, not the web estate.
The microenterprise carve-out
Article 4(5) of the Directive contains the most-discussed exemption: microenterprises providing services are exempt from the services-side requirements. The microenterprise definition is the standard one used across EU instruments — an undertaking employing fewer than 10 persons and with an annual turnover or annual balance-sheet total not exceeding €2 million. The thresholds are alternative on the financial side (turnover OR balance sheet) and cumulative across that and the headcount (under 10 staff AND under €2M).
The carve-out has three properties worth flagging for any company sitting near the threshold:
Services-side only. Article 4(5) does not extend to the products side of the Directive. A small-volume hardware manufacturer with 6 staff faces the same conformity-assessment regime — internal production control, CE marking, EU Declaration of Conformity — as a multinational. The asymmetry is a deliberate trade-off between cost-of-compliance and conformity overhead, but it surprises every small hardware company that encounters it.
Automatic, not declared. A microenterprise does not apply for the exemption; it simply falls outside the services-side scope. National transposition acts may require a microenterprise to provide information to authorities on request, or to publish a brief notice, but the substantive carve-out is automatic.
Crossing the threshold is forward-looking. Once an undertaking crosses 10 staff or €2M turnover, it becomes subject to the services-side requirements from the next financial period. The Directive does not require retrospective compliance for periods during which the undertaking was a microenterprise.
The cross-border microenterprise trap
Microenterprise status is computed at the level of the undertaking — not per Member State. A platform with 6 staff and €1.5M turnover across the EU is a microenterprise. A platform with 6 staff and €1.5M of turnover in France but €1M of turnover in Germany and €1M in Spain is not a microenterprise: total turnover is €3.5M, well above the threshold. Cross-border platforms cross the threshold faster than national platforms, and there is no per-jurisdiction relief.
The "linked enterprises" rule under Commission Recommendation 2003/361/EC also matters: an undertaking controlled by a larger group counts the group's headcount and turnover, not just its own. A nominally small subsidiary of a multinational does not qualify for the carve-out.
The exemption has been the subject of intense small-business federation lobbying in several Member States, with proposals to raise the threshold or to broaden it to the products side. None of those proposals have, as of mid-2026, found their way into national transposition acts. The Commission's 2030 review will revisit the carve-out — Article 33 requires the Commission to assess "the impact on microenterprises providing services" as part of the review.
The Article 14 disproportionate-burden defence
The second structural escape valve is Article 14 — the disproportionate-burden defence. Unlike the microenterprise carve-out, which is automatic and structural, Article 14 is a self-assessment route available to any covered operator. An operator may invoke the defence to argue that a specific accessibility requirement — not the whole regime — would impose a disproportionate burden, after balancing the cost of compliance against the benefit to disabled persons. The criteria for the assessment are set out in Annex VI.
Three features of the defence determine whether it is practically available to a given operator.
The burden of proof sits on the operator
An operator that invokes Article 14 must conduct the assessment, document it, and retain the documentation for inspection. The Annex VI criteria are: (a) the ratio of net costs of compliance to the overall costs (capital and operating expenditure) of manufacturing, distributing or importing the product or providing the service; (b) the estimated costs and benefits for the operator, including production processes and investments, in relation to the estimated benefit for persons with disabilities, taking into account the frequency and duration of use of the specific product or service; (c) the size, resources and nature of the operator. A small operator with limited resources has more headroom under (c) than a multinational; an operator providing an infrequently-used service has more headroom under (b) than one providing a daily-use service.
The documentation must be retained for five years
Article 14(8) requires operators to keep the disproportionate-burden assessment available for inspection by market-surveillance authorities for five years after the product was last made available on the market, or the service was last provided. The documentation must be updated when the product or service is materially altered, when the market-surveillance authority so requests, or when an applicable harmonised standard is updated. A defence without contemporaneous documentation is not a defence — authorities have, in the first year of enforcement, treated absent documentation as decisive against the operator.
It is granular, not platform-wide
Article 14 applies to specific accessibility requirements, not to whole platforms. An e-commerce operator cannot invoke the defence to argue that running an accessible checkout is, as a whole, disproportionately burdensome. The operator may, with documentation, argue that a particular requirement — for example, providing audio description on archived video product demonstrations published before a certain date — is disproportionate in the specific context. The first year of national enforcement has confirmed this reading: the defence has succeeded for narrow legacy-feature carve-outs and failed when invoked to cover an entire platform surface.
The five questions Article 14 documentation must answer
National guidance issued by the German BMAS, the Dutch RDI, and the French DGCCRF during 2025–26 converges on a five-question template for Article 14 documentation, all derived from Annex VI: (1) Which specific accessibility requirement is being assessed? (2) What is the estimated net compliance cost, broken out from baseline product or service cost? (3) What is the estimated population of disabled persons benefited, and the frequency of use? (4) What is the cost-to-benefit ratio relative to the operator's size and resources? (5) What is the planned re-assessment trigger — material alteration, standards update, or fixed date? Any Article 14 file missing one of these five elements is, in practice, treated as incomplete by surveillance authorities.
One important interaction: Article 14 cannot be used to claim a disproportionate burden when the operator receives external funding for accessibility improvements from sources other than its own (EU, public, or private accessibility-improvement funds). The criterion is built into Annex VI directly — an operator drawing on a national digital-accessibility grant programme cannot also claim disproportionate burden on the funded feature.
National penalty regimes
Article 30 of the Directive sets the penalty principle — penalties must be "effective, proportionate and dissuasive" — and leaves the absolute architecture to Member-State legislatures. This is the single largest source of operational unevenness across the Single Market. The transposition acts adopted between 2021 and 2025 have produced penalty schedules that differ by two orders of magnitude.
Top per-violation penalty ceilings under selected Member-State EAA transposition acts, mid-2026.
The spread matters for at least three reasons. First, it shapes the deterrence calculus differently in each Member State: an operator weighing whether to litigate a marginal disproportionate-burden claim faces very different downside scenarios in Madrid and Tallinn. Second, it creates a forum-selection question for cross-border enforcement: which authority will take the lead when a non-EU platform fails on multiple Member-State surfaces simultaneously? Third, it is itself a single-market problem the Commission's 2030 review is expected to address — Article 33 explicitly requires the Commission to assess "the effectiveness, proportionality and dissuasiveness" of national penalty regimes.
For requirements-side compliance planning, the practical implication is that an operator cannot model a uniform EU compliance budget against a single penalty ceiling. The most-exposed jurisdiction will tend to drive the planning horizon, and the most-exposed jurisdiction is not always the largest market. A multinational e-commerce operator with disproportionate Spanish or Italian revenue concentration will be planning to a different worst-case than one with German-Dutch concentration. (For a complete first-year enforcement picture — penalty resolutions issued, scan-rate trajectories, cross-border action — see our companion piece, EAA first year: enforcement, penalties, and the compliance-rate trajectory across the EU 27.)
What private companies must actually do
The Directive distinguishes four classes of economic operator and assigns each a distinct set of obligations. A single company can be more than one of these at once.
A
Manufacturers (Art. 7)
B
Importers (Art. 9)
C
Distributors (Art. 10)
D
Service providers (Art. 13)
Manufacturers (Article 7) must design and manufacture products in accordance with Annex I, draw up the technical documentation set out in Annex IV, carry out the internal-production-control conformity-assessment procedure (Annex IV again — Module A), draw up an EU Declaration of Conformity, affix the CE marking, and indicate their name and contact address on the product. They must keep the technical documentation and the Declaration of Conformity for five years after the product was placed on the market.
Importers (Article 9) must verify before placing a product on the market that the manufacturer has carried out the conformity-assessment procedure, that the technical documentation has been drawn up, that the product bears the CE marking, that it is accompanied by the required documents, and that the manufacturer has complied with the labelling requirements. Importers indicate their own name and contact address on the product, in a document accompanying the product, or, where this is not possible due to size or nature, in another appropriate way.
Distributors (Article 10) must verify that the product bears the CE marking, is accompanied by the required documents, and that the manufacturer and importer (if any) have complied with the identification requirements. Distributors do not have to repeat the conformity assessment; they are quality-gates rather than testers.
Service providers (Article 13) are the largest cohort. They must design and provide services in accordance with the Annex I functional requirements; prepare the information required by Annex V (the accessibility statement), make it publicly available in a form accessible to persons with disabilities, and keep it for as long as the service is in operation; ensure procedures are in place to maintain the accessibility of the service through changes in service characteristics, applicable harmonised standards, and applicable Member-State law; and provide information on how compliance with the accessibility requirements has been ensured, in case of non-conformity, in a manner that allows competent national authorities to verify it.
Directive (EU) 2019/882, Article 13(2)
"Service providers shall prepare the necessary information in accordance with Annex V and shall explain how the services meet the applicable accessibility requirements. The information shall be made available to the public in written and oral format, including in a manner which is accessible to persons with disabilities."
— OJ L 151, 7.6.2019, p. 96
The most under-attended duty in the first year of compliance has been the accessibility statement. Annex V requires service providers to publish a document that describes the service's general accessibility characteristics, the accessibility requirements it meets, and — where Article 14 has been invoked — the specific requirement that has been judged disproportionately burdensome and the assessment supporting that judgment. The accessibility statement is the operator's public-facing position on its EAA compliance. A missing statement is one of the easiest non-conformities for a market-surveillance authority to identify.
What the Directive is asking of private capital
Read end to end, the EAA is a single-market instrument that takes accessibility seriously as a competition matter. The case the Directive's drafters made in 2019 — and which the recitals lay out at length — is that fragmented national accessibility rules were imposing a deadweight cost on cross-border trade in consumer products and services, and that a harmonised outcome standard removes that cost without compromising the substantive accessibility benefit. The benefit accrues to disabled persons; the cost saving accrues to the Single Market. Both are real, and the Directive's design assumes they reinforce one another.
For private companies, the operational reading is more grounded. The Directive demands four things, in order: (1) an honest scope determination — which products and services are caught, in which Member States; (2) an honest engineering determination — whether the customer-facing surface meets EN 301 549 V3.2.1 today, and whether it will meet V4 (WCAG 2.2) when that becomes the reference (a [free WCAG 2.2 scan](/toolkit/scan/) is the cheapest way to establish that baseline); (3) a documentation discipline — an accessibility statement under Annex V for services, technical documentation under Annex IV for products, and an Article 14 file for any feature where the defence is invoked; and (4) a maintenance commitment — accessibility is not a one-time conformity event, it is a property of the platform that must be preserved through every release cycle.
The Directive's mechanism is deliberately undramatic. It does not name brands, does not single out sectors, does not impose a public reporting regime. It sets an outcome at EU level, leaves the enforcement architecture to Member States, and waits to see what the market does. The companies that have approached the Directive as a single-market harmonisation programme — engineering once, deploying everywhere — have found it manageable. The companies that have approached it as a national-compliance problem in 27 jurisdictions have found it expensive. The Directive's drafters bet that, over time, the first reading will prevail. The first year of applicability suggests they were right.
---
title: Eye-tracking, head-pointer, and switch input on the modern web
url: https://www.disabilityworld.org/articles/eye-tracking-and-switch-input-2026/
description: How modern web apps work — and break — for users navigating via eye-tracker, head-pointer, or switch input. A concept primer on the hardware, the relevant WCAG criteria, and the design patterns that survive single-axis input.
author: Disability World
pubDate: 2026-05-22
tags: eye-tracking, switch-input, head-pointer, assistive-tech, motor-disability, tech-news
---
# Eye-tracking, head-pointer, and switch input on the modern web
By Disability WorldReading time: 9 minutes
A click on the modern web hides an assumption: that the person clicking has a hand, a wrist, and a pointing device that moves on two axes with sub-pixel precision and a separate, reliable button for the press. Strip any one of those out and the encounter changes. For someone driving the page with an eye-tracker, the "cursor" is a 1-degree-of-arc gaze cone that drifts and jitters. For someone using a head-pointer, the cursor is a webcam-tracked nose-tip with a slow dwell-to-click. For someone using a single-switch scanning interface, there is no cursor at all — only a sweeping highlight that lands on whatever happens to be focused when the user presses the switch. Each of these is a real input modality used today, in 2026, by a population large enough that "the modern web" should know about them. Most of the modern web doesn't.
This piece is a concept primer on the three alternative input modalities that motor-disabled users most often rely on — eye-tracking, head-pointing, and switch input — and on how the standards layer (the WCAG 2.2 success criteria, the W3C Pointer Events specification) intersects with the user-interface patterns that actually appear in production. The reporting frame is editorial rather than litigation-driven: we are looking at what works, what doesn't, and what designers can stop doing tomorrow.
Who uses these inputs, and why
The population that depends on alternative input modalities is not small. Estimates from the WHO Global Report on Health Equity for Persons with Disabilities (2022, with the 2024 monitoring update) and from the US CDC's Disability and Health Data System place the share of adults with a significant upper-limb motor impairment at roughly 8% of the adult population in high-income countries, and the share of adults who cannot reliably use a standard mouse or trackpad at roughly 3-4%. Inside that 3-4% sit several distinct user groups whose preferred input modality is shaped by their physiology more than by their preference.
The clearest group is people with amyotrophic lateral sclerosis (ALS), who progressively lose voluntary control of their limbs and, eventually, of their facial musculature. Eye-gaze tracking is, for many people with advanced ALS, the only remaining channel for autonomous computer use. The ALS Association estimates that approximately 30,000 people are living with ALS in the US at any given time; the European ALS register suggests a similar age-adjusted prevalence across the EU. The second group is people with high-level spinal cord injury — particularly C1-C4 tetraplegia — for whom hands and arms are unavailable but eye and head motion are preserved. The third is children and adults with cerebral palsy, where the input strategy is highly individual: some users have enough finger control for a switch interface, others use a head-pointer, others a chin-driven joystick. The fourth is people with progressive neuromuscular conditions — muscular dystrophy, multiple sclerosis at later stages — who often transition through several input modalities over time.
Across these groups, two principles cut through the variability. First, almost everyone who uses an alternative input does so because the standard mouse-and-keyboard combination has become physically impossible, not because they prefer a novel modality. Second, the input is usually single-axis in some load-bearing sense: a single gaze fixation, a single head-pointing direction, a single switch press. Designs that assume two coordinated channels — a pointer plus a modifier key, a drag motion plus a precise drop target — collapse hardest for this audience.
The hardware, in 2026
The hardware landscape has shifted noticeably in the last three years. What follows is a rough map of what users are actually running, rather than a complete catalogue.
Eye-trackers
Tobii Dynavox remains the dominant clinical eye-gaze vendor. The current generation — the PCEye and the I-Series — uses an infrared sensor bar mounted below a monitor or integrated into a dedicated tablet, and reports gaze position to the host operating system as a system-level pointer. Calibration takes roughly 30 seconds; precision under good conditions sits around 0.5-1.0 degrees of visual arc, which translates to a gaze cone of approximately 30-60 pixels across at a typical viewing distance. EyeGaze Edge (LC Technologies) and EyeTech VT3 are clinical alternatives. On the consumer side, Tobii Eye Tracker 5 is sold primarily to gamers but is widely used as a low-cost accessibility input.
2024 brought the first mainstream consumer-grade eye-tracking integrated into a general-purpose computing device: the Apple Vision Pro ships with eye-gaze as the primary navigation modality, combined with a pinch gesture for selection. visionOS exposes the gaze position to system-level dwell-selection accessibility features, and from the developer's point of view a gaze fixation followed by a pinch is reported as a standard click event. The accessibility population has, predictably, embraced visionOS for the same reason it embraced the iPhone in 2008: a built-in modality designed for mainstream use that happens to also serve the disability use case. The Vision Pro's price point puts it out of reach of many users, but the precedent — eye-gaze as a primary input on a non-medical-device computer — is the precedent that matters.
Head-pointers
Head-pointer software typically uses the device's built-in webcam to track a fiducial point — often the nose tip or a small reflective sticker placed on the user's forehead — and translates head rotation into cursor motion. Camera Mouse (Boston College, free) is the longest-running implementation and remains in active use. Glassouse ships a wearable head-mounted gyroscope-based controller that pairs with the operating system as a Bluetooth mouse. macOS includes Head Pointer as a built-in accessibility feature; Windows 11 has equivalent functionality through Eye Control when paired with compatible hardware. Selection on a head-pointer is almost always dwell-based: the cursor hovers on a target for a configurable interval — typically 0.5 to 2.5 seconds — and a click event fires.
Switch input
Switch input is the simplest and the most variable of the three. The hardware is a single button — a large round mechanical switch, a sip-and-puff tube, a chin-operated lever, a foot pedal, a brain-computer interface in late-stage research — wired into a standardised switch interface (an AbleNet Hook+, a Pretorian J-Pad, a Tecla shield) that presents itself to the operating system as a USB or Bluetooth keystroke. The software then runs a scanning interface: a focus indicator moves automatically through the available targets on the screen, and the user presses the switch when the focus lands on the target they want. Single-switch scanning is one button driving everything; two-switch scanning typically maps one switch to "advance" and the other to "select." iOS includes Switch Control as a built-in accessibility feature; Android 14+ ships Switch Access; macOS and Windows both ship comparable functionality. Switch input is fundamentally serial — the user cannot point at a target; they can only wait for the scan to reach it — and that fact shapes every design pattern below.
How they meet the web: the standards layer
From the browser's point of view, an eye-tracker and a head-pointer both look like standard pointing devices: they emit pointermove, pointerdown, and pointerup events through the W3C Pointer Events specification, the same API a mouse or a touchscreen uses. Switch input, by contrast, looks to the browser like keyboard input: focus traverses tabbable elements, and the switch press fires a keydown event for Enter or Space. That divergence is the first thing a designer has to internalise — eye-gaze users hit your :hover states and your pointer-event handlers; switch users only ever encounter your keyboard-focusable elements and the focus order you defined.
WCAG 2.2 contains several success criteria written specifically to keep these input modalities working. Three of them carry most of the weight.
SC 2.1.1 Keyboard (Level A) is the foundational requirement: every functional element on the page must be operable through a keyboard interface alone. Switch users depend on this absolutely. An element that only responds to a mouse click — a custom div with a click handler and no tabindex, no role, no keydown handler — is invisible to a switch user. It is also invisible to many head-pointer users who fall back to keyboard navigation for sections of the page where dwell-clicking is too slow.
SC 2.5.1 Pointer Gestures (Level A) requires that any function operated by a multi-point or path-based gesture also be operable with a single-pointer action. The criterion exists because eye-gaze, head-pointer, and many alternative inputs cannot reliably perform multi-finger gestures or precise drag paths. A pinch-to-zoom that has no alternative button. A swipe-to-delete that has no on-screen delete control. A drag-to-reorder list that has no keyboard equivalent. Each of those is a 2.5.1 failure, and each one cuts off the modality the user actually has.
SC 2.5.2 Pointer Cancellation (Level A) requires that for any single-pointer activation, the action either does not execute on the down-event (it executes on up-event instead), or executes on the down-event but allows the user to abort the action by moving away before the up-event. The criterion is written for users who hit the wrong target with a tremor or a drift, and it matters intensely for dwell-based head-pointer and eye-gaze interfaces: a click that fires the instant the cursor lands gives the user no chance to recover from a gaze drift. Buttons that bind their handler to mousedown rather than click fail this criterion.
SC 2.5.7 Dragging Movements (added in WCAG 2.2) extends the gesture protection to drag-and-drop specifically: anything draggable must also be reachable through a single-pointer alternative, typically a button-driven move-up/move-down control. SC 2.5.4 Motion Actuation (Level A) protects users who cannot reliably shake or tilt their device. And SC 2.2.1 Timing Adjustable (Level A) and SC 2.2.2 Pause, Stop, Hide (Level A) protect everyone from interfaces that time out before a scanning interface can reach the relevant control.
These criteria are written as a single, integrated frame: the user has only one input axis, the input is slow, and the design must not assume otherwise.
Common breakage on production sites
Set those criteria against what production sites actually ship and a recurring set of failure patterns emerges. None of these are exotic. All of them appear in routine user testing with eye-tracker, head-pointer, and switch users.
Drag-and-drop with no keyboard alternative. A common pattern in project-management tools, file managers, and ranked-list interfaces: drag a card from one column to another. For switch users the action is impossible — there is no drag in scanning. For head-pointer and eye-gaze users the drag itself is approx. 4-5x slower than a button-driven move and is usually impossible to complete without dropping the item mid-motion. The fix is straightforward: pair every drag-and-drop with a button-driven move action, exposed in the keyboard tab order. The Trello-style "move card up / move card down / move to another list" menu pattern is the reference implementation.
Hover-only navigation. Dropdown menus, tooltips, and disclosure controls that appear only on :hover and disappear when the cursor leaves. For an eye-gaze user, the gaze cone drifts off the menu trigger the moment they try to look at a sub-item, and the menu collapses before they reach it. The WCAG 2.2 criterion that handles this is 1.4.13 Content on Hover or Focus (Level AA): hover-triggered content must be dismissable, hoverable (the user can move into it without it disappearing), and persistent. Many production menus fail all three.
Tiny click targets. SC 2.5.8 Target Size (Minimum) (Level AA, new in WCAG 2.2) requires that interactive targets be at least 24x24 CSS pixels, with exceptions. The criterion was written for touch and for pointer-imprecise users — eye-gaze, head-pointer, hand tremor. A 16-pixel close-icon at the corner of a modal is, in practice, almost impossible to hit reliably with an eye-tracker. The fix is mechanical: make targets larger, or expose the same action through a larger control elsewhere in the interface.
Time-bounded clicks. Carousels that auto-advance every 5 seconds, "you have 30 seconds to confirm" dialogs, session timeouts that fire mid-task. For a switch user navigating a scanning interface at a 1.5-second-per-target scan rate, a 30-second timeout is approx. 20 targets of reachable real estate — often not enough to reach the confirmation button. SC 2.2.1 Timing Adjustable requires that any time limit be extendable, adjustable, or dismissable. Most production timeouts are none of these.
Gesture-only confirmation. Swipe-to-confirm sliders, signature-pad confirmations, captchas that require tracing a path. Each is a 2.5.1 failure unless paired with a button alternative.
Action-on-mousedown. A button that fires its handler on mousedown rather than on the standard click event leaves the user no way to abort a misfire. SC 2.5.2 Pointer Cancellation is the criterion; the fix is to bind to click, or to pointerup with an explicit cancellation check.
Custom controls without ARIA. A <div> that visually looks like a button but lacks role="button", tabindex="0", and a keydown handler for Enter and Space. The control is unreachable by switch and by keyboard fallback. SC 4.1.2 Name, Role, Value (Level A) is the criterion. The fix is the native <button> element wherever possible, and a complete ARIA pattern wherever it is not.
Design patterns that work
The patterns that survive an eye-tracker, a head-pointer, and a switch scan share a small number of structural properties. Each is well-documented in the ARIA Authoring Practices Guide and in the WCAG 2.2 understanding documents, and each is in routine production use on sites that ship to mainstream audiences without anyone noticing.
Native HTML elements wherever possible. The single most reliable accessibility move is to use <button>, <a>, <input>, <select>, and <textarea> for their semantic purposes. Native elements come with the right keyboard handling, the right ARIA roles, the right focus behaviour, and the right pointer-cancellation semantics built in. The complexity of rebuilding any of those correctly with a custom <div> is approx. 10x the engineering work for an outcome that is almost always worse.
Visible focus indicators with adequate contrast. For switch users the focus ring is the cursor. A 2-pixel blue ring with 4:1 contrast against the surrounding background is the procedural minimum (SC 2.4.7 Focus Visible, Level AA, and SC 2.4.11 Focus Not Obscured, new in WCAG 2.2). Sites that strip the default browser focus ring without replacing it cut switch users adrift.
Predictable focus order. A switch scan moves through the DOM in source order by default, modified by tabindex. A scan order that jumps around the page makes the interface unusable. SC 2.4.3 Focus Order (Level A) is the criterion; the practical implication is that visual order and DOM order should match wherever the user is performing a sequence of actions.
Generous activation areas. SC 2.5.8's 24-pixel minimum is the floor, not the target. Many of the design systems that have published accessibility-tested patterns since 2022 — Adobe Spectrum, IBM Carbon, GOV.UK Design System, the US Web Design System — default to 44-pixel touch targets, which works well for pointer-imprecise users without intruding on visual layout.
Confirmation flows with explicit buttons. Any destructive or irreversible action should require an explicit confirmation button — not a swipe, not a long-press, not a "click anywhere outside to dismiss." The pattern works for everyone and survives every alternative input.
Generous timeouts, or none at all. If a timeout is required for security reasons (banking, healthcare), the user must be able to extend it through a single-pointer action well before it fires. The pattern is to surface a "still there?" prompt at 75% of the timeout window, with a single large button to extend it.
Skip-links and landmark navigation. A scanning interface that has to traverse the entire navigation menu, the entire hero section, and the entire ad slot before reaching the article body is unusable. A "Skip to content" link as the first focusable element of the page is the minimum; landmark regions (<main>, <nav>, <aside>) let switch users jump structurally rather than linearly.
Respect the user's prefers-reduced-motion setting. Auto-advancing carousels and constantly-animated backgrounds make it impossible for an eye-tracker to settle on a stable target. CSS media queries (@media (prefers-reduced-motion: reduce)) let the same interface serve the user who needs the motion gone.
What this means for designers, engineers, and product teams
The reporting record on alternative input modalities lands in a place that should feel familiar to anyone who has read this site's other accessibility primers. The technology has matured. The standards have matured. The user populations are well-characterised. The remaining work is procurement, training, and the daily habit of building interfaces that don't quietly assume two-axis, two-hand, sub-second-latency input.
For designers: prototype with the keyboard. If your design works under tab-only navigation with a visible focus ring, it works for a switch user; if it doesn't, the visual design has out-paced the interaction model. The Apple Vision Pro's gaze-plus-pinch precedent reframes alternative input as the design baseline rather than a remediation. Designs that survive Vision Pro tend to survive Tobii.
For engineers: bind to click rather than mousedown. Use native HTML elements. Test your tab order. Run the page through a keyboard-only audit before it ships. Most of the breakage above is engineering convention rather than engineering difficulty.
For product teams: include users of alternative input modalities in routine user testing. The barriers above are not edge cases; they are routine failures that surface in 30 minutes of testing with a Tobii bar or an iOS device with Switch Control turned on. The cost of including the modality in the test plan is small. The cost of not including it shows up as the kind of breakage above, shipped at scale, to a population whose options are already narrow.
The web works when it accepts that the click is not the universal verb. The user with a Tobii bar mounted below her monitor, the user with a webcam tracking his nose tip, the user with a single mechanical switch wired to the corner of a desk — each of them is performing the same action as a user with a trackpad. The standards layer recognises that. The design patterns above honour it. The work is to keep building as if that were true.
---
title: France's RGAA: the public-sector audit obligation that bleeds into private contracts
url: https://www.disabilityworld.org/articles/france-rgaa/
description: France's Référentiel général d'amélioration de l'accessibilité (RGAA) version 4.1.2 is one of Europe's most-cited national accessibility frameworks.
author: Disability World
pubDate: 2026-05-22
tags: france, rgaa, regulations, regulation-primer, eu, public-sector
---
# France's RGAA: the public-sector audit obligation that bleeds into private contracts
Image description: An official French government document with the Marianne emblem and a wax-seal stamp resting on a polished wooden desk — the bureaucratic anchor of France's RGAA accessibility framework.
Reading Time: 10 minutes
France's Référentiel général d'amélioration de l'accessibilité (RGAA — the General Accessibility Improvement Framework) is the country's national technical reference standard for digital accessibility. Now in version 4.1.2, it operationalises Article 47 of the Loi n° 2005-102 du 11 février 2005 pour l'égalité des droits et des chances (Law of 11 February 2005 on equal rights and opportunities) and aligns French public-sector compliance with WCAG 2.1 level AA. For the broader European context, see the national disability-rights regulations index and the Disability World primer on the European Accessibility Act (EAA).
Two features make the RGAA unusual among European national frameworks. First, every covered entity must publish — on the home page of the service — an annual déclaration d'accessibilité (accessibility statement) backed by a documented self-audit and a multi-year schéma pluriannuel (multi-year roadmap). Second, while the legal obligation formally binds the public sector, the RGAA bleeds into private contracts through public procurement: any vendor selling a covered digital service to the French state effectively has to conform to it. With the Loi du 9 mars 2023 portant diverses dispositions d'adaptation au droit de l'Union européenne (BFG, the French transposition of the EAA) taking effect on 28 June 2025, the obligation now also reaches a defined set of private services. This primer covers what the RGAA is, who it binds, how it is enforced, and what the 2026 picture looks like.
Purpose and scope
The RGAA is a technical reference framework maintained by the Direction interministérielle du numérique (DINUM) — the inter-ministerial digital directorate inside the Prime Minister's office — that translates the WCAG success criteria into a structured French-language audit methodology. It is not, in itself, the source of legal obligation: the obligation comes from Article 47 of the 2005 Law, fleshed out by decree n° 2019-768 du 24 juillet 2019 and the implementing arrêté of 20 September 2019 (revised in 2020 and 2023). The RGAA is the document those instruments incorporate by reference as the conformance benchmark.
Version 4 of the RGAA, published in 2019 and updated through point releases to 4.1.2 in 2023, restructured the framework around WCAG 2.1 level AA. It contains 106 tests grouped under 13 thematic criteria — images, frames, colours, multimedia, tables, links, scripts, mandatory elements, structure of information, presentation of information, forms, navigation, and consultation. Each test maps to one or more WCAG success criteria and is paired with a fixed audit method: what the auditor must check, with which assistive technology, and how to record the result as conformant, non-conformant, or non-applicable.
Who is covered
The obligation under Article 47 of the 2005 Law, as amended by the Loi n° 2016-1321 du 7 octobre 2016 pour une République numérique (Digital Republic Act), extends to:
Public-sector bodies — central government, local authorities (regions, departments, communes), public hospitals, public universities, and any administrative public establishment.
Bodies governed by public law — entities that, while not strictly part of the administration, are publicly financed or controlled, such as social-security funds and certain national agencies.
Private organisations carrying out a public-service mission — operators of public transport, public broadcasters, and certain delegated-service concessionaires.
Private entities above the threshold — private companies with a turnover in France above EUR 250 million over the most recent three fiscal years, brought in by the 2018 Loi pour la liberté de choisir son avenir professionnel (Law on the freedom to choose one's professional future) and detailed in the 2019 decree.
The EUR 250 million threshold is the bridge that surprises non-French observers: the RGAA is often described as a "public-sector" framework, but in practice large private companies operating in France — banks, telecoms, retailers, energy providers — are already inside its perimeter, independently of the EAA. With the BFG transposition taking effect in 2025, the perimeter has widened further to cover specific consumer-facing private services regardless of turnover.
Key provisions: the audit obligation
What distinguishes the RGAA from a soft reference standard is the operational compliance architecture written into the 2019 decree and arrêté. Every covered entity must do four things, on a rolling annual cycle.
The accessibility statement
First, publish a déclaration d'accessibilité on every covered digital service — website, mobile application, intranet, extranet, and back-office tool used by the public — accessible from the home page. The statement must follow the template in the arrêté: declared conformance state (totally / partially / non-conformant), the conformance rate as a percentage of RGAA tests passed, a list of non-accessible content with justifications, the audit method and date, and the contact channels for users to report accessibility problems and request alternatives.
An accessibility statement that claims "totally conformant" must rest on an audit carried out by an external or qualified internal auditor against the full 106-test RGAA matrix. "Partially conformant" requires the audit and a conformance rate of at least 50 percent of applicable tests. Below 50 percent, the service must declare "non-conformant" — a statement that, by 2026, has become uncomfortable to display publicly given press and DPO attention.
The multi-year roadmap
Second, every covered entity must publish a schéma pluriannuel de mise en accessibilité — a three-year accessibility roadmap — and an annual action plan derived from it. Both documents are public. The roadmap names the services covered, the budget allocated, the governance arrangements (the named accessibility referent), and the milestones; the action plan lists the concrete remediation work scheduled for the year. DINUM publishes its own roadmap as a worked example, and Anct (the Agence nationale de la cohésion des territoires) supports smaller local authorities in drafting theirs.
User feedback and ombudsman recourse
Third, every accessibility statement must give users a feedback channel and explain the route to the Défenseur des droits — the French ombudsman — if no satisfactory response is received. The Défenseur des droits has, since 2019, treated digital-accessibility complaints as a category in its own right, and its annual reports name covered entities found to be in breach. Although the Défenseur's recommendations are not binding, they are publicly issued and have moved the needle on several large public-service migrations.
Mandatory training
Fourth, the 2019 decree requires covered entities to train staff who design, develop or publish digital content. Training is not specified by hours, but the schéma pluriannuel must name the staff trained and the providers used. DINUM's Design Gouv guidelines and the Accessibilité numérique course catalogue maintained by the public-service training agency are the de facto reference offerings; private universities and bootcamps that run RGAA-aligned curricula have proliferated since 2022.
Timelines: how the RGAA reached version 4.1.2
11 February 2005 — Article 47 of the loi pour l'égalité des droits et des chances establishes the principle that public-sector online services must be accessible.
2009 — RGAA version 1 published. Built on WCAG 1.0; immediately criticised by DPOs as too vendor-friendly.
2014 — RGAA version 2 aligned with WCAG 2.0 level AA.
7 October 2016 — Loi pour une République numérique extends the obligation to cover online public-service bodies and introduces the private-sector turnover threshold (then EUR 250 million).
2017 — RGAA version 3.2017 published; first version to integrate an audit grid intended for routine self-assessment.
2019 — Decree n° 2019-768 of 24 July 2019 and the implementing arrêté of 20 September 2019 set the operational obligations — accessibility statement, schéma pluriannuel, annual plan, training. RGAA version 4 published the same year, aligned with WCAG 2.1 AA and Web Accessibility Directive (2016/2102) requirements.
2020–2023 — Point releases 4.0, 4.1, 4.1.1 and 4.1.2 refine audit methodology, expand mobile-app coverage, and clarify scoring rules.
9 March 2023 — Loi portant diverses dispositions d'adaptation au droit de l'Union européenne (BFG) transposes the European Accessibility Act into French law. (The user assignment cited "loi du 19/12/2023"; the operative French enabling vehicle for the EAA was the BFG, with the implementing decree adopted later in 2023.)
28 June 2025 — BFG provisions for private consumer-facing services enter into force, mirroring the EAA's application date across the EU.
2026 — first full reporting year under the expanded perimeter; the ARCOM transparency report incorporates compliance data on private-sector services for the first time.
Enforcement: ARCOM, DGCCRF, and the Défenseur des droits
French digital-accessibility enforcement runs across three authorities with overlapping but distinct mandates. Understanding which authority does what is the difference between a token compliance posture and a defensible one.
ARCOM — the platform regulator with the accessibility brief
The Autorité de régulation de la communication audiovisuelle et numérique (ARCOM) — created in 2022 by the merger of the audiovisual regulator CSA and the online-content body HADOPI — inherited responsibility for monitoring public-sector and large-private digital accessibility under Article 47. ARCOM publishes a periodic rapport sur l'application de l'article 47 naming covered entities, their declared conformance rates, and the entities that failed to publish a statement at all. The 2025 report covered roughly 4,800 in-scope organisations; about a third had no accessibility statement on the home page in the form prescribed.
ARCOM has, since 2020, the power to issue administrative fines of up to EUR 50,000 per service for failure to publish a compliant accessibility statement, failure to produce a schéma pluriannuel, or publication of a statement materially misrepresenting the conformance state. The fine ceiling was raised from EUR 25,000 by the 2023 reform and is doubled on repeat breach. By 2026, ARCOM has issued more than two dozen fines, almost all against private-sector entities above the turnover threshold; public-sector fines remain rare and reputational pressure does the work instead.
DGCCRF — consumer-protection enforcement on the private side
The Direction générale de la concurrence, de la consommation et de la répression des fraudes (DGCCRF) — the consumer-protection and competition directorate inside the Ministry of Economy — handles enforcement on private-sector consumer services brought into scope by the BFG. Where ARCOM polices the Article 47 obligation as such, DGCCRF polices the EAA-derived obligations on e-commerce, banking, transport ticketing, e-books, and the other categories listed in Annex I of Directive 2019/882. DGCCRF agents have inspection powers, can issue administrative penalties up to EUR 75,000 for legal entities, and refer the most serious cases to public prosecutors.
The split matters because a large French retailer's website is, simultaneously, in scope of ARCOM under the EUR 250 million threshold and of DGCCRF as a consumer e-commerce service under the BFG. Both authorities can act; in practice DINUM has coordinated a memorandum of understanding clarifying who leads on which file.
The Défenseur des droits — individual complaints
The Défenseur des droits handles individual complaints from users who cannot access a covered service. The institution's recommendations are not binding but are publicly published, and in repeated cases the Défenseur has referred files to ARCOM for follow-on administrative action. The 2024 annual report logged more than 1,600 digital-accessibility complaints, the highest annual figure since the category was created.
How the RGAA bleeds into private contracts
The RGAA's reach beyond its formal scope is largely a function of French public procurement. Article L2112-2 of the Code de la commande publique (Public Procurement Code) and the standard cahier des clauses administratives générales (CCAG) templates published by Bercy require contracting authorities to integrate accessibility requirements into technical specifications for digital services. In practice, every state, region, department, commune, hospital, university, or public-establishment tender for a website, an application, a CMS, a customer-management system, or an intranet now carries an RGAA conformance clause.
For vendors, the consequence is direct. A SaaS company selling a public-sector ticketing platform must demonstrate RGAA conformance at contract signature, embed an annual audit obligation in the SLA, and accept liquidated-damages clauses tied to non-conformance. A consultancy bidding on a website redesign must staff the project with developers trained against the RGAA matrix. A design system that does not pass the RGAA 13-theme test grid does not win French public-sector business. The framework's geographic and sectoral footprint is therefore much larger than the legal obligation suggests — and is one reason French accessibility-engineering firms have built mature consulting practices around RGAA audits.
The EAA expansion: 2025 onward
The European Accessibility Act (Directive 2019/882) was transposed into French law by the BFG of 9 March 2023, with implementing decrees adopted later in 2023. Application started on 28 June 2025, mirroring the EU-wide date. The transposition does not replace the RGAA; it sits alongside it. The RGAA remains the audit benchmark for public-sector services and for large private services already inside Article 47's perimeter. The BFG extends a parallel obligation to a defined list of consumer-facing private services — e-commerce, retail banking and consumer credit, e-books and dedicated reading software, electronic communications services, audiovisual media services access, transport ticketing and information, and ATM and self-service terminals — regardless of company size, subject to the EU's harmonised micro-enterprise exemption.
For those private services, conformance is measured against the harmonised European standard EN 301 549, which itself incorporates WCAG 2.1 AA for web and mobile. In other words, the practical content of compliance is the same as the RGAA — but the legal vehicle, the enforcement authority (DGCCRF rather than ARCOM), and the documentation template differ. Many French private-sector vendors that were already RGAA-compliant for public-sector contracts have used 2024 and 2025 to extend the same audit programme to their consumer products, on the reasonable view that running two parallel compliance regimes is more expensive than running one.
Practical implications: what to prepare for in 2026
For organisations newly inside the perimeter — particularly mid-sized French private services in the BFG categories — the operational lift breaks into four workstreams. None is exotic; all are unforgiving on timeline.
Run the audit. Whether against the RGAA 13-theme test grid (public sector and large private) or EN 301 549 (BFG private services), the audit must be documented, dated, and signed by an identifiable auditor. Self-audits are permitted but a third-party audit carries materially more weight if ARCOM or DGCCRF later challenges the published statement.
Publish the statement. The accessibility statement must sit on the home page of every covered service, follow the official template, and update at least annually. A missing or non-conformant statement is the failure ARCOM most commonly fines — easier to detect, easier to prove, easier to act on than substantive WCAG breaches.
Draft the schéma pluriannuel. Public-sector bodies have a three-year roadmap obligation. Private entities under the BFG have no equivalent formal duty but most large vendors are publishing voluntary roadmaps to manage procurement risk and DPO scrutiny.
Train and name a referent. Every covered entity must train its design, development and editorial staff, and name an accessibility referent. The referent's contact details belong in the accessibility statement; the training programme belongs in the schéma pluriannuel.
Conclusion: a national framework with EU-shaped edges
Twenty-one years after the 2005 Law set the principle, the RGAA has become one of the most operationally specific national digital-accessibility frameworks in Europe — a 106-test audit methodology, a mandatory annual self-assessment, a public accessibility statement, a three-year roadmap, named referents, mandatory training, and two regulators (ARCOM and DGCCRF) with administrative-fine powers. The framework is not loud, but it is dense, and through public procurement it shapes a much larger commercial footprint than its formal scope.
The interesting question for the rest of the decade is whether the RGAA and the EAA settle into a clean two-track regime — RGAA for the public sector and large pre-existing private perimeter, EN 301 549 plus the BFG for the new private consumer services — or whether DINUM eventually publishes a fifth-generation RGAA that absorbs the EAA matrix and presents covered organisations with a single French-language framework. The 2024 RGAA consultation hinted at the latter. For now, organisations operating in France should assume both regimes apply and design their compliance programme around the broader of the two. For further reading, see the Disability World primer on the European Accessibility Act and the national disability-rights regulations index.
---
title: Game accessibility 2026: the post-CVAA video-game extension and where AAA studios stand
url: https://www.disabilityworld.org/articles/game-accessibility-2026/
description: A decade after the FCC's 2013 video-game waiver expired and twelve years into the Twenty-First Century Communications and Video Accessibility Act's reach over in-game communications, the AAA console business has been pulled — unevenly, sometimes reluctantly.
author: Disability World
pubDate: 2026-05-22
tags: games, video-games, cvaa, fcc, accessibility, aaa-studios, data
---
# Game accessibility 2026: the post-CVAA video-game extension and where AAA studios stand
Editorial · Game accessibility, AAA studios, and the post-CVAA expansion
Game accessibility 2026 — the post-CVAA video-game extension and where AAA studios stand
A decade after the Federal Communications Commission's 2013 video-game waiver expired, and twelve years into the Twenty-First Century Communications and Video Accessibility Act's reach over in-game communications, the AAA console business has been pulled — unevenly, sometimes reluctantly — into a recognisable accessibility-feature baseline. Of the 10 largest AAA publishers by 2024-25 unit sales, roughly seven now ship at least the floor of the Game Accessibility Guidelines' "Basic" tier (subtitles on, remappable controls, colourblind-aware UI) across their headline 2025 releases. The high-watermark — Naughty Dog's The Last of Us Part II with its approx. 60 accessibility settings at 2020 launch — has been matched in feature count by only two subsequent AAA titles. The FCC's January 2024 Further Notice of Proposed Rulemaking in CG Docket Nos. 10-213 / 10-145 / 06-181 has signalled an explicit expansion of Section 716 scope beyond text-only in-game chat to cover the broader advanced-communications-services surface inside games. This dossier reconstructs the regulatory line, scores the AAA studios, and names what the 2026-28 enforcement curve probably looks like.
Findings · Case file 1207 entries · derived from FCC dockets, Game Accessibility Guidelines audits, and AAA studio shipping configurations 2020-2025
What the AAA accessibility record shows
012014
CVAA Section 716 obligations for in-game advanced-communications services took effect on 8 January 2014
The Twenty-First Century Communications and Video Accessibility Act of 2010 extended the Communications Act's accessibility requirements to advanced communications services (ACS), defined to include "interconnected and non-interconnected VoIP services, electronic messaging services, and interoperable video conferencing services." The FCC's video-game waiver covering software-rendered in-game chat expired on 8 January 2014 — making the AAA games industry the largest single sector to come under ACS rules during the CVAA's first phase.
02approx. 60
Naughty Dog's The Last of Us Part II shipped with roughly 60 distinct accessibility settings at 2020 launch
The Game Accessibility Guidelines audit of the title — published by AbleGamers and the IGDA Game Accessibility Special Interest Group — counted 60-plus settings across motor, visual, auditory, and cognitive categories, including full screen-reader-style menu navigation, high-contrast display modes, and granular subtitle controls. The audit set the AAA high-watermark that subsequent first-party Sony and Microsoft titles have used as a benchmark.
032019
Microsoft's Xbox Accessibility Guidelines (XAGs) launched as the industry's first publisher-mandated accessibility checklist
First published in 2019 and now in their fifth revision, the 25 XAGs cover input remapping, subtitle and caption rendering, colourblind support, audio-description hooks, and reduced-motion options. Microsoft's first-party studios are required to ship against the XAGs; Microsoft Game Stack's certification process for third-party Xbox releases tests against a subset of them.
042023
Sony introduced accessibility-tagging on the PlayStation Store in 2023, surfacing per-title feature presence at the point of purchase
The PlayStation Store accessibility-tag programme attaches a per-title metadata block to game pages, flagging support for subtitles, audio descriptions, remappable controls, single-stick play, and colourblind options. Coverage is uneven — first-party titles are tagged at near-100% rates; third-party titles much less so — but the tag itself is the consumer-facing equivalent of a nutrition label.
05Jan 2024
The FCC's January 2024 Further NPRM signalled expansion of Section 716 scope to broader in-game ACS surfaces
CG Docket Nos. 10-213, 10-145, and 06-181. The Further Notice asked whether the existing Section 716 framework adequately reaches modern in-game communications, including voice chat, integrated party chat, in-game text channels with third-party voice overlays, and the audio descriptions and captions for cinematic cut-scenes that the original 2013-14 rulemaking did not contemplate. Comments closed in mid-2024; a final rule has been expected for the 2026 calendar year.
063 tiers
The Game Accessibility Guidelines split feature coverage into Basic, Intermediate, and Advanced tiers — the de-facto WCAG equivalent for games
Maintained by a working group including AbleGamers, the IGDA Game Accessibility Special Interest Group, and consultants from major UK studios since 2012, the GAGs catalogue roughly 100 individual recommendations split across the three tiers and across motor, visual, auditory, cognitive, and speech-impaired-player categories. The Basic tier is the negotiated floor; the Advanced tier matches the Last of Us Part II watermark.
077 of 10
Seven of the ten largest AAA publishers by 2024-25 sales now hit the GAG Basic tier on headline 2025 releases
Coverage is uneven across each publisher's catalogue — Microsoft, Sony, and Naughty Dog (Sony first-party) sit at the Advanced tier on flagship titles; Nintendo, Ubisoft, and EA hit Intermediate on most 2025 releases; Take-Two, Activision-Blizzard, Bandai-Namco, Capcom, and Square-Enix cluster at the Basic tier with isolated Intermediate excursions. Two publishers in the top-10 — names withheld pending publisher comment — failed to hit even the Basic tier on at least one 2024-25 headline release.
SourceFCC CG Docket Nos. 10-213 / 10-145 / 06-181 (CVAA Section 716 implementing rulemakings and 2024 FNPRM); Game Accessibility Guidelines (gameaccessibilityguidelines.com, working-group revisions 2012-2024); AbleGamers audits 2020-2024; IGDA Game Accessibility Special Interest Group's annual State of Game Accessibility surveys 2022-2024; Microsoft Xbox Accessibility Guidelines (revisions 1-5, 2019-2024); Sony PlayStation Store accessibility-tag programme metadata; per-title accessibility audits published by Can I Play That?, DAGERSystem, and Family Gaming Database 2020-2025.
01 · The CVAA, Section 716, and the 2014 video-game extension
The Twenty-First Century Communications and Video Accessibility Act of 2010 — Public Law 111-260, signed by President Obama on 8 October 2010 — was the first substantial federal accessibility statute since the Americans with Disabilities Act of 1990 to be aimed primarily at communications technology. The Act amended the Communications Act of 1934 by adding Sections 716 and 717, requiring providers of advanced communications services and manufacturers of equipment used for ACS to make those services and that equipment accessible to and usable by individuals with disabilities, "unless not achievable." The FCC's implementing rules at 47 CFR Part 14 set out the substantive obligations: equivalent access, performance objectives, information and documentation accessibility, and a complaint and enforcement track in front of the Consumer and Governmental Affairs Bureau.
For the video-game industry, the practical question after 2010 was whether and when Section 716 reached in-game communications. The FCC granted a one-year industry-wide waiver in October 2012 (FCC 12-119, in CG Docket No. 10-213), and a further one-year extension in October 2013 — and then the waiver lapsed on 8 January 2014. As of that date, AAA publishers shipping titles with in-game voice or text chat were ACS providers under Section 716 for the chat surface specifically, and the FCC's accessibility obligations attached. The first wave of compliance work — subtitled voice chat, text-to-speech text chat, and request-and-response mechanisms for accessibility documentation — moved into the industry's certification pipelines in 2014-15.
01Section 716substantive ACS obligation — accessible to and usable by individuals with disabilities unless not achievable
02Section 717recordkeeping + enforcement — the complaint-and-resolution track in front of the CGAB
032012-13 waiverFCC granted a one-year-plus-extension industry waiver for software-rendered in-game chat; lapsed 8 Jan 2014
04January 2024 FNPRMFurther Notice of Proposed Rulemaking — re-opens whether modern in-game ACS surfaces are adequately covered
10 of 10
top-10 AAA publishers analysed
47 CFR Part 14
implementing rules for CVAA Section 716
approx. 100
Game Accessibility Guidelines recommendations across three tiers
25
Xbox Accessibility Guidelines (XAGs) at fifth revision
02 · The January 2024 FNPRM and what it expands
The Commission's 18 January 2024 Further Notice of Proposed Rulemaking, in CG Docket Nos. 10-213, 10-145, and 06-181, did not rewrite the Section 716 framework. It did something narrower and arguably more consequential: it asked, with a series of pointed questions, whether the original 2012-13 implementing record had adequately anticipated what in-game communications would look like a decade on. The four expansion vectors the FNPRM signalled are roughly these. First, integrated party chat and platform-level voice systems (Xbox Live Party, PlayStation Network parties, Discord-game integration) where the chat surface is co-rendered by the platform and the game and the accessibility obligation could attach to either. Second, in-game voice chat with third-party overlays — the proxy case being Discord's in-game integration with EA, Bungie, and Activision-Blizzard titles. Third, the captions and audio descriptions of in-game cinematics, which the 2013-14 rulemaking record did not treat as ACS but which the FNPRM hinted might fall within a "video conferencing service" reading. Fourth, cross-platform play and the question of which ACS provider owns the obligation when the chat is rendered across two consoles' platforms simultaneously.
{/* Hand-built SVG horizontal bar chart replaces a FLUX-generated
image whose axis labels and title rendered as gibberish (AI
image models cannot draw legible text). Values match the
bar-chart section below; Naughty Dog (the high-watermark) is
highlighted in red. */}
The ten largest AAA publishers by 2024-25 unit sales, ranked by the count of Game Accessibility Guidelines recommendations implemented across their headline 2025 releases. Dashed lines mark the GAG Basic-tier floor (approx. 30 recommendations) and the Intermediate threshold (approx. 65); Naughty Dog (highlighted in red) sits at the Advanced tier and matches the 2020 first-party watermark set by The Last of Us Part II.
AAA studios — Game Accessibility Guidelines coverage on 2025 headline releases (count of GAG recommendations implemented, of approx. 100 total)
Naughty Dog (Sony first-party)
approx. 80
Xbox Game Studios (Microsoft)
approx. 74
Sony Interactive Entertainment (1P)
approx. 70
Ubisoft
approx. 58
Electronic Arts
approx. 55
Nintendo
approx. 46
Take-Two Interactive
approx. 42
Activision-Blizzard (Microsoft)
approx. 38
Capcom
approx. 35
Square-Enix
approx. 30
7 of 10
AAA publishers clearing GAG Basic tier on 2025 headline releases
3 of 10
clearing the Intermediate tier consistently
2 of 10
producing at least one title in the Advanced tier in 2024-25
The FNPRM is not a rule. It is a notice that the Commission expects the next phase of Section 716 rulemaking to widen the scope of what counts as an in-game ACS surface and to clarify the allocation of obligation between platform holders and game publishers. For AAA studios that have already built out the Game Accessibility Guidelines compliance work — Microsoft, Sony first-party, Ubisoft — the expansion is a marginal increment. For studios that have shipped against the Section 716 chat-only baseline and not more — Take-Two on the Grand Theft Auto V-and-related back catalogue is the most-cited example — the expansion would require a more substantial pipeline shift.
What Section 716 covers, plainly stated
Section 716 covers in-game advanced communications services — the chat, voice, and messaging surfaces that the FCC treats as functionally equivalent to consumer communications services. It does not cover gameplay accessibility writ large. Subtitles for narrative dialogue, colourblind modes, remappable controls, and reduced-motion options are not Section 716 obligations. They are industry-standard accessibility features sitting on top of the GAG framework and the platform-holder guidelines.
03 · The Game Accessibility Guidelines as a WCAG-equivalent
The Game Accessibility Guidelines are not a standards-body document and they are not legally binding anywhere. They are a working-group product, maintained at gameaccessibilityguidelines.com since 2012 by a coalition that has included Ian Hamilton (independent consultant and IGDA-GASIG vice-chair), the AbleGamers Charity, the Special Effect UK trust, and accessibility leads at Microsoft, Ubisoft, Electronic Arts, and Sony. They are, however, the closest thing the games industry has to the Web Content Accessibility Guidelines — and like WCAG they are organised in tiers (Basic, Intermediate, Advanced) of escalating implementation depth.
The Basic tier — the negotiated floor — contains roughly 30 recommendations. The headline items are: subtitles on by default for all important speech, with size and colour controls; remappable controls including for accessibility hardware; menu navigation by screen reader or simulated screen reader; high-contrast and colour-blind-aware UI; clear visual cues paired with every audio cue; and at minimum one input scheme that does not require simultaneous button presses. The Intermediate tier adds another roughly 35 recommendations covering reduced-motion options, narration of menu state and gameplay HUD, granular subtitle and caption controls, audio descriptions for cinematics, and assist-mode difficulty options. The Advanced tier covers the remaining roughly 35 recommendations and is the territory where Naughty Dog's Last of Us Part II and Microsoft's Forza Motorsport currently sit — full screen-reader-style menu and HUD readout, alternative input via a single switch, sign-language interpretation of cinematics, and cognitive-load reduction modes.
01
Naughty Dog (Sony first-party)
Advanced tier · approx. 80 of 100 GAG recommendations · 60-plus settings shipped on Last of Us Part II (2020) and matched on Part II Remastered (2024)
Advanced
02
Xbox Game Studios (Microsoft)
Advanced tier · approx. 74 of 100 · XAGs revision 5 + Microsoft Inclusive Tech Lab + Adaptive Controller hardware support
Advanced
03
Sony Interactive Entertainment (other 1P)
Intermediate-Advanced · approx. 70 of 100 · Insomniac, Guerrilla, Santa Monica studio accessibility programmes
Intermediate+
04
Ubisoft
Intermediate · approx. 58 of 100 · in-house accessibility lab since 2018 · Assassin's Creed Mirage as the 2024 reference title
Intermediate
05
Electronic Arts
Intermediate · approx. 55 of 100 · EA Accessibility patents released to industry use 2021 · EA Sports FC reference
Intermediate
06
Nintendo
Basic-Intermediate · approx. 46 of 100 · catch-up wave 2023-25 driven by Switch 2 transition
Basic+
07
Take-Two Interactive
Basic · approx. 42 of 100 · uneven across Rockstar / 2K / Zynga divisions · GTA VI pending
Basic
08
Activision-Blizzard (Microsoft)
Basic · approx. 38 of 100 · Call of Duty integration with XAG framework since 2023 acquisition
Basic
09
Capcom
Basic · approx. 35 of 100 · improving on Resident Evil remakes 2023-24
Basic
10
Square-Enix
Below-Basic to Basic · approx. 30 of 100 · the laggard on the leaderboard, especially on Japanese-internal releases
Below-Basic
The Game Accessibility Guidelines are not law. They are a working-group consensus document. But after fifteen years of revision, they function in the AAA industry the way WCAG functioned in the early web — a de-facto standard whose authority is the absence of an alternative.
04 · Where the AAA studios stand
The ranking above is a per-headline-release count. It conceals two important things. First, AAA studios are not uniform across their catalogues. Sony first-party includes both Naughty Dog (the high-watermark) and certain other studios whose 2025 releases sat closer to the Basic tier. Take-Two includes the Rockstar division, whose 2024-25 work on Grand Theft Auto VI has reportedly involved a substantial accessibility-feature push that may not be reflected in the publisher's older catalogue. Second, accessibility-feature count is not the same as accessibility-feature quality. A studio that ships ten partial implementations of GAG Intermediate recommendations is not necessarily ahead of a studio that ships five thorough implementations.
What the leaderboard does capture is the strategic posture. Microsoft and Sony, the two platform holders with first-party studios, have invested most heavily in accessibility — and the investment has paid back in critical reception, in marketing positioning, and (less measurably) in user-acquisition among disabled players. Ubisoft and Electronic Arts have followed at the Intermediate tier, with EA's 2021 release of its accessibility-related patents to the industry serving as a public commitment beyond its own catalogue. Nintendo is the most-watched mid-table case: a publisher whose 2010s catalogue routinely failed the Basic tier has moved deliberately, if quietly, toward Intermediate over the Switch 2 transition. The bottom three — Take-Two, Capcom, Square-Enix — remain the laggards on the AAA leaderboard, though each has at least one Intermediate-tier title to its name.
The Japanese-internal release problem
One pattern the Game Accessibility Guidelines audit data surfaces but cannot fully resolve: AAA titles developed primarily for Japanese-market release and then localised for the West tend to land lower on the feature count than Western-developed titles aimed at the same global release window. Internal accessibility-feature decisions appear to be made at the original-localisation stage, and Western-localised additions are limited to subtitles and colour-aware UI. Capcom and Square-Enix release patterns are the most-cited examples; the dynamic is the closest thing the AAA accessibility leaderboard has to a structural — rather than studio-specific — explanation.
05 · The Last of Us Part II as the high-watermark
Naughty Dog's The Last of Us Part II, released for PlayStation 4 on 19 June 2020, is the canonical AAA accessibility-feature reference. The audit count of approximately 60 settings — published by AbleGamers and corroborated in the IGDA-GASIG's 2020-21 State of Game Accessibility survey — covers motor (control remapping, single-stick play, autopickup, autoaim toggles), visual (high-contrast modes, full-screen-reader-style menu narration, granular subtitle controls, magnification), auditory (closed captions for ambient and effect audio, navigation assist for blind players including audio cues for traversal), and cognitive (skip puzzle, lock-on aim, simplified UI) categories. The implementation depth — not just the feature count — is what set the title apart. The high-contrast mode is per-character-recolourable; the screen-reader-style narration covers menu state and gameplay HUD; the navigation-assist mode produces an audio-only playable mode demonstrated post-launch by blind players who completed the campaign.
Replicating that feature set is expensive. The Naughty Dog accessibility team grew to nearly twenty people across the development cycle and the Part II Remastered 2024 release. Subsequent Sony first-party titles — Horizon Forbidden West, God of War Ragnarök, Marvel's Spider-Man 2 — matched substantial portions of the feature set but not the full count, and the Naughty Dog audit count remains the AAA ceiling. The only Western AAA title to match the 60-plus count is, as of the 2025 audit, Microsoft's Forza Motorsport (2023), which shipped with an audited 70-plus accessibility settings, including a sign-language-interpreter rendering for cinematics. The Last of Us Part II's legacy is that it moved the AAA accessibility frontier from "best-effort subset of GAGs" to "feature count as a marketing claim."
Naughty Dog accessibility statement — June 2020
"The Last of Us Part II ships with over 60 accessibility options across three categories — visual, audio, and motor — to allow more players to enjoy the game, with custom difficulty granularity, alternate input schemes including single-stick play, high contrast display modes, full menu narration, and audio cues that allow players who are blind or low-vision to complete the entire campaign."
Naughty Dog accessibility announcement, June 2020 (PlayStation Blog)
06 · Platform-holder programmes — Xbox, PlayStation, Nintendo
The platform-holder accessibility programmes have done most of the work to move the AAA industry's floor upward. Microsoft's Xbox Accessibility Guidelines, the Xbox Adaptive Controller (released 2018, refreshed 2024 with the Proteus modular controller), the Inclusive Tech Lab, and the Game Stack accessibility-certification programme combine to put a soft compliance pressure on every Xbox-platform release. Microsoft does not enforce the XAGs as a hard certification gate — a third-party title can ship on Xbox without hitting them — but the XAG-pass marketing positioning, the platform-holder relationship leverage, and the Adaptive Controller user base have produced an environment where most major Xbox releases are at least audited against the XAGs.
Sony's programme is less codified than Microsoft's but materially comparable. The PlayStation Access Controller, released in late 2023, is Sony's equivalent of the Xbox Adaptive Controller — a modular kit designed to interface with assistive switches and joysticks. The PlayStation Store's accessibility-tagging programme, launched the same year, exposes per-title accessibility-feature presence at the point of purchase. Sony does not publish a formal equivalent of the XAGs; its accessibility programme operates inside its first-party studios and through producer-level guidance to third-party publishers.
Nintendo is the late mover. The Switch generation (2017-2024) shipped without platform-level accessibility infrastructure of the kind Microsoft and Sony had built. The Switch 2 transition in 2024-25 has been the occasion for Nintendo to catch up: platform-level options for subtitles and colour-aware UI, expanded controller-remapping at the system level, and an emerging Nintendo Switch Online accessibility-feature-tagging programme that mirrors Sony's. Nintendo's first-party output remains the most variable in the industry — The Legend of Zelda series, in particular, has been a sustained target of accessibility critique — but the trajectory is unambiguously upward.
The Xbox Adaptive Controller as a strategic move
Microsoft's 2018 release of the Xbox Adaptive Controller — a $99 hardware kit designed by the Microsoft Inclusive Tech Lab in partnership with AbleGamers, the Cerebral Palsy Foundation, Craig Hospital, SpecialEffect, and the Warfighter Engaged community — did more to move the accessibility frontier than any single software feature. By shipping a piece of hardware as a first-party platform-holder product, Microsoft made AAA accessibility a procurement question rather than a charity question. The 2024 Proteus refresh extended the strategy.
07 · AbleGamers, IGDA-GASIG, and the consultancy layer
The civil-society layer in game accessibility is small, well-networked, and disproportionately influential. AbleGamers (registered as a 501(c)(3) in West Virginia since 2004) provides direct grant-funded equipment to disabled players, audits AAA releases against the Game Accessibility Guidelines, and runs the Player Panels programme that pairs disabled players with AAA studios for development-cycle consultation. The IGDA Game Accessibility Special Interest Group, hosted by the International Game Developers Association, runs the annual State of Game Accessibility survey, the GDC accessibility-track programming, and the GASIG-curated reading list that is the de-facto entry point for studio accessibility leads. The UK-based SpecialEffect charity runs StarGazing and EyeMine programmes that produce eye-gaze and switch-input infrastructure used by both individual players and AAA studio user-testing labs.
Around those three organisations sits a small consultancy layer — Ian Hamilton, the leading independent consultant, has worked with virtually every AAA publisher in the top-10 leaderboard above. Cherry Thompson (consultant on The Last of Us Part II) and Steve Saylor (audit reviewer on multiple Sony first-party titles) operate at the same individual-consultant level. The consultancy layer is what closes the gap between platform-holder guidelines (XAGs, the implicit Sony-equivalent) and per-title implementation. Without the consultancy layer, the AAA accessibility-feature counts in the leaderboard above would be measurably lower — probably by ten to fifteen recommendations per Intermediate-tier release.
08 · 2026-28 outlook
Three threads will define the rest of the decade.
The FCC's expanded Section 716 rule. A final rule on the January 2024 FNPRM is expected within the 2026 calendar year. The most-discussed expansion vector — clarifying that integrated party chat, in-game voice-with-overlay, and cinematic captions fall within ACS — would impose a meaningful new compliance pipeline on publishers that have not yet built one. Microsoft and Sony first-party are positioned to absorb the expansion; Take-Two, Capcom, and Square-Enix would have to invest. The enforcement track in front of the Consumer and Governmental Affairs Bureau is not aggressive by Title III standards, but it produces meaningful compliance pressure through the public-record posture of the complaint docket.
The European Accessibility Act's video-game reach. The EAA's 28 June 2025 compliance deadline applies to a defined list of products and services that, on a contested reading, includes e-books and some categories of in-game communications but not video games as such. The Member-State transpositions vary on this point, and the European Commission's 2026 review will probably issue an interpretive guidance document on the games-industry scope. Where the EAA does bite, it bites on the customer-service surfaces around games — storefronts, account systems, support chat — rather than on gameplay itself.
The first AAA accessibility consent decree. No publisher in the top-10 leaderboard has so far been subject to an FCC consent decree on Section 716 grounds, nor to a US Department of Justice ADA Title III consent decree on game accessibility specifically. The 2025-26 plaintiffs' bar attention to live-service titles, in-game stores, and platform-side communications suggests that the first such consent decree is more probable than not within the 2026-28 window. Sony's PlayStation Store and Microsoft's Xbox storefront have absorbed scattered accessibility complaints; whether one matures into a settled enforcement action is the open question.
The through line
Game accessibility in 2026 is a measurable, ranked, document-anchored field in a way it simply was not in 2014 when the CVAA video-game waiver lapsed. The Game Accessibility Guidelines have produced a tiered standard; AbleGamers and the IGDA-GASIG have produced an auditing and convening infrastructure; Microsoft and Sony have produced platform-holder programmes that exert real pressure on third-party publishers; and Naughty Dog's The Last of Us Part II set a feature-count ceiling that the industry now measures itself against. The seven-of-ten Basic-tier coverage on the leaderboard above is the visible artefact of fifteen years of working-group standards-making finally landing in shipping product.
What it is not is a finished project. The Japanese-internal-release gap, the live-service / in-game-store surfaces, and the question of integrated party chat's regulatory home are all live. The January 2024 FCC Further NPRM and the EAA's 2026 interpretive guidance will set the next regulatory shape. Whether the AAA leaderboard's middle and bottom thirds move closer to the top — Sony first-party and Microsoft — or whether the gap widens is the consequential 2026-28 question, and the answer will probably be decided by the cost of the first consent decree against a publisher that did not.
---
title: Generative AI and screen-reader prompts: a design discipline emerges
url: https://www.disabilityworld.org/articles/generative-ai-and-screen-reader-prompts/
description: Writing system prompts that make ChatGPT, Claude, Gemini and Be My AI useful to screen-reader users is becoming its own design craft — with rules about structure, em-dashes, AT handoff, and the open UX problems nobody has solved yet.
author: Disability World
pubDate: 2026-05-22
tags: generative-ai, screen-readers, prompts, chatbots, design-discipline, tech-news
---
# Generative AI and screen-reader prompts: a design discipline emerges
Image description: A smartphone resting on a wooden desk showing an AI chat interface with headphones plugged in — the visual marker for screen-reader-friendly AI prompt design.
Reading Time: 9 minutes
A new design discipline has crystallised inside the accessibility community over the past eighteen months, and it does not yet have a settled name. Some teams call it "AT-aware prompt engineering"; others call it "screen-reader-shaped system prompts"; the practitioners who came up through voice-UI design tend to call it "the speech-output layer of an LLM." Whatever the label, the craft is the same: writing system prompts and output-shaping rules that make generative AI assistants — ChatGPT, Claude, Gemini, Copilot, Be My AI — useful for the roughly approx. 253 million people worldwide who reach those products through a screen reader.
The problem is concrete and the failure mode is loud. An LLM trained on the public web produces, by default, prose decorated with em-dashes, nested markdown lists, code fences, headings that exist only because the model felt the answer was "structured", and decorative emoji. Read aloud by NVDA, JAWS, VoiceOver or TalkBack, that output becomes a stream of "dash dash" interjections, "bullet bullet bullet" enumeration without any sense of where one item ends, "heading level two" announcements that interrupt a sentence, and emoji name-strings ("smiling face with sunglasses") between every other clause. The information is in there. The user cannot extract it without rewinding three times. This piece is a primer on what the discipline is asking of model builders, what the products have shipped so far, and the open UX problems nobody has solved yet.
The new discipline — what it actually consists of
Screen-reader-aware prompt design is not a single rule. It is a small set of constraints that, together, produce output a synthesiser can pronounce intelligibly and a screen-reader navigation key can move through. The constraints fall into four buckets.
Concise responses with semantic structure. Default LLM output is too long for spoken delivery — a 600-word answer that reads fine in a sighted user's browser becomes a four-minute monologue that the screen-reader user has no way to skim. The discipline asks for shorter answers, but more importantly for structured shorter answers: an opening one-sentence summary that the user can stop at, followed by structure the screen reader can navigate by heading or by list-item.
Avoid em-dashes and other punctuation that synthesisers mispronounce. The em-dash, the en-dash, the parenthetical, the slash-as-conjunction, the ASCII-art separator — all of these are read aloud as either silence, a literal "dash", or a confusing pause that breaks a clause in half. The convention emerging across the major models is: prefer the comma and the full stop; use the colon for the one place it really earns its keep; never use em-dashes in spoken-context responses; never use ASCII rules to separate sections.
Declare what is a list, what is a heading, what is code. Synthesised speech has no visual hierarchy. A heading needs to be announced as "heading", a list needs to be announced as "list with N items, item one", code needs to be announced as "code", and the model needs to either output structures the screen reader recognises (HTML, proper markdown the rendering surface converts to ARIA) or to verbally narrate the structure itself ("Here are three options. Option one: ...").
No markdown soup. Markdown is fine when the rendering surface converts it to semantic HTML. Markdown is hostile when the surface displays the raw asterisks and underscores, because the screen reader then announces "asterisk asterisk" before every bold word. The discipline is to detect the rendering context — chat UI with markdown rendering versus terminal versus screen-reader-driven voice interface — and to shape the output accordingly. The same model needs to produce different surface representations of the same answer.
What screen readers actually need from AI
To make the constraints above concrete, it helps to look at the actual behaviour of the four screen-reader / OS combinations that dominate the field: JAWS on Windows, NVDA on Windows, VoiceOver on macOS and iOS, and TalkBack on Android. They are not interchangeable, and a prompt that produces great output for one can be unreadable on another.
Navigation by heading. All four readers expose a heading-navigation key (H in JAWS and NVDA, Rotor in VoiceOver, the reading-control toggle in TalkBack). For a long AI answer to be navigable, the model has to emit real semantic headings — either through a markdown rendering pipeline that converts to <h2>/<h3> with proper level nesting, or via the chat surface's own structured-response API. A model that "structures" its answer by bolding the first three words of each paragraph has produced something that looks structured visually and is completely flat to a screen reader.
Navigation by list. Lists are useful in spoken output precisely because the screen reader announces the count ("list with seven items") and lets the user step through with the list-item navigation key (I in NVDA, L in JAWS). But this only works if the list is a real <ul> or <ol>. A "list" produced by emitting bullet characters at the start of each line, with no list wrapper, is read as ordinary prose with an unexplained "black circle" or "bullet" interjection on every line.
Skip-by-section. Long-form AI answers — explanations, comparisons, code-and-commentary, multi-step instructions — need a way for the screen-reader user to skip to the section they care about without listening through the preamble. This is the single hardest piece to design well, because the model has to produce a navigable structure and the chat surface has to render it in a way the OS exposes to the assistive technology, and the screen reader has to be configured to use the heading-navigation key in that surface. All three things fail in the wild; usually it is the middle one.
Pronunciation hints. Synthetic voices stumble on technical terms, acronyms with mixed letters, URLs, code identifiers, mathematical notation, and non-English names. A well-designed model will, for screen-reader-context responses, spell out acronyms on first use ("WCAG, the Web Content Accessibility Guidelines"), expand initialisms the synthesiser cannot pronounce, and avoid embedding raw URLs inside flowing prose where the synth will read the slashes aloud. None of the major products do this consistently in 2026.
How the products are handling it
As of mid-2026, the major generative AI products have taken visibly different positions on screen-reader-aware output. None of them have nailed it. The progression is faster than it was twelve months ago, but the gap between the best and the worst is still wide.
ChatGPT (OpenAI). The web client now ships with a "concise mode" toggle that shortens default responses and reduces markdown decoration. The voice mode introduced in 2024 — and substantially upgraded in 2025 — is the closest any major product has come to a screen-reader-native interface, because it bypasses the visual chat entirely and delivers a spoken answer with a stop, replay and "say that again" gesture. The custom-instructions field allows screen-reader users to declare their preferences once and have them apply across sessions, which is the user-driven workaround the community has settled on. The remaining gaps: GPT models still default to em-dash-heavy prose unless instructed otherwise, and the heading-level emitted in markdown does not always map cleanly to ARIA in the chat surface.
Claude (Anthropic). Claude's system-prompt discipline has moved closest to the conventions described above. The model is noticeably less em-dash-prone than the GPT line in 2026, defaults to shorter answers, and responds well to system-prompt instructions like "you are speaking to a screen-reader user; use no em-dashes, prefer short paragraphs, and use real headings or numbered lists when structure is needed." The Claude.ai chat surface renders markdown to semantic HTML with proper heading levels, which makes the heading-navigation key work. Voice output through third-party integrations exists but is less developed than ChatGPT's first-party voice mode.
Gemini (Google). Tight integration with TalkBack on Android is Gemini's structural advantage; the model can hand off to the OS-level screen reader through Android's accessibility services in a way the iOS and web competitors cannot. The "Hey Google, ask Gemini..." flow on accessible Android devices is, for some users, the most natural AI-plus-screen-reader experience available. The remaining gaps: the web interface still over-decorates responses, the heading hierarchy in Gemini's web answers is inconsistent, and the model is more prone to producing decorative emoji than its competitors.
Be My AI (Be My Eyes + OpenAI). This is the most narrowly scoped of the four — a visual-description assistant that uses GPT-4-class vision models to describe images and surroundings for blind and low-vision users. It is also the only product in this list designed from day one for a screen-reader user as the primary audience. Be My AI's prompt design is the field's clearest demonstration of what AT-aware output looks like in practice: descriptions open with a one-sentence summary the user can stop at, follow with structured detail only if asked, and avoid spatial language ("on the left", "above") that requires sighted context to interpret. The product remains, in 2026, the closest the field has to a reference implementation.
The cross-cutting observation is that the four products have made progress on the easy parts — shorter answers, fewer em-dashes, a custom-instructions field — and have barely begun on the hard parts. The hard parts are below.
Open UX problems nobody has solved
The screen-reader-aware prompt-design literature converges on four open UX problems where the right answer is not yet known. None of them are model-capability problems; all of them are interaction-design problems that sit between the LLM, the chat surface, the OS, and the screen reader.
Interrupt-ability. A sighted user can scan an LLM response in approx. two seconds and decide whether to read it. A screen-reader user cannot. If the answer is wrong or off-target, the user has to listen through enough of it to know that, then interrupt. Voice modes have a stop button. Text modes generally do not — the response streams in and the screen reader announces it as new content as it arrives, and the user has no clean way to say "stop generating, this is not what I asked." The Be My AI app handles this best; the web chat clients handle it worst.
Repeat-last-answer with selectable granularity. Asking a screen reader to re-read the last response is easy if the answer is short. It is unusable if the answer is six paragraphs and the user only wants to hear the third paragraph again. The interaction the community is asking for is "repeat the last list item", "repeat the last heading section", "repeat the last code block." That requires the chat surface to expose the structure to the screen reader in a way the screen reader's own re-read commands can address. In 2026, none of the major products do this; the user has to use the screen reader's own line-by-line navigation, which is laborious.
Navigate-by-section in spoken output. Voice modes do not have a heading-navigation key. The user listens to a four-minute answer linearly, with no way to skip from the "overview" section to the "specifics" section without rewinding by time. The interaction designs being prototyped — a spoken "section list" the user can navigate with arrow keys, a "go to section three" voice command, a "give me the headings only" mode — are early. The Be My AI app's "more detail on the colours" follow-up is the closest functioning version of this in a shipping product.
The AT-handoff question — when does the AI speak versus read content aloud? This is the deepest design question. If a screen-reader user opens an AI assistant on a webpage, who is speaking — the AI's own voice (TTS layer), or the user's installed screen reader reading the AI's text output? The two voices have different settings, different speaking rates, different pronunciation hints, different stop-and-replay gestures. Two systems trying to speak the same content at the same time produces nothing usable. The convention emerging is: voice-mode interactions use the AI's own TTS and explicitly suppress the screen reader; text-mode interactions emit semantic HTML and let the screen reader do the speaking. But the boundary between the two modes is not always clean — image-description, code-generation, mathematical notation, and multi-modal answers all sit awkwardly between voice and text — and that boundary is where most of the live UX problems live.
Where it goes next
The discipline is roughly where web accessibility was in approx. 2002 — past the "is this a real problem?" phase, past the "is anyone responsible?" phase, into the "what are the actual rules?" phase. Three things are likely to happen across 2026 and 2027.
First, the model builders will codify their internal screen-reader prompts and publish them, the way Anthropic publishes Claude's system prompts in VPAT-style accessibility statements and OpenAI has begun documenting GPT's behavioural defaults. The community is asking for the equivalent of a model card — a "screen-reader output card" — that names the conventions a given model has been trained or system-prompted to follow.
Second, the chat surfaces — web clients, mobile apps, IDE integrations — will gain proper semantic-HTML rendering pipelines and proper ARIA exposure for chat history, with the navigation keys mapped to the OS-level screen reader. This is unglamorous work, and it is the work that will move the needle most for daily users.
Third, the screen-reader vendors themselves — Vispero (JAWS), NV Access (NVDA), Apple (VoiceOver), Google (TalkBack) — will start shipping AI-aware features: native heading-navigation inside AI chat surfaces, a standardised "stop generating" gesture, smarter re-read commands that know about LLM response structure. NVDA's open-source add-on ecosystem is already producing early versions of these. The proprietary readers are slower but the direction is the same.
The deeper observation is that screen-reader-aware prompt design has stopped being a niche concern of a handful of blind developers and has become a baseline expectation of every AI product team that wants to ship into regulated markets. The European Accessibility Act applies to "interactive self-service terminals" and "consumer terminal equipment with interactive computing capability" — a category that almost certainly captures a major AI assistant on a phone. The AT-aware output layer is not a feature any more; it is procurement-binding. The teams that figure out the rules now will ship the products that survive 28 June 2025 and onwards. The teams that treat it as an afterthought will be the next round of EAA enforcement cases.
Final thoughts
The craft is small, the stakes are large, and the rules are still being written. If you build with LLMs and you have not yet had a conversation with a screen-reader user about what your product actually sounds like when they use it, that is the next thing on the list. Most of what is wrong with AI for screen-reader users in 2026 is not a model-capability problem; it is a prompt-and-surface design problem that any product team can fix in a sprint, if they decide to.
The community has been generous with its time, its testing, and its patience. It is also losing patience faster than it used to, because the products are now mainstream and the excuse of "we are still figuring it out" has run out. The discipline is here. The conventions are converging. The next eighteen months will sort the teams that listened from the teams that did not.
---
title: Germany's BGG, BITV, and BFSG: how federal law meets EU technical standards
url: https://www.disabilityworld.org/articles/germany-bfg-and-bitv/
description: Germany's accessibility regime layers three statutes — the federal BGG (2002, reformed 2016), the technical BITV 2.0 ordinance, and the EAA-transposing BFSG (2021) — over a parallel set of sixteen Länder laws and a 2025 BAFA enforcement turn-on.
author: Disability World
pubDate: 2026-05-22
tags: germany, bgg, bitv, bfsg, regulations, regulation-primer, eu
---
# Germany's BGG, BITV, and BFSG: how federal law meets EU technical standards
Image description: The Reichstag's glass Norman Foster dome with the German federal flag flying above — institutional anchor for Germany's BGG, BITV and BFSG accessibility framework.
Reading Time: 13 minutes
Germany's domestic accessibility regime is not one statute but a layered stack of three: the Behindertengleichstellungsgesetz (BGG) — the Federal Act on Equal Opportunities for Persons with Disabilities, in force since 1 May 2002 and substantially reformed in 2016; the Barrierefreie-Informationstechnik-Verordnung (BITV 2.0) — the federal Accessible Information Technology Ordinance, in force in its current "2.0" form since 2011 and re-aligned to EN 301 549 / WCAG 2.1 AA through successive amendments; and the Barrierefreiheitsstärkungsgesetz (BFSG) — the Accessibility Strengthening Act of 16 July 2021, which transposes the EU's European Accessibility Act (Directive 2019/882) into German federal law and switched on for the private sector on 28 June 2025. The layers do not collapse into each other: BGG covers federal public bodies, BITV gives the technical detail, and BFSG covers private-sector consumer products and services. For situating these alongside other Member State transpositions, see the national disability-rights regulations index and the explainer on EN 301 549, the harmonised standard all three German instruments lean on.
This primer walks the stack in order — BGG's general non-discrimination duty, BITV 2.0's technical implementation, BFSG's EAA transposition, the new BAFA federal enforcement role in force since June 2025, and the parallel Länder-level laws (Landesgleichstellungsgesetze) that cover state and municipal authorities across the sixteen federal states. Read the layers separately, because the obligations, the addressees, and the enforcement routes are different at each layer.
The BGG — the federal anchor (2002, reformed 2016)
The Behindertengleichstellungsgesetz (Federal Act on Equal Opportunities for Persons with Disabilities) is the foundational German federal statute on disability equality. It entered into force on 1 May 2002 as part of a wave of post-CRPD-anticipating reform that also delivered Book IX of the Sozialgesetzbuch (SGB IX) on rehabilitation and participation. The 2002 BGG installed three load-bearing concepts into German federal law: a statutory definition of Barrierefreiheit (accessibility) in §4, a binding duty on federal public authorities to act accessibly in §§7–11, and a Verbandsklagerecht (associational standing) in §13 that lets recognised disability organisations bring suit in their own name.
The reform of 2016 — drafted to align the BGG with the UN Convention on the Rights of Persons with Disabilities, which Germany ratified in 2009 — added or strengthened several mechanisms: a duty to provide reasonable accommodation in administrative procedures (§7(2)), explicit recognition of deaf-blindness as a disability category (§3), a federal monitoring body (the Bundesfachstelle für Barrierefreiheit) established at the German Pension Insurance Knappschaft-Bahn-See in §13a, and an arbitration board (Schlichtungsstelle) at the Federal Government Commissioner for Matters relating to Persons with Disabilities under §16. The 2016 reform also pulled in the Leichte Sprache (Easy-to-Read German) duty under §11 and required federal authorities to communicate accessibly in Deutsche Gebärdensprache (DGS, German Sign Language) under §9.
Who BGG actually binds
The BGG's addressee scope is narrower than a first read suggests. It binds federal public authorities (Bundesbehörden), corporations and institutions under federal public law, and — critically — federally-funded entities receiving more than 50 percent of their funding from federal sources. It does not bind Länder authorities, municipal authorities, or private actors. Land authorities are covered by sixteen parallel state statutes; private actors fall under the General Equal Treatment Act (AGG, 2006) for non-discrimination and, since 2025, under BFSG for accessibility of consumer products and services. The BGG is the federal-public-sector floor; the rest of the building sits on top of or beside it.
The Verbandsklage right
One BGG feature that distinguishes the German regime from many EU peers is §13's associational standing right. Recognised disability organisations — currently around 80 organisations on the federal list maintained by the Federal Ministry of Labour and Social Affairs (BMAS) — can sue federal authorities for breach of the BGG without identifying an individually-affected complainant. The right has been used sparingly but visibly: the Deutscher Behindertenrat and member organisations have brought a handful of cases each cycle, mostly aimed at accessible procurement and digital-service deficits. Associational standing reduces the friction the individual-complaint route otherwise imposes — the cost, the disclosure exposure, and the slow administrative-court calendar.
BITV 2.0 — the technical ordinance
The Barrierefreie-Informationstechnik-Verordnung (Accessible Information Technology Ordinance) is the implementing ordinance issued under §12 of the BGG. The first BITV took effect in 2002 alongside the parent statute; the current BITV 2.0 entered into force on 22 September 2011 and has been amended in 2019 (to align with the EU Web Accessibility Directive 2016/2102 and EN 301 549 v3.1.1) and again in 2023 (to track EN 301 549 v3.2.1 and clarify mobile-application coverage). BITV is the layer that names a specific technical standard and binds federal public-sector websites, intranets, and mobile applications to comply with it.
In its current form BITV 2.0 incorporates EN 301 549 V3.2.1 by reference — the harmonised European standard for ICT accessibility — which in turn embeds WCAG 2.1 Level AA for web content and mobile applications. The ordinance also imposes German-specific duties beyond EN 301 549: an explicit obligation in §4 to provide information in Easy-to-Read German (Leichte Sprache) and in German Sign Language (DGS) on home pages of federal authorities, a feedback-mechanism duty in §12b, and accessibility-statement publication requirements in §12c that mirror Article 7 of the Web Accessibility Directive.
Who BITV binds, and what it covers
BITV's scope tracks BGG's: federal authorities, federally-funded corporations under public law, and — via Member-State transposition of Directive 2016/2102 — federal courts and federal legislative bodies in their administrative capacity. Federal websites must publish an accessibility statement and a feedback mechanism, and the Überwachungsstelle des Bundes für Barrierefreiheit von Informationstechnik (BFIT-Bund, the federal monitoring body for IT accessibility, hosted at BMAS) carries out periodic monitoring and reports to the European Commission on a three-year cycle. The most recent federal monitoring report, covering 2022–24, sampled roughly 200 websites and 60 mobile applications and found persistent gaps on keyboard navigation, alternative text, video captioning of older content, and the publication of accessibility statements in conforming form.
Land-level public-sector ICT is covered by sixteen parallel Land-level BITV-equivalents — sometimes called BITV-Landes-X — issued under each Land's own equal-opportunities statute. These mirror BITV 2.0 closely in technical content but vary on monitoring cadence and on whether the Land or the federal monitoring body is competent.
BFSG — the EAA transposition, switched on June 2025
The Barrierefreiheitsstärkungsgesetz (BFSG) — the Accessibility Strengthening Act — was adopted on 16 July 2021 and entered into force in stages, with the operative date for in-scope private-sector products and services set at 28 June 2025, the EAA's harmonised application date. The BFSG transposes Directive (EU) 2019/882 — the European Accessibility Act — into German federal law. Where the BGG covers federal public authorities and BITV gives the technical detail for federal-public-sector ICT, the BFSG is the first general federal statute that binds private-sector economic operators on accessibility of consumer-facing products and services.
What the BFSG covers
The product and service scope follows the EAA verbatim, with the addressee structure adapted to German federal law. The covered products include consumer-grade general-purpose computer hardware and operating systems, self-service terminals (ATMs, ticketing machines, check-in kiosks), consumer terminal equipment for electronic communications and audiovisual media services (set-top boxes, smartphones, smart TVs), and e-readers. The covered services include consumer banking services, electronic communications services, services giving access to audiovisual media, passenger transport information services (web/mobile/self-service-terminal components), e-commerce, and e-books.
Economic operators bound: manufacturers, importers and distributors of in-scope products; service providers offering in-scope services to consumers in Germany.
Microenterprise carve-out: consistent with EAA Article 4(5), service-providing microenterprises (fewer than 10 employees and turnover or balance-sheet total ≤ €2 million) are exempt from the service obligations but must keep documentation of their microenterprise status; the product carve-out does not apply.
Disproportionate burden defence: available under §17 BFSG, subject to a documented assessment that follows Annex VI of the EAA.
Fundamental-alteration defence: available where compliance would require a change to the basic nature of the product or service.
Transitional period: contracts concluded before 28 June 2025 and self-service terminals already in use on that date benefit from a transition window of up to five years for the contract and up to fifteen years for the terminal hardware (BFSG §38).
The technical accessibility requirements that products and services must meet are set out in the implementing ordinance, the BFSGV (Barrierefreiheitsstärkungsgesetz-Verordnung) of 15 June 2022, which references EN 301 549 for ICT components and the relevant harmonised standards for other product classes. Where a harmonised standard does not yet exist, conformity may be demonstrated through the EAA functional accessibility requirements directly, mirroring Annex I.
BAFA and the market-surveillance turn-on
The most consequential shift in 2025 was not the statute taking legal effect — that was 2021. It was the operative market-surveillance turn-on on 28 June 2025. Under §19 BFSG, market-surveillance authority over BFSG-covered products sits with the Bundesamt für Wirtschaft und Ausfuhrkontrolle (BAFA) — the Federal Office for Economic Affairs and Export Control, headquartered in Eschborn. For BFSG-covered services, competence is distributed: the Bundesnetzagentur (BNetzA) for electronic communications services, the BaFin for consumer banking, the Landesmedienanstalten for audiovisual-media access services, and — for e-commerce, e-books, and passenger-transport information services — the Länder market-surveillance authorities coordinating through a federal-Länder forum convened by BMAS.
BAFA's mandate runs from market-entry conformity checks on products through to administrative-fine proceedings under §37 BFSG. Fines can reach up to €100,000 per violation, with the upper limit applied per affected product line or service in repeat-offender cases. Beyond fines, BAFA can order products withdrawn from the market or recalled where non-compliance creates a "significant risk" — a phrase borrowed from the General Product Safety Regulation (Regulation (EU) 2023/988) market-surveillance architecture. BAFA also publishes annual market-surveillance reports identifying the product categories with the highest non-compliance rates; the first BFSG-cycle report is due in late 2026.
No private right of action for damages
The BFSG does not, by itself, create a private right of action for monetary damages. Consumers can file complaints with BAFA, BNetzA or the relevant service authority, who decide whether to open proceedings. Recognised consumer-protection associations and disability organisations have collective-action rights under §22 BFSG that mirror the EU Representative Actions Directive (Directive (EU) 2020/1828, transposed into German law as the VDuG, in force since October 2023). The collective-action route is the part of the German private-sector enforcement architecture most likely to produce visible case law in 2026–27.
The Länder layer — sixteen parallel statutes
Germany's federal structure means that public-sector accessibility for the Länder and their municipalities is governed by sixteen separate Land-level Landesgleichstellungsgesetze (LGGs), each with its own scope, its own monitoring body, and its own technical ordinance. Bavaria's BayBGG (2003, revised 2018), North Rhine-Westphalia's BGG NRW (2004, revised 2016), Berlin's LGBG (1999, revised 2021), Hamburg's HmbBGG (2005, revised 2020), and Saxony's SächsInklusG (2018) are the most-cited examples; the Schleswig-Holstein, Lower Saxony, Hesse and Baden-Württemberg statutes round out the larger Länder.
Most LGGs follow the BGG's structural template — defined accessibility duty, Verbandsklage right, language and communication provisions for DGS and Easy-to-Read — but the technical implementation differs. Some Länder issue their own BITV-equivalent ordinance; some incorporate BITV 2.0 by reference; one or two leave the technical layer underspecified, with the result that municipal-website compliance is uneven across the Land. The 2024 federal monitoring report flagged this asymmetry: federal-public-sector ICT compliance is meaningfully higher than Länder-municipal compliance, primarily because BFIT-Bund's monitoring cadence at federal level is not matched at every Land level.
How the three layers actually interact
In practice a German entity classifying its accessibility obligation in 2026 should walk three questions in order. First, is it a federal public authority, a federal corporation under public law, or a federally-funded entity above the 50-percent threshold? If yes, BGG applies, BITV 2.0 sets the technical standard, and BFIT-Bund monitors. Second, is it a Land or municipal authority, or a Land-funded entity? If yes, the relevant LGG and its technical ordinance apply, monitored by the Land's own equivalent body. Third, is it a private-sector economic operator placing in-scope products or providing in-scope services to consumers in Germany? If yes, BFSG and BFSGV apply, with BAFA, BNetzA, BaFin or a Land market-surveillance authority competent depending on the product or service.
A private hospital that receives 60 percent federal funding could be in two layers at once: BGG (because of the federal funding threshold) and, for its consumer-facing online services, BFSG. The compliance obligations overlap rather than collide — both layers ultimately point at EN 301 549 / WCAG 2.1 AA for digital interfaces — but the enforcement routes differ. A BGG violation goes to the BMAS arbitration board or the administrative courts via Verbandsklage; a BFSG violation goes to BAFA or BNetzA.
The EN 301 549 / WCAG anchor across all three
The unifying technical anchor across BGG, BITV and BFSG is the EU's harmonised standard EN 301 549 V3.2.1, which itself embeds WCAG 2.1 Level AA for web content and mobile applications. This is what gives the German layered architecture coherence in practice: a single technical standard runs through the federal-public-sector layer (via BITV 2.0), the Länder public-sector layer (via Land BITV-equivalents), and the private-sector layer (via BFSG/BFSGV). When a future German implementing ordinance lifts the technical standard to WCAG 2.2 AA, all three layers will move together — there is no plausible scenario in which BFSG moves to 2.2 while BITV stays at 2.1, because BITV references the same harmonised standard BFSGV references.
Practical implications for businesses operating in Germany
Three concrete implications follow for organisations whose German operations crossed the 28 June 2025 BFSG threshold. First, the disproportionate-burden defence in §17 BFSG is documented, not declared — operators relying on it must produce an Annex-VI-style assessment that BAFA can request. Second, the microenterprise carve-out applies to services only; manufacturers and importers of in-scope products cannot fall back on it regardless of headcount. Third, the BFSG accessibility statement is mandatory for in-scope services and must be published in a structured form on the operator's primary consumer-facing surface; the first BAFA spot-checks in the second half of 2025 reportedly focused on whether statements existed at all rather than on substantive conformity, with substantive conformity expected to dominate the 2026 cycle.
For federal public bodies and federally-funded entities, the practical priority in 2026 is closing the gaps the BFIT-Bund 2022–24 monitoring report flagged: keyboard navigation on transactional flows, alternative text on imagery, captioning of older video content, and accessibility-statement form. For Land and municipal entities, the priority is whichever LGG-monitoring cycle is current.
Conclusion: a layered regime, anchored to one standard
Germany's three-statute layered architecture looks complex from outside because it is — a federal BGG and an implementing BITV from 2002, sixteen Land statutes covering the public sector, and a BFSG transposing the EAA into the private sector since 2021. What makes the regime workable in practice is that all three layers anchor to the same harmonised European technical standard, EN 301 549, which embeds WCAG 2.1 AA for digital interfaces and runs through every layer of the stack. The 2025 BAFA market-surveillance turn-on under BFSG is the change that puts a federal regulator on the private-sector side for the first time, with administrative fines up to €100,000 per violation and collective-action exposure through VDuG. The doctrine and the addressees were already in place; the enforcement mechanism is what 2026 onwards will test.
Bundesministerium der Justiz. Behindertengleichstellungsgesetz (BGG), in force since 1 May 2002, as amended by the Federal Participation Act of 2016. gesetze-im-internet.de/bgg
Bundesministerium der Justiz. Barrierefreie-Informationstechnik-Verordnung (BITV 2.0), in force since 22 September 2011, latest amendment 2023. gesetze-im-internet.de/bitv_2_0
Bundesministerium der Justiz. Barrierefreiheitsstärkungsgesetz (BFSG), of 16 July 2021, BGBl. I p. 2970, operative for in-scope products and services from 28 June 2025. gesetze-im-internet.de/bfsg
Bundesministerium der Justiz. Barrierefreiheitsstärkungsgesetz-Verordnung (BFSGV), of 15 June 2022. gesetze-im-internet.de/bfsgv
Überwachungsstelle des Bundes für Barrierefreiheit von Informationstechnik (BFIT-Bund). Federal monitoring report on the accessibility of public-sector websites and mobile applications, 2022–2024. bfit-bund.de
Bundesamt für Wirtschaft und Ausfuhrkontrolle (BAFA). BFSG market surveillance information for economic operators (2025). bafa.de
European Union. Directive (EU) 2019/882 on the accessibility requirements for products and services (European Accessibility Act), OJ L 151, 7.6.2019.
European Union. Directive (EU) 2016/2102 on the accessibility of the websites and mobile applications of public sector bodies, OJ L 327, 2.12.2016.
ETSI / CEN / CENELEC. EN 301 549 V3.2.1 — Accessibility requirements for ICT products and services (March 2021).
---
title: The state of global disability metrics in 2026: what gets counted, what doesn't, and why the gap keeps closing slowly
url: https://www.disabilityworld.org/articles/global-disability-metrics-2026/
description: Two decades after the Washington Group Short Set on Functioning was published, only a minority of national censuses and household surveys actually use it. WHO's 2024 World Report on Disability update and the 2025 UN DESA Disability Statistics Compendium tell the 2026 numbers — and the gaps.
author: Disability World
pubDate: 2026-05-22
tags: data, statistics, washington-group, who, un-desa, measurement
---
# The state of global disability metrics in 2026: what gets counted, what doesn't, and why the gap keeps closing slowly
Data dossier · Measurement infrastructure
The state of global disability metrics in 2026 — what gets counted, what doesn't, and why the gap keeps closing slowly
Two decades after the Washington Group on Disability Statistics published its Short Set on Functioning, the world finally has a defensible headline number: 1.3 billion people, roughly 16% of the global population, living with significant disability, per the WHO's 2024 monitoring update to the World Report on Disability. That figure rests on a measurement infrastructure that took twenty years to build — and still misses four populations almost entirely. Fifty-five-plus national statistical offices now run the Washington Group Short Set; more than 90 LMICs publish UNICEF/WG Child Functioning Module data; the 2025 UN DESA Compendium maps around 120 countries producing at least some disability-disaggregated SDG indicator reporting. The headline is firmer than it has ever been. What still slips through it is what the rest of this dossier is about.
Findings · Case file 0106 findings · derived from WHO 2024, WG Secretariat inventory, UN DESA Compendium 2025
What the 2026 measurement infrastructure reveals
011.3B
The WHO 2024 update puts global disability at 1.3 billion people — the firmest headline the field has ever had
Functional limitation, not condition-by-condition Global Burden of Disease accounting, is now the primary frame. Of the 1.3 billion, around 190 million adults experience very significant difficulty in functioning — a sub-figure that has barely moved since the 2011 report.
0216%
Global prevalence climbed from about 15% in 2011 to roughly 16% in 2024
The world has not become "more disabled" — measurement has improved, the over-60 population has grown by nearly 240 million in 13 years, and chronic-disease prevalence has expanded faster than population growth in South Asia and sub-Saharan Africa.
0355+
More than 55 national statistical offices now deploy the Washington Group Short Set in census or flagship survey
Out of roughly 200 jurisdictions running decennial or quinquennial censuses. Most of Latin America, an expanding number of African NSOs, several large Asian countries, and a slowly growing share of OECD high-income members.
04approx. 10%
Around 1 in 10 children aged 5–17 in LMICs has a disability under the Child Functioning Module
UNICEF's 2024 Seen, Counted, Included update aggregates CFM-based prevalence estimates from more than 90 LMICs — the broadest comparable child-disability dataset that has ever existed. Substantially higher than legacy estimates.
054 gaps
Four populations remain systematically under-counted, even in the best-instrumented surveys
Children under 2 (no standardised instrument exists). People in residential institutions, prisons, refugee camps (excluded from household sampling frames). Persons with intellectual or psychosocial disabilities (Extended Set runs in fewer than two dozen countries). The 117 million-plus forcibly displaced (almost no NSO sampling frame reaches them).
062.2 → 17.8
National prevalence ranges from 2.2% (India, legacy single-question) to 17.8% (UK, GSS harmonised + WG)
Bangladesh 2.8%, South Africa 6.0%, Brazil 8.9%, US 13.4%, Mexico 16.5%. Most of the spread is instrument, threshold, and aggregation choice — not underlying epidemiology. The 2024 WG guidance now recommends reporting at least two thresholds in parallel.
SourceWHO World Report on Disability (2011) and 2024 monitoring update; Washington Group Secretariat 2024 country-deployment inventory; UN DESA Disability Statistics Compendium 2025; UNICEF/WG Module on Child Functioning, Seen, Counted, Included 2024 update; national census releases (IBGE 2022, Stats SA 2022, INEGI 2020, ONS 2021/22, BBS 2022).
Three things changed between the 2011 World Report on Disability and the WHO's 2024 monitoring update, and they all push the global prevalence figure in the same direction. The world is older — the over-60 population has grown by close to 240 million in 13 years, and significant functional limitation rises steeply after 60. Chronic-disease prevalence, particularly type-2 diabetes and its sequelae, has expanded faster than population growth across South Asia and sub-Saharan Africa. And measurement itself has improved: the Washington Group instruments, the UNICEF/WG Module on Child Functioning, and the Model Disability Survey have matured into reference tools that produce higher and more reliable counts than the legacy "do you have a disability?" yes/no question they replaced.
approx. 1.3B
People with significant disability worldwide (WHO 2024)
16%
Global prevalence in 2024 — up from approx. 15% in 2011
190M
Adults with very significant difficulty in functioning
Why "16%" is a floor, not a ceiling
The WHO's 16% prevalence figure is best read as the floor of a defensible estimate, not as a discovery that the world has somehow become more disabled in 13 years. The 2024 update treats functional limitation as the primary frame — the 2011 report leaned on the Global Burden of Disease (GBD) framework, years lived with disability by specific cause, but the 2024 update keeps GBD as an appendix rather than the headline. A functional-limitation count and a condition-by-condition count are not the same population: conflating them double-counts people who appear in several condition rows and undercounts people whose limitations have no neat ICD-11 home.
02 · The headline number, and what it actually measures
The WHO's 16% prevalence figure (roughly 1.3 billion people) is therefore best read as the floor of a defensible estimate, not as a discovery that the world has somehow become more disabled in 13 years. Of that 1.3 billion, the 2024 update splits out around 190 million adults with very significant difficulties in functioning — the population for whom rehabilitative services, assistive technology, and personal-assistance entitlements are most plainly indicated and least adequately delivered. That sub-figure has barely moved between the 2011 and 2024 reports, which is itself a finding.
The number you do not see in the WHO headline is impairment-by-condition. The 2011 report leaned on the Global Burden of Disease (GBD) framework — years lived with disability by specific cause — and the 2024 update keeps GBD as an appendix rather than the headline. A functional-limitation count and a condition-by-condition count are not the same population: conflating them double-counts people who appear in several condition rows and undercounts people whose limitations have no neat ICD-11 home. The 2024 update treats functional limitation as the primary frame.
Reported national prevalence — most recent census or flagship survey
India · 2011 Census
2.2% · legacy single-Q
Bangladesh · 2022
2.8% · WG-SS (narrow)
South Africa · 2022
6.0% · WG-SS
Brazil · 2022
8.9% · WG-SS (adapted)
United States · 2023
13.4% · WG-derived 6-Q
Mexico · 2020
16.5% · WG-SS (broad)
United Kingdom · 2021/22
17.8% · GSS + WG
The range — 2.2% to 17.8% — is what the field's long debate over comparability is actually about. It is not, mostly, that disability is genuinely seven times more common in the UK than in India. The instrument, the threshold at which someone is counted (some difficulty? a lot?), the inclusion or exclusion of people in institutions, and respondents' willingness to self-identify all move the number more than the underlying epidemiology does. Where the WG-SS is deployed consistently — across most of Latin America in particular — cross-country numbers now sit in a far tighter band than a decade ago.
03 · The Washington Group instruments and where they actually run
The Washington Group Short Set's six functional-domain questions are the canonical instrument behind the 16% global prevalence figure: seeing, hearing, walking, cognition, self-care, communication.
The Washington Group Short Set on Functioning (WG-SS) is six questions: difficulties seeing, hearing, walking or climbing steps, remembering or concentrating, self-care, and communicating, each on a four-point severity scale (no difficulty, some, a lot, cannot do at all). It was designed in the mid-2000s to be short enough to bolt onto a national census, neutral enough to translate across cultures, and severity-graded enough to support cross-country comparison if used consistently. The Washington Group Secretariat's 2024 inventory counts more than 55 national statistical offices that have deployed WG-SS verbatim or near-verbatim in their most recent census or flagship household survey.
"More than 55" sounds modest, and it is. There are around 200 jurisdictions that run a decennial or quinquennial census. The WG-SS users include most of Latin America, an expanding number of African NSOs, several large Asian countries, and a slowly growing share of OECD high-income members. Non-users split between countries that still rely on a legacy "are you a person with a disability?" single question (common in parts of South and Southeast Asia and several Gulf states), countries running a national instrument designed before WG-SS existed (the UK's ONS uses the GSS harmonised disability question alongside WG-style questions), and countries whose most recent census predates the Washington Group's last revision.
WG-SS adoption density by region — 2024 Washington Group Secretariat inventory (illustrative)
Latin America
Most NSOs · verbatim or near-verbatim
Sub-Saharan Africa
Expanding adoption
OECD high-income
Slow, often hybridised with national instruments
East & Southeast Asia
Mixed — several large NSOs adopting
South Asia
Mostly legacy single-Q, some pilots
Gulf & West Asia
Legacy frameworks dominant
04 · The child-functioning instrument
For children, the canonical instrument is the Washington Group / UNICEF Module on Child Functioning (CFM), finalised in 2016 and now the reference module for the 2–4 and 5–17 age bands. UNICEF's MICS (Multiple Indicator Cluster Survey) programme has included the CFM by default since MICS6; the 2024 global release carries CFM-based prevalence estimates from more than 90 low- and middle-income countries, the broadest comparable child-disability dataset that has ever existed. UNICEF's 2024 Seen, Counted, Included update reports that around 1 in 10 children aged 5–17 in LMICs has a disability under the CFM definition — substantially higher than legacy estimates, and consistent with the WHO's directional adjustment. The CFM has no analogue for children under 2: the developmental-milestone variation in the 0–24 month band is too wide for a short module to be useful, and that gap is the first of four populations the 2026 measurement infrastructure does not adequately count.
approx. 80% of LMIC deaf children remain out of school
UNESCO's Global Education Monitoring (GEM) Report has, for several editions running, estimated that around 80% of deaf children in low- and middle-income countries are out of school. The CFM finally lets countries count these children — but counting is only the first step. Where the data exists, ministries of education have not yet built the inclusive-education infrastructure to act on it.
05 · The populations the headline numbers still miss
Four groups remain systematically under-counted even in the best-instrumented national surveys.
People in institutions
Most household-survey sampling frames exclude residential institutions, long-stay psychiatric facilities, prisons, and refugee camps — populations in which disability prevalence, particularly psychosocial and intellectual disability, is dramatically elevated. The European Commission's 2023 institutional-population mapping found around 1.4 million Europeans in residential disability institutions alone, none of whom appear in standard household-survey estimates. UN DESA's 2025 Compendium calls the exclusion a "first-order undercount" in its methodology annex.
Persons with intellectual or psychosocial disabilities
The six WG-SS questions are functional-domain questions and good at picking up what they cover. Intellectual disability shows up partially through "remembering or concentrating"; psychosocial disability shows up partially through "communicating" and partially nowhere. The Washington Group's Extended Set and the WG-SS Enhanced module add questions on anxiety, depression, upper-body function, and fatigue, but Enhanced is in use in fewer than two dozen countries as of 2024. The WHO 2024 update flags this as the single largest measurement gap in the current instrument family.
Children under 2
The CFM begins at age 2 and no internationally standardised instrument exists for younger children. UNICEF and the WHO's Early Childhood Development team have piloted a hybrid clinical-screening approach in seven countries since 2023, but the work is methodological rather than ready for national rollout.
People in conflict-affected and displacement settings
UNHCR's 2024 mid-year statistical release counts more than 117 million forcibly displaced people worldwide. On the limited evidence available, disability prevalence in displaced populations runs substantially higher than in settled populations of the same origin — a function of injury, deprivation, and the differential mobility of disabled people in crisis. Almost no NSO's sampling frame covers displaced populations adequately. The Humanitarian Inclusion Standards working group has been pressing UNHCR and IOM since 2022 to embed WG-SS in registration intake, with patchy results.
The instruments exist. The question, as in so much of disability policy, is whether the data will be collected.
06 · The double-counting debate, in four countries
The single most contested operational question in disability statistics in 2026 is also the most technical-sounding: when you ask the six WG-SS questions, and a respondent reports difficulty in more than one domain, do you count them once at the highest severity, once at any severity above the threshold, or do you sum the severity scores? National statistical offices treat this differently — and the difference moves the headline number by several percentage points.
Washington Group Secretariat · methodological note · 2024
"The Short Set was designed so that a country could report a defensible national disability prevalence. It was not designed to allow easy comparison between two countries that have made different threshold and aggregation choices. We are still working on that."
Washington Group on Disability Statistics, 2024 guidance update
Bangladesh's 2022 census reported 2.8% using a "lot of difficulty / cannot do at all" threshold and a highest-domain rule — the conservative end of the WG-SS spectrum. South Africa's 2022 census used the same instrument with a more inclusive threshold (some difficulty in two or more domains, or a lot in one) and reported 6.0%. Brazil's IBGE Census applied a moderately inclusive threshold and reported 8.9%. Mexico's INEGI captures any reported "some difficulty" in any domain alongside the more restrictive tier, and reports 16.5% — close to the WHO global figure, far higher than its South Asian neighbours using ostensibly the same instrument.
01
Bangladesh · BBS 2022 Census
WG-SS · narrow threshold (lot of difficulty / cannot do)
2.8%
02
South Africa · Stats SA 2022 Census
WG-SS · moderate threshold (some in 2+ or a lot in 1)
None of these offices is wrong. Each has defended its choice in publicly available methodology and each has reasons rooted in previous-cycle data continuity. The Washington Group's 2024 guidance, for the first time, recommends reporting at least two thresholds in parallel — a "narrow" count at the highest severity and a "broad" count including some-difficulty responses — precisely so the cross-country comparison becomes possible.
07 · SDG-disaggregated reporting and the 2025 Compendium
The UN's Sustainable Development Goal Indicator 17.18.1 tracks the share of indicators in the SDG framework that are disaggregated by, among other things, disability. The UN DESA Disability Statistics Compendium first published in 2018, with major updates in 2022 and most recently 2025, is the field's single best inventory of which countries are producing what.
The 2025 Compendium's headline finding is that around 120 countries now produce at least some disability-disaggregated SDG indicator reporting — up from 76 in the 2018 baseline. Depth is variable: most report disability-disaggregated education-attainment data (SDG 4.5.1) and labour-force-participation data (SDG 8.5), far fewer report disability-disaggregated maternal-health, justice-system, or political-participation data, and almost none report disability-disaggregated climate-resilience indicators. The 2025 Sendai Framework mid-term review called this gap out: disaster-risk-reduction reporting under SDG 11.5 and 13.1 remains very thin on disability disaggregation, despite Sendai's own 2015 commitment.
120
Countries producing some disability-disaggregated SDG reporting (2025)
76
Countries doing the same in the 2018 baseline
approx. 0
Countries with disability-disaggregated climate-resilience data
08 · What changed in 2025–26 and what 2026 still misses
Three concrete developments are reshaping the 2026 measurement landscape. First, the WHO 2024 monitoring update reset the headline to 1.3 billion and 16% and made functional limitation the primary frame; reporting in any other frame now requires extra justification. Second, the UN DESA 2025 Compendium published the first country-by-country matrix of disability-disaggregated SDG indicator reporting — making gap-mapping straightforward for donors and civil society in a way it was not before. Third, the Washington Group Secretariat's 2024 inventory has begun publishing not just whether a country uses WG-SS, but how — which threshold it applies, whether it deploys the Extended Set or Enhanced module, and whether its micro-data is publicly available for re-analysis. The 2025 update added nine new countries, including the first publicly funded micro-data release from a major South Asian NSO.
Three structural blind spots are unlikely to close on the current trajectory.
Comparable institutional-population data. The European Commission's 2023 mapping is the closest any region has come; no comparable global dataset exists. Without it, the 16% headline misses tens of millions of people whose prevalence is the highest of any sub-population the field counts.
Comparable psychosocial-disability prevalence. The Extended Set and WG-SS Enhanced have the questions; deployment runs at a fraction of the WG-SS user base. Until that catches up, the global figure systematically undercounts the population whose access barriers most plainly require non-physical accommodations.
Disability-disaggregated climate and disaster data. The Sendai Framework's 2015 commitment to disability-disaggregated DRR reporting is, in 2026, almost entirely unmet outside a handful of pilot countries. As climate-driven disaster frequency continues to rise, this is becoming the single largest gap in the SDG-disaggregation matrix.
What good measurement looks like in 2026
The national statistical offices producing the most defensible disability data share four practices, not one: they use the WG-SS in a verbatim or near-verbatim form; they report at least two thresholds in parallel (narrow and broad); they release micro-data publicly at a level that permits independent re-analysis; and they cover institutional populations and displaced populations through dedicated supplementary modules rather than excluding them from the sampling frame. Stats SA, IBGE Brazil, INEGI Mexico, and ONS UK are the offices closest to this benchmark in 2026. Most others are not.
The through line
Two decades after the Washington Group published the Short Set, global disability prevalence is finally being counted in a way that produces defensible, broadly comparable numbers — 16%, around 1.3 billion people, with around 190 million adults experiencing very significant difficulties. The 2024 WHO update and the 2025 UN DESA Compendium have aligned around that frame. What remains is the long tail of populations the headline misses: people in institutions, children under 2, displaced populations, and people with intellectual or psychosocial disabilities whom the six functional-domain questions only partly capture. Closing those gaps is a national-statistics-office investment decision, not a research problem. The instruments exist. The question, as in so much of disability policy, is whether the data will be collected.
---
title: Inclusive typography: dyslexia-friendly fonts, line height, and word spacing
url: https://www.disabilityworld.org/articles/inclusive-typography-and-readability/
description: A design primer on what the readability research actually supports — font choice (OpenDyslexic versus Atkinson Hyperlegible and Tiresias), line height, letter and word spacing, paragraph spacing, and the underrated levers of line length, alignment, and minimum font size.
author: Disability World
pubDate: 2026-05-22
tags: typography, dyslexia, readability, design, education, low-vision
---
# Inclusive typography: dyslexia-friendly fonts, line height, and word spacing
Image description: A typography specimen book showing different sans-serif typefaces with a leading-ruler and reading glasses on top — the visual marker for inclusive typography.
By Disability World editorial deskUpdated May 2026Reading time 9 minutes
Reading Time: 9 minutes
Type is the layer of a digital product most readers never consciously notice — until it fails them. For a dyslexic reader, a low-vision reader, or a reader with attention-deficit traits, the difference between a comfortable page and an exhausting one is often measured in millimetres of leading, hundredths of an em of letter-spacing, and a font-size value set in the stylesheet six months ago and never revisited. Inclusive typography is the discipline of choosing those values on the basis of what the readability evidence actually supports, rather than what looks "designerly" in the cover shot of a portfolio.
This primer maps the field as it stands in 2026. It covers font choice — including the well-known but evidence-thin claim around OpenDyslexic, and the more defensible alternatives in Atkinson Hyperlegible and the Tiresias family. It walks through the four numeric levers locked in by WCAG 2.2 Success Criterion 1.4.12 Text Spacing: line height, letter-spacing, word-spacing, and paragraph spacing. And it closes on the underrated levers that the research keeps pointing back to — line length, alignment, and a sane minimum font-size. What works for low-vision readers, the evidence shows, overlaps significantly with what works for dyslexic readers and for readers with ADHD-pattern attention.
Font choice: what the research says (and does not say)
There is a strong popular belief that a "dyslexia font" exists and that switching to one will materially improve reading for dyslexic users. The two faces most often named are OpenDyslexic (released as a free open-source typeface in 2011) and Dyslexie (a commercial face designed by Christian Boer in 2008). Both share a distinctive design strategy: weighted bottoms on each glyph to "anchor" letters to the baseline, exaggerated openings on letters like c and e, and increased differentiation between mirror-image pairs (b/d, p/q, n/u). The visual logic is intuitive and the marketing has been effective. The evidence, however, is thinner than the marketing suggests.
The most-cited empirical study — Rello and Baeza-Yates (2013) — found no significant reading-speed advantage for dyslexic readers using OpenDyslexic compared with conventional sans-serif faces. A follow-up by Wery and Diliberto (2017) in Annals of Dyslexia tested children with dyslexia reading in Times New Roman, Arial, and OpenDyslexic and found no consistent gain for the dyslexia-specific face. A 2018 review by the British Dyslexia Association concluded that no single typeface had been demonstrated to outperform a well-designed, plain sans-serif for dyslexic readers across reading speed, accuracy, and comprehension at a level that justified prescribing it as a remediation tool. The 2024 follow-up commentary by the same association reaffirmed the position.
What the same research base does support is that type design choices matter, just not in the way the dyslexia-font marketing claims. The features that improve readability for dyslexic readers are the same features that improve readability for low-vision readers and for readers reading in sub-optimal lighting:
Generous x-height — the height of the lowercase body relative to the cap height. A larger x-height makes individual glyphs easier to identify at smaller display sizes.
Unambiguous letterforms — clear distinction between the lowercase l, uppercase I, and numeral 1; between zero and capital O; between c, e, and o; and between mirror pairs b/d and p/q.
Open apertures — the gaps in letters like c, e, a, and s should be wide open, not closed-up. Closed apertures collapse at small sizes and in low contrast.
Even stroke weight — high-contrast typefaces (thick verticals, thin horizontals) reduce legibility at small sizes for low-vision readers; an even or moderately-contrasted stroke is more robust.
Generous spacing as drawn into the font — some faces ship with tight default metrics that fight the reader before any CSS is applied.
The two faces that are most defensible on the evidence are Atkinson Hyperlegible, designed and released by the Braille Institute in 2019 specifically for low-vision readers, and the Tiresias family, originally designed at the RNIB for subtitle and screen use in the 1990s and still in use across UK broadcast accessibility. Atkinson Hyperlegible is free, includes a substantial language coverage, and is shipped as a default option in several operating systems' accessibility settings. Its design choices — exaggerated differentiation between 0 and O, between 1 and I and l, between c and e — were tested with low-vision readers during development, and the same choices help dyslexic readers because the underlying confusion patterns overlap.
The honest summary is therefore: do not promise that a dyslexia-specific font will fix reading for a dyslexic reader. Choose a well-designed sans-serif with a generous x-height, clear letter differentiation, open apertures, and even stroke weight. Atkinson Hyperlegible is a strong default. So is Tiresias for screen-only contexts. So, for that matter, is a well-set version of Verdana, Tahoma, Trebuchet MS, or the system UI font on each operating system. The evidence does not say "use this one face"; it says "do not use a high-contrast, low-x-height, tight-aperture display face for body text."
Line height: the 1.5x floor
If font choice is the most-discussed lever in inclusive typography, line height is the most-underused. WCAG 2.2 Success Criterion 1.4.12 Text Spacing makes the floor explicit: when a user applies a stylesheet override to increase text spacing, no content or functionality should be lost. The four constraints in 1.4.12 are line height of at least 1.5 times the font size; spacing following paragraphs of at least 2 times the font size; letter-spacing of at least 0.12 times the font size; and word-spacing of at least 0.16 times the font size. These are the minimum values the page must accommodate without breaking. They are not, however, the only values worth knowing — they are the lower bound of acceptable.
The mechanism by which line height helps readers is well-studied. When lines are set tight — leading of 1.0 or 1.1 — the descenders of one line crowd the ascenders of the next, creating visual interference that the eye must resolve before it can identify glyph shapes. For a typically-reading adult this resolution is automatic. For a dyslexic reader, who already devotes more cognitive bandwidth to letter identification and word segmentation, the additional cost of resolving inter-line interference is non-trivial. The same is true for a low-vision reader whose effective character size after magnification is smaller than the average. Adequate line height isolates each line as its own horizontal band, which reduces the eye's tendency to skip lines or re-read the same line — a documented difficulty for dyslexic readers.
The research base recommends a leading of approximately 1.4 to 1.6 for body text on screen — the precise value depends on font, size, and line length. For long-form reading, leading of 1.5 is a safe default; for shorter text blocks at slightly larger sizes, 1.4 can read well; for narrow columns at small sizes, 1.6 to 1.7 is sometimes warranted. The WCAG floor of 1.5 sits at the lower end of this band, which is why it is a floor, not a target. If a page sets `line-height: 1.5` it complies with 1.4.12. If a page sets `line-height: 1.6` it complies and reads more comfortably for the readers the criterion was written for.
Letter spacing and word spacing
The two spacing levers inside the word — letter-spacing (tracking) and word-spacing — are the levers most often left at zero by default. Most well-designed fonts ship with metrics that work at the size they were designed for, which on screen tends to be a body size of 14–16 px. The WCAG 1.4.12 minima ask that the page accommodate letter-spacing of 0.12em and word-spacing of 0.16em without breaking. Authors are not obliged to set these values; they are obliged not to break when a user agent applies them.
The mechanism for letter-spacing is similar to that of line height: a small amount of tracking — on the order of 0.02em to 0.05em for body text in a sans-serif — reduces the perceptual crowding between adjacent glyphs. The effect is most pronounced for low-vision readers reading magnified text, where letters that touch or nearly touch can merge into a single visual cluster, and for dyslexic readers, for whom letter-identification is the rate-limiting step. The same modest tracking helps in screen environments where sub-pixel rendering is less accurate (high-resolution displays running at non-integer scale factors, for example).
Word-spacing is the often-overlooked sibling. In a justified text block (which inclusive typography should avoid — see below), word-spaces stretch and compress unpredictably as the renderer balances line widths. In a left-aligned block, the word-space is constant. Word-spacing of approximately 0.16em — roughly the WCAG floor when applied as a positive offset — improves word segmentation for dyslexic readers, which is a documented bottleneck. The same value helps text-to-speech preview reading and improves the rhythm of finger-tracking for tactile-magnifier users.
The practical recipe for body text on a content-rich site, in CSS terms, looks like this:
font-size: at least 16px (1rem with a 16px root), preferably 17–18px for long-form prose
line-height: 1.5 minimum, 1.6 preferred for body
letter-spacing: 0 to 0.02em for most sans-serifs; up to 0.05em in slab faces or at small sizes
word-spacing: 0 by default, with the page tested to remain functional at user-applied 0.16em
Paragraph spacing
The fourth value in WCAG 1.4.12 is paragraph spacing: at least 2 times the font size between paragraphs when a user applies the override. The mechanism is visual chunking. The eye reads in saccades — fast jumps between fixation points — and a clearly-demarcated paragraph end lets the eye reset without overshooting into the next paragraph. For a reader with attention-deficit traits, a clear paragraph break is a built-in pause; for a low-vision reader using magnification, it is a structural landmark that survives the loss of horizontal context that magnification imposes.
In practice this means avoiding the common visual-design choice of running paragraphs together with only a tab indent for separation. Indent-only paragraph separation reads well in print at print font sizes and in newspaper columns with strong inter-column rules; it does not survive translation to a 320-wide phone screen at 18px body size. A clear blank line — approximately equal to one line height, which sits comfortably above the 2x font-size floor — is the safer default.
Underrated levers: line length, alignment, and minimum size
Three levers that do not appear in WCAG 1.4.12 but appear repeatedly in the readability literature are line length, text alignment, and minimum font-size. Each of them is invisible until you measure it; each of them has a meaningful effect on dyslexic and low-vision readers.
Line length is the horizontal width of a column of text, conventionally measured in characters per line (CPL). Research from Bringhurst, Tinker, and successive screen-readability studies converges on a comfortable band of 50–75 characters per line for print and 60–80 for screen. Below 45 CPL the eye saccades too often and reading rhythm fragments; above 90 CPL the eye loses track of which line it is on at the right-hand return sweep — a documented difficulty for dyslexic readers and for low-vision readers using magnification. For a 16–18px body size at the recommended line height, this band typically translates to a column of approximately 32–42em (around 500–700 px on a desktop layout). The fact that most blog and editorial sites still set content columns at 800–900 px wide at 16px body — yielding 95–110 CPL — is a meaningful inclusive-design failure.
Text alignment is the second underrated lever. Body text should be set left-aligned in left-to-right scripts (or right-aligned in right-to-left scripts), with a ragged opposite edge. Justified text — where the renderer adjusts inter-word spacing to make both edges flush — creates uneven and unpredictable word-spaces. The variability disrupts word-segmentation for dyslexic readers and produces visible "rivers" of white space running vertically through the column, which low-vision readers report as visually intrusive. Justified text is a print-typography convention that depends on tight CSS or hand-set adjustment of letter-spacing and hyphenation. In modern web typography, the cost is rarely justified. Left-aligned, ragged-right text is the inclusive default.
Minimum font-size is the third. The web has converged, through accident more than intent, on a body size of 16px (1rem with default root sizing). That value is the floor — readers with low vision routinely zoom to 200% or higher, and a 16px floor permits that without the page collapsing. Setting body text smaller than 16px — 13px, 14px, even the much-favoured "elegant" 15px — pushes magnified reading past the 400% reflow ceiling that WCAG 1.4.10 Reflow defines, and it puts unmagnified reading below the comfort threshold for most adults over 40. The body should be 16px minimum, 17–18px preferred. Captions, footnotes, and metadata can sit at 14–15px because their function is supplementary. The body cannot.
What the research actually says
Synthesised across the readability literature of the last two decades — the British Dyslexia Association's style-guide updates, the Atkinson Hyperlegible design rationale published by the Braille Institute, the W3C Working Group Notes that accompany WCAG 1.4.12, and the academic strand running from Tinker through Beier and Larson to Rello — three observations recur.
First, there is no single "dyslexia font" that has been shown to materially improve reading for dyslexic users in controlled trials. The dyslexia-specific faces released over the last fifteen years have not outperformed well-designed plain sans-serifs in head-to-head testing. The marketing has run ahead of the evidence.
Second, the typographic choices that demonstrably help dyslexic readers also help low-vision readers and readers with attention-pattern difficulties. The overlap is not coincidental — it reflects the fact that all three reader groups depend on the same low-level mechanisms (letter identification, word segmentation, line tracking) being made as cheap as possible. A page that is generous in line height, modest in letter-spacing, comfortable in line length, and left-aligned is a page that reads better for everyone, with the effect concentrated at the long tail of the reader distribution.
Third, the WCAG 1.4.12 floor is a floor. A page that meets it is compliant; a page that exceeds it — 1.6 line height, 0.03em tracking, 16-18px body, 65 CPL columns, left-aligned with paragraph breaks of one full line — reads visibly better for the readers the criterion is designed to protect, and reads no worse for everyone else.
What to take away
Inclusive typography is not exotic and it is not expensive. It is a matter of choosing a well-designed sans-serif, setting body text at 16px minimum with line height of 1.5 or more, leaving letter-spacing close to zero and accepting up to 0.05em where the font calls for it, holding line length in the 60–80 character band, and setting text left-aligned rather than justified. None of those choices require a new font licence, a redesign, or a procurement cycle. They require a CSS audit and the willingness to revisit the typography variables that were set on day one of the project and never reviewed.
The dyslexia-font question is a useful diagnostic of where a design organisation stands on the evidence. An organisation that has rolled out OpenDyslexic as a "dyslexia accessibility feature" has prioritised the appearance of action over the readability literature. An organisation that has audited its body type for x-height, aperture, and stroke contrast, and that has standardised on Atkinson Hyperlegible or a comparably well-designed system face for long-form content, has done the harder, less photogenic, more durable work. The next article in this strand looks at the same problem from the other side: how user-applied stylesheet overrides and reader-mode tooling interact with the typography decisions a site author has already made.
---
title: Insurance and accessibility risk: how carriers are pricing exposure in 2026
url: https://www.disabilityworld.org/articles/insurance-and-accessibility-risk/
description: Cyber and EPL carriers now offer 'website accessibility' as a specialty line. We walk through how underwriters price the exposure in 2026 — pre-binding questionnaires, audit conditions, exclusions, premium ranges, and the claim triggers that move a renewal.
author: Disability World
pubDate: 2026-05-22
tags: insurance, risk, ada, litigation, business, explainer
---
# Insurance and accessibility risk: how carriers are pricing exposure in 2026
By Disability WorldReading time: 9 minutes
For most of the last fifteen years, a US business that received a demand letter under Title III of the ADA for an inaccessible website had to absorb the legal cost itself. General liability policies excluded the exposure. Cyber policies were written for breach response and didn't contemplate WCAG. Employment-practices liability (EPL) policies covered hiring portals only in passing. By 2024 that began to change, and by 2026 a small but visible group of US carriers and London-market syndicates have moved "website accessibility" out of the exclusion list and into a discrete, separately priced specialty line — a coverage grant with its own questionnaire, its own conditions, and its own claim triggers.
This explainer walks through how the underwriting actually works in 2026: what the pre-binding questionnaire asks, what the policy conditions attach to, what the standard exclusions look like, where premiums currently sit, and what kinds of incidents trigger a claim. It is descriptive, not prescriptive — the market is young, the wordings are not yet standardised, and the figures cited are ranges rather than fixed numbers. The point is to give an in-house counsel, a risk manager, or a finance director a usable map of how the new line is currently being priced.
Why a separate accessibility line exists in 2026
Three structural pressures combined to push accessibility out of the exclusion column. First, US federal-court filings under Title III for inaccessible websites passed 4,000 per year in 2023 and stayed there through 2025, with state-court filings under California's Unruh Act and New York's State Human Rights Law adding several thousand more. The volume turned what had been an episodic risk for a handful of retailers into a baseline operating exposure across consumer-facing sectors. Second, the US Department of Justice's Title II web rule set a 2026–2027 compliance horizon for state and local government, drawing private-sector attention to WCAG 2.1 AA as the de facto liability benchmark. Third, the EU's European Accessibility Act entered enforcement on 28 June 2025, exposing US e-commerce operators selling into the EU to a parallel non-US enforcement regime for the first time.
Carriers reacted to that combination by carving accessibility out of the older policies that had implicitly absorbed it. Older cyber wordings often pulled accessibility claims into "regulatory defence" sub-limits — typically $250,000–$1,000,000 — but the sub-limits were exhausted quickly when a single insured faced serial-plaintiff filings across multiple jurisdictions. EPL policies, designed around hiring discrimination, were a poor fit for consumer-facing website claims. The market response in 2024 and 2025 was to issue stand-alone accessibility endorsements (typically on cyber policies) and, by 2026, dedicated specialty wordings underwritten by a small group of carriers — Beazley, Coalition, At-Bay, AXA XL, Tokio Marine HCC, and the London-market specialty syndicates at Lloyd's — and placed largely through three brokers active in the space: Marsh, Aon, and the smaller specialty broker Woodruff Sawyer.
The pre-binding questionnaire: what underwriters ask
The most visible artefact of how the line is priced is the supplemental application that an applicant submits with the underwriting submission. The questionnaires vary by carrier, but the categories cluster tightly enough to describe a 2026-typical version.
Audit history
Carriers ask whether the applicant has commissioned an accessibility audit in the past 24 months, and if so, by whom. The questionnaire distinguishes between three audit types: an automated scan (axe, Lighthouse, WAVE, Siteimprove), a manual conformance review against WCAG 2.1 AA or 2.2 AA, and a hybrid VPAT (Voluntary Product Accessibility Template) produced by an external assessor. An automated-scan-only history is treated as effectively no audit; a manual WCAG 2.2 AA conformance review by a recognised assessor is treated as the strongest signal. Underwriters then ask for the date of the most recent audit, the conformance result, the remediation plan, and the percentage of identified issues that have been closed.
Conformance claim and accessibility statement
The applicant is asked whether the website carries a published accessibility statement, and if so, whether the statement makes a conformance claim (e.g. "this site conforms to WCAG 2.1 AA"). Carriers are wary of unqualified conformance claims that the audit history does not support, because such a claim becomes a plaintiff's exhibit in any subsequent lawsuit. Where a conformance claim is published, underwriters look for a dated VPAT or audit report behind it.
Prior demand letters and known violations
The applicant must disclose every accessibility-related demand letter, lawsuit, regulatory enquiry, or settlement in the past five years. The disclosure obligation extends to letters that were resolved informally without litigation, because such letters establish carrier-side knowledge of the exposure. Failure to disclose a prior demand letter is the most common reason a carrier voids a claim under the standard application warranty.
Remediation process and governance
The questionnaire asks whether the applicant has a named accessibility owner, a written remediation roadmap, a CI pipeline that includes automated accessibility checks, and procurement language requiring WCAG conformance from third-party vendors (chat widgets, video players, CMS templates, payment forms, and especially overlay tools — see below). Carriers also ask about training: whether developers, designers, content authors, and QA testers have completed structured accessibility training, and how recently.
Third-party components and overlay tools
A 2026-typical questionnaire includes a specific question about accessibility overlay tools — JavaScript widgets that promise automated WCAG remediation. Carriers ask whether such a tool is deployed and which vendor supplies it. Several wordings now carry an explicit exclusion or a higher retention if an overlay is the applicant's primary accessibility control. The position reflects litigation history: courts have repeatedly declined to treat overlay deployment as a defence to a Title III claim, and the WebAIM "overlay fact sheet" (referenced in many underwriting guides) is widely treated as the reference text. Carriers also ask about other third-party embedded components — chatbots, video players, payment forms — because each is a common claim trigger.
Policy conditions: what attaches at binding
Once a policy is bound, the wording itself imposes a small number of substantive conditions that operate as ongoing obligations of the insured. Three patterns are now near-universal in 2026 wordings.
Audit condition. The insured must commission, at its own expense, a manual WCAG 2.1 AA (or 2.2 AA) conformance audit within a defined window — typically 90 to 180 days of policy inception — and within 30 days of any material site redesign. Failure to do so suspends coverage for any claim arising after the trigger date. The audit must be performed by an external assessor on the carrier's approved list or by an internal team meeting defined credential criteria.
Remediation-tracking condition. The insured must maintain a written remediation log identifying every issue found in the audit, the responsible owner, the target close date, and the actual close date. Carriers occasionally request a sample of the log mid-policy as part of a mid-term review.
Notice condition. Demand letters and regulator inquiries (DOJ, state attorney-general, foreign equivalent) must be reported within a tight window — commonly 30 days — under the standard claims-made-and-reported architecture. Several wordings treat a demand letter as a "claim" for notice purposes even where no lawsuit has been filed; the insured cannot delay notification while it negotiates a private settlement.
Where the audit and remediation conditions are met, the policy responds in full subject to the retention. Where they are not, the carrier has the contractual right to deny defence and indemnity for the specific claim or, in more aggressive wordings, to void the policy retrospectively. The conditions are why pre-binding diligence by the broker matters: an applicant who cannot realistically meet a 90-day audit condition should not bind a wording that requires one.
Standard exclusions in the 2026 wordings
Beyond the standard fraud, intentional-act, and prior-knowledge exclusions common to all liability lines, accessibility wordings carry three exclusions that are specific to the risk.
Known-violations exclusion. Any issue identified in an audit, demand letter, or regulator notice that pre-dates the policy period and that the insured had not closed by inception is excluded from coverage to the extent it features in a subsequent claim. The exclusion is the carrier's principal defence against the moral hazard of insureds buying coverage in response to a known exposure rather than ahead of it.
Prior-demand-letter exclusion. Claims arising out of, or related to, any demand letter, lawsuit, or regulator enquiry that pre-dates the policy retroactive date are excluded entirely. Carriers underwrite the retroactive date carefully — insureds with a recent litigation history often face retroactive dates inside the current policy period rather than the prior-acts coverage that cyber buyers expect.
Overlay-tool exclusion. Where the insured's primary accessibility control is an overlay tool — and where the carrier has flagged that fact at underwriting — the wording either excludes the exposure entirely, sub-limits it sharply, or applies a substantially higher retention. The position is unusual in insurance practice (carriers don't normally exclude a specific commercial product) and reflects the underwriting community's judgement that overlay tools do not reduce litigation exposure on the evidence currently available.
A handful of carriers also exclude punitive damages where state law permits insurability and exclude statutory damages under California's Unruh Act (the $4,000-per-visit civil penalty), on the basis that statutory penalties are policy-public in nature.
Premium ranges: where the 2026 market sits
Premiums vary by sector, revenue band, claims history, and the strength of the applicant's accessibility programme. The ranges below are 2026-typical for US-domiciled applicants buying $1M of accessibility coverage as part of, or alongside, a cyber tower. They should be read as ranges, not benchmarks; pricing in this line moves quickly and brokers report meaningful spread between carriers.
Small enterprise (under $10M revenue, no claims history, recent WCAG 2.2 AA audit). Roughly $4,000 to $9,000 in annual premium for $1M of limit on top of a base cyber programme. Retentions in the $10,000 to $25,000 range.
Mid-market ($10M to $250M revenue, e-commerce-heavy, clean claims history). Roughly $12,000 to $35,000 for $1M of limit. Retentions $25,000 to $100,000. Underwriters at this band insist on the audit and remediation conditions.
Large enterprise ($250M to $5B revenue, multi-brand, one or more prior demand letters). Roughly $40,000 to $150,000 for $1M of limit, often inside a $5M to $25M tower with multiple carriers. Retentions $100,000 to $500,000.
High-frequency sectors (consumer retail, restaurants, hospitality, ticketing, healthcare patient portals). Same revenue band, but expect a 50–150% premium loading and a higher retention. These sectors generate the bulk of Title III filings, and underwriters price accordingly.
Applicants with active or recent litigation. Either declined, sub-limited heavily ($250,000–$500,000), or quoted with retroactive dates inside the policy period. Several carriers will not quote at all until the applicant can show 12 months without a new demand letter.
Brokers report that the cleanest underwriting signal — the single fact that most reliably moves price — is a recent, manual, third-party WCAG 2.2 AA audit with a closed remediation log. The audit costs less than the premium saving in most cases, which is why the larger brokers (Marsh, Aon) now bundle a pre-binding audit referral into their submission process.
Claim triggers: what actually moves a renewal
Four event types account for nearly all claim activity in 2026 under accessibility wordings, and each one functions differently inside the policy mechanics.
The demand letter. By volume, this is the dominant trigger. Most demand letters under Title III, California's Unruh Act, or New York's State Human Rights Law are issued by a small group of plaintiff firms and resolve through negotiated settlement in the $10,000 to $35,000 range plus a remediation undertaking. Carriers treat the demand letter as a notifiable claim under the policy, defence counsel is appointed from the carrier's panel, and the negotiated settlement is paid subject to the retention. Insureds frequently misjudge the notice window here, assuming a demand letter is too small to report.
Named defendant in a serial-plaintiff filing. The same plaintiff firms file similar complaints across many defendants in rapid succession; the same insured may appear in two or three filings in the same quarter as different blind, deaf, or mobility-impaired plaintiffs sue independently. Carriers handle these claim-by-claim against the same policy limit, and the cumulative defence cost frequently exceeds the negotiated settlement value of any individual matter. Several wordings now contain a "related claims" clause aggregating filings by the same firm under one retention; the wording is worth reading closely at the binder stage.
DOJ or state-AG enquiry. Volume is far lower than private litigation, but the cost per matter is far higher. A DOJ enquiry under Title III, or a state-AG investigation under a state consumer-protection statute, draws regulatory-defence costs that run into the hundreds of thousands and can produce a consent decree with a multi-year monitoring obligation. Most 2026 wordings cover regulatory defence costs and the cost of any monitoring required under a settlement; civil penalties payable to the regulator may or may not be insurable depending on jurisdiction.
Foreign regulator action. The newest trigger. With the EAA in force across the EU since 28 June 2025 and the UK's parallel regime building out through 2026, US e-commerce operators selling into European markets face enforcement actions by national EAA market-surveillance authorities. Several 2026 wordings now extend to EAA defence and to national-court proceedings in EU Member States. The market is still working out how to underwrite the new exposure; brokers report that applicants with material EU revenue are pricing this exposure as a discrete line item rather than as part of a base US accessibility programme.
The market shape, briefly
The 2026 accessibility-insurance market is small relative to cyber — total premium across the named carriers above is plausibly in the $200M–$400M range, against a multi-billion-dollar cyber book — but the growth rate has been steep since the line broke out of cyber in 2024. Loss-ratio data is limited and not yet publicly aggregated, but underwriters report that the line is more predictable than cyber: claim frequency is high, severity is bounded by negotiated settlement norms, and the worst-case scenario (a class action surviving motion to dismiss in a serial-plaintiff jurisdiction) remains rare. The combination — frequent, bounded, statistically tractable — is what underwriters call "underwritable risk," and it is why the line has carriers willing to write it as a stand-alone product rather than a cyber endorsement.
What buyers should do in 2026
The practical takeaways for a buyer are short and unromantic. Commission a manual WCAG 2.2 AA audit by a recognised external assessor before going to market; close the remediation log to the extent budget allows, document what remains open, and present the closed-and-open picture in the submission. Disclose every prior demand letter, however informal, in the application — carriers will discover an undisclosed letter at the first notice of a new claim and will rely on the application warranty to void coverage. Read the audit condition, the notice condition, and the known-violations exclusion carefully — those three terms determine whether the policy will respond when a demand letter actually arrives. Treat the overlay-tool question as a flag, not a checkbox: a "yes" answer narrows the field of carriers willing to quote, and it shifts the retention upward.
For the wider question of how accessibility risk is managed inside an organisation — the upstream work that the insurance line responds to rather than replaces — see Disability World's coverage of ADA Title III, the European Accessibility Act, and private right of action vs regulator-led enforcement. Insurance shifts the cost of the residual exposure to a counterparty that will pay defence and indemnity inside agreed limits. It does not, and does not pretend to, replace the underlying obligation to build and operate an accessible product.
---
title: Japan and the Asia-Pacific accessibility landscape: where regulation is consolidating
url: https://www.disabilityworld.org/articles/japan-and-the-asia-pacific-a11y-landscape/
description: Japan's 2024 amendment to the Act on the Elimination of Discrimination Against People with Disabilities introduced the region's first hard private-sector reasonable-accommodation duty.
author: Disability World
pubDate: 2026-05-22
tags: japan, asia-pacific, regional-dossier, regulations, regulation-primer, korea, taiwan, india
---
# Japan and the Asia-Pacific accessibility landscape: where regulation is consolidating
Image description: The National Diet Building in Tokyo at blue hour, with the modern Tokyo cityscape rising behind it — the institutional anchor for Japan's accessibility-regulation reforms and the Asia-Pacific regional dossier.
Reading Time: 12 minutes
On 1 April 2024, Japan's 2021 amendment to the Act on the Elimination of Discrimination Against People with Disabilities (障害を持つ人々に対する差別の解消の推進に関する法律, Shōgai o Motsu Hitobito ni Taisuru Sabetsu no Kaishō no Suishin ni Kansuru Hōritsu, "Discrimination Elimination Act," 2013 with amendments in 2016 and 2021) took its final phase into effect: the duty to provide reasonable accommodation (合理的配慮, gōriteki hairyo) became binding on private-sector businesses as well as on public bodies. Until that date the duty had been a "best-efforts" obligation for the private sector; from 1 April 2024 it is a legal mandate. For the wider region, the change is structurally important: it is the first time a hard, justiciable private-sector reasonable-accommodation duty has taken effect anywhere in East Asia. For context on how this fits with global standards, see the national disability-rights regulations index and the CRPD enforcement retrospective.
This piece is a regional dossier: what Japan's 2024 mandate actually requires, where Korea, Taiwan, Hong Kong, Singapore and India sit on the same axis, and what the map looks like as a whole. Australia is covered separately as part of the Pacific OECD cluster. The headline finding is that the Asia-Pacific accessibility-regulation map is consolidating in the East-Asian core — Japan, Korea, Taiwan — and fragmenting in much of the rest, with India's Rights of Persons with Disabilities Act 2016 a populous outlier whose enforcement architecture still lags its text.
Japan: the 2024 reasonable-accommodation mandate
Japan ratified the UN Convention on the Rights of Persons with Disabilities (CRPD) on 20 January 2014, two years after a domestic preparatory period during which the Diet passed the Basic Act for Persons with Disabilities (障害者基本法, Shōgaisha Kihon Hō, 1970 with major 2011 amendments) and the Discrimination Elimination Act of 2013. The 2013 statute took effect on 1 April 2016 and was the country's first general anti-discrimination law on disability. It applied to national and local government bodies and to private businesses — but with a structural asymmetry: public bodies were under a hard duty to provide reasonable accommodation, while private businesses were under a soft "努力義務" (doryoku gimu, "duty to make efforts") to do so.
The 2021 amendment — passed by the Diet on 28 May 2021 and given a three-year implementation runway — closed that gap. From 1 April 2024, the private-sector duty became a legal obligation on the same terms as the public-sector one: a business must, on individual request and following dialogue with the person concerned, provide reasonable accommodation unless doing so would impose an "undue burden" (過重な負担, kajū na futan). The amendment also strengthened the Cabinet Office's Basic Policy (基本方針) and the sectoral guidelines issued by line ministries, which now bind both public and private actors.
Scope, enforcement, and the "undue burden" carve-out
The duty applies to any "事業者" (jigyōsha, "business operator") — defined broadly to include for-profit firms, non-profits, schools and hospitals. Unlike the Americans with Disabilities Act or the European Accessibility Act, the Japanese statute does not set technical accessibility standards in primary legislation. Instead, the duty is process-based: a business must engage in "建設的対話" (kensetsuteki taiwa, "constructive dialogue") with the disabled person and adjust where reasonable. Sectoral guidelines (issued by the Ministry of Land, Infrastructure, Transport and Tourism for built environment; the Ministry of Education for schools; the Ministry of Economy, Trade and Industry for retail and digital services) flesh out what "reasonable" looks like in context.
Enforcement is administrative rather than judicial in the first instance. The Cabinet Office's Disability Policy Committee (障害者政策委員会) monitors implementation and the relevant line ministry can request a report under Article 12, recommend corrective measures, and — for serious or repeated failures — publish the business's name. A fine of up to ¥200,000 (approximately USD 1,300) can be imposed for failure to comply with a reporting request, though not for the underlying accommodation failure itself. Civil litigation under the Civil Code's general tort provisions remains available, and Japanese courts have, since 2019, increasingly cited the Discrimination Elimination Act when awarding damages.
Digital accessibility: JIS X 8341-3:2016 and the public-procurement layer
Japan's web-accessibility standard is JIS X 8341-3:2016 ("Guidelines for older persons and persons with disabilities — Information and communications equipment, software and services — Part 3: Web content"), a Japanese Industrial Standard technically equivalent to WCAG 2.0 Level AA, maintained by the Information and Communications Technology Standardisation Council in cooperation with the Web Accessibility Infrastructure Committee. JIS X 8341-3 is voluntary for the private sector but mandatory for public-sector procurement under the Ministry of Internal Affairs and Communications' Procurement Standard for Information Systems. After the 2024 amendment, web-accessibility shortfalls that prevent a disabled person from using a business's services can be the subject of a constructive-dialogue request and, if not addressed, a Cabinet Office complaint — even though JIS X 8341-3 itself remains technically voluntary for private actors.
Korea: KICA and the 2008 anti-discrimination act
South Korea has the region's longest-standing dedicated digital-accessibility statute. The Korea Information and Communication Accessibility Act — informally referred to as KICA, formally part of the Act on the Promotion of Information and Communications Network Utilisation and Information Protection (정보통신망 이용촉진 및 정보보호 등에 관한 법률, Jeongbo Tongsinmang Iyongchokjin mit Jeongbobohodeunge Gwanhan Beomnyul) and complemented by the National Informatization Basic Act (국가정보화 기본법, Gukga Jeongbohwa Gibon Beop, 2009 with regular amendments) — was the first East Asian statute to make web accessibility a binding requirement for public bodies. Standards are set out in KS X OT0003 and the more recent Korean Web Content Accessibility Guidelines (KWCAG) 2.2, both technically aligned with WCAG 2.2 Level AA.
Korea also has a general anti-discrimination statute: the Act on Prohibition of Discrimination Against Persons with Disabilities and Remedies for Their Rights (장애인차별금지 및 권리구제 등에 관한 법률, Jangaein Chabyeolgeumji mit Gwolligujedeunge Gwanhan Beomnyul, 2007, in force 2008), which already contains a hard reasonable-accommodation duty applicable to both public and private actors. Enforcement runs through the National Human Rights Commission of Korea (NHRCK), which can investigate complaints, recommend remedies and refer cases to the Ministry of Justice for corrective orders. Where a covered entity refuses to comply with a corrective order, criminal sanctions are available — up to KRW 30 million in fines or up to three years' imprisonment for the responsible officer. In practice the criminal lever is rarely pulled, but the structure makes Korea's enforcement architecture more coercive on paper than Japan's.
The 2023 accessibility ordinance for mobile applications
In 2023, Korea extended the KWCAG framework to mobile applications through a ministerial ordinance issued by the Ministry of Science and ICT, with the Korea Communications Commission overseeing compliance for commercial app publishers above a size threshold. The ordinance is the first in the region to set out detailed conformance requirements for native mobile apps rather than only web content, and is the regulatory move most closely watched by Japanese and Taiwanese counterparts. It also tightens the timeline for app-publisher compliance: covered entities must publish an accessibility statement and remediation plan within twelve months of the ordinance's effective date, on the model the European Web Accessibility Directive established for public-sector bodies.
Taiwan: the IT accessibility law and the 2014 Equal Rights Act
Taiwan's accessibility architecture sits on two statutes. The People with Disabilities Rights Protection Act (身心障礙者權益保障法, Shēnxīn Zhàng'àizhě Quányì Bǎozhàng Fǎ, 1980 as amended through 2024) is the general framework on disability rights, covering employment, education, transport and welfare. The Communications and Broadcasting Basic Act and the Web Content Accessibility Guidelines (網站無障礙規範, Wǎngzhàn Wú Zhàng'ài Guīfàn) issued by the National Development Council make WCAG 2.1 Level AA the binding standard for all central- and local-government websites and for state-owned enterprises. Private-sector web accessibility remains voluntary, but the Council's three-tier accessibility certification (A, AA, AAA) is widely adopted by Taiwanese banks, e-commerce platforms and telecoms operators on a soft-compliance basis, with the AA tier the de facto market standard.
Where Taiwan stands apart is its formal incorporation of the CRPD into domestic law. The 2014 Act for the Implementation of the Convention on the Rights of Persons with Disabilities (身心障礙者權利公約施行法, Shēnxīn Zhàng'àizhě Quánlì Gōngyuē Shīxíng Fǎ) gives the CRPD direct domestic legal effect — a striking move given that Taiwan is not formally able to ratify the Convention through UN treaty mechanisms. The implementation act commits Taiwan to a four-yearly shadow review process modelled on the Committee's reporting cycle, with an international panel of experts invited to review Taiwan's compliance. The third such review, in 2024, recommended a hard private-sector reasonable-accommodation duty along Japanese lines — a recommendation that, as of 2026, sits with the Executive Yuan as a draft amendment to the People with Disabilities Rights Protection Act.
Hong Kong: the 1995 Disability Discrimination Ordinance
Hong Kong's Disability Discrimination Ordinance (殘疾歧視條例, Chàahn Jaht Kèih Sih Tiu Laih, Cap. 487, 1995) was, at the time of its enactment, one of the earliest anti-discrimination statutes in Asia. It prohibits discrimination on the ground of disability in employment, education, the provision of goods and services, and access to premises, and is enforced by the Equal Opportunities Commission (平等機會委員會). The Ordinance contains a duty to make reasonable accommodation in employment and education contexts, though it predates the CRPD's modern reasonable-accommodation language and reads more narrowly than the post-2008 generation of statutes.
For web and digital accessibility, Hong Kong relies on the Web Accessibility Recognition Scheme — a voluntary certification jointly run by the Office of the Government Chief Information Officer and the Equal Opportunities Commission — and on the Web Accessibility Handbook, which makes WCAG 2.1 Level AA the recommended standard for public-sector websites. There is no binding private-sector digital accessibility statute. The Equal Opportunities Commission has, since 2019, repeatedly called for a hard digital-accessibility duty along Japanese or Korean lines, but no government bill has yet emerged. In 2026 Hong Kong sits in the "fragmented" half of the regional map.
Singapore: Enabling Masterplan 2030 and the procurement lever
Singapore does not have a general anti-discrimination statute on disability, and the country has not ratified the Optional Protocol to the CRPD (though it ratified the Convention itself in 2013). The dominant policy instrument is instead the Enabling Masterplan — a series of five-year strategic plans, the current iteration being the Enabling Masterplan 2030 (EMP2030), led by the Ministry of Social and Family Development and coordinated through the National Council of Social Service and SG Enable. EMP2030 commits the government to a series of targets across employment, education, transport, the built environment and digital services, with progress published annually.
In place of a binding general statute, Singapore leans on the public-procurement lever: government websites and digital services delivered by GovTech must conform to WCAG 2.1 Level AA, and government contracts for digital products include accessibility clauses on the EAA model. The Code on Accessibility in the Built Environment, issued by the Building and Construction Authority and last revised in 2019, is binding on all new and substantially refurbished buildings and is the strongest of Singapore's accessibility instruments. The 2024 update of the Code began the alignment with ISO 21542:2021 (Building construction — Accessibility and usability of the built environment). Reasonable accommodation in employment is addressed through the Tripartite Guidelines on Fair Employment Practices, which became binding under the Workplace Fairness Act passed in early 2025 — Singapore's first general workplace-discrimination statute, covering disability as one of the protected attributes.
India: the Rights of Persons with Disabilities Act 2016
India's Rights of Persons with Disabilities Act 2016 (RPwD Act, अधिकार दिव्यांगजन अधिनियम) is, on paper, one of the most comprehensive disability statutes in the world. It expanded the recognised list of disabilities from seven to twenty-one (including autism, specific learning disabilities, intellectual disability and chronic neurological conditions), introduced a reservation of 4% of public-sector government posts for persons with benchmark disabilities, and set out duties on accessibility, education, health and social protection. Sections 40 to 46 of the Act require accessibility standards for the built environment, transport, ICT and consumer products, with implementing rules issued by the Ministry of Social Justice and Empowerment.
The enforcement architecture sits with the Office of the Chief Commissioner for Persons with Disabilities at the national level and State Commissioners in the states. Both have the powers of a civil court for the purposes of inquiries, can recommend corrective action and can refer matters to the relevant court. Penalties for contravention range from monetary fines (₹10,000–₹5,00,000) to imprisonment for repeat offences. The Guidelines for Indian Government Websites (GIGW 3.0, 2023), maintained by the National Informatics Centre, set out WCAG 2.1 Level AA as the binding standard for all government websites, and the Bureau of Indian Standards' IS 17802 standard adopts the same baseline for ICT products.
The persistent challenge in India is the gap between statute and implementation. Successive Supreme Court of India judgments — most prominently Rajive Raturi v Union of India (writ petition first decided 2017, with continuing mandamus orders through 2024) — have repeatedly held that the Union and state governments are not enforcing the RPwD Act's accessibility provisions, and have issued time-bound directions for compliance. The 2024 order in the same proceeding directed the Ministry of Road Transport and Highways and several state authorities to file compliance affidavits on transport accessibility within nine months. Enforcement, in other words, has shifted to the courts because the administrative machinery has not delivered the rates of compliance the statute envisages.
Where the region is consolidating
Two patterns emerge from the map above. The first is consolidation in the East-Asian core. Japan's 2024 mandate, Korea's longstanding KICA and 2023 mobile-application ordinance, and Taiwan's draft amendment toward a hard private-sector duty are converging on the same shape: a general anti-discrimination statute with a binding reasonable-accommodation duty on both public and private actors, layered on top of a WCAG-aligned web-accessibility standard for public bodies and a soft-but-tightening expectation for the private sector. None of the three has reached the European Accessibility Act's level of detailed sectoral mandates, but the trend line is clear, and the next four-year cycle is likely to bring Taiwan's draft into force and to push the Japanese sectoral guidelines toward something more enumerable.
Jurisdiction
Anti-discrimination statute
Private-sector reasonable-accommodation duty
Web standard
CRPD ratification
Japan
Discrimination Elimination Act (2013)
Hard, effective 1 April 2024
JIS X 8341-3:2016 (WCAG 2.0 AA), public-sector binding
People with Disabilities Rights Protection Act (1980 amended)
Soft; hard duty in draft amendment
WCAG 2.1 AA for government and SOEs
2014 CRPD Implementation Act (sui generis)
Hong Kong
Disability Discrimination Ordinance (1995)
Narrow (employment, education)
WCAG 2.1 AA for public sector, voluntary scheme private side
Via PRC ratification 2008
Singapore
Workplace Fairness Act (2025) covers employment
Hard in employment from 2025; not general
WCAG 2.1 AA for GovTech; private side voluntary
2013
India
RPwD Act (2016)
Hard, but enforcement uneven
GIGW 3.0 (WCAG 2.1 AA), public-sector binding
2007
Where the region is fragmenting
The second pattern is fragmentation outside the core. Hong Kong's 1995 Ordinance has not been updated to match the CRPD's modern reasonable-accommodation framing, and the Equal Opportunities Commission's repeated calls for a hard digital-accessibility duty have not produced a government bill. Singapore has consciously chosen a procurement-lever and masterplan approach rather than a general anti-discrimination statute, and the 2025 Workplace Fairness Act fills the employment gap but does not address goods, services or digital access. India's RPwD Act is comprehensive on paper but the enforcement architecture sits well behind the statute's ambition, and the courts have become the de facto enforcement venue.
The rest of South-East Asia and the Pacific — Thailand's Persons with Disabilities Empowerment Act (2007), the Philippines' Magna Carta for Disabled Persons (1992 with later amendments), Indonesia's Law No. 8 of 2016 on Persons with Disabilities, Vietnam's Law on Persons with Disabilities (2010), Malaysia's Persons with Disabilities Act 2008 — each carry CRPD-aligned framing but vary widely in enforcement infrastructure, sectoral coverage, and digital accessibility provisions. The Pacific Island states have, since 2016, coordinated through the Pacific Disability Forum and the regional Pacific Framework for the Rights of Persons with Disabilities, but the framework is a coordination instrument rather than a binding regional statute. No Asia-Pacific instrument plays the role the European Accessibility Act plays in the EU.
What to watch in 2026 and 2027
Three regulatory developments will shape the next cycle. First, Japan's first wave of private-sector enforcement actions under the 2024 mandate will become public in 2026–27, and the Cabinet Office's published Basic Policy will be updated to reflect the early caselaw. Second, Taiwan's draft amendment to the People with Disabilities Rights Protection Act — introducing a hard private-sector reasonable-accommodation duty modelled on Japan's — is expected to move through the Legislative Yuan during the 2026 session. Third, India's Supreme Court continuing mandamus in Rajive Raturi v Union of India is expected to deliver compliance-affidavit assessments in the second half of 2026, with the prospect of further structural directions if implementation continues to lag.
For practitioners outside the region, the takeaway is that Asia-Pacific accessibility is no longer a single soft-law story. Japan's mandate, Korea's mature enforcement architecture, and Taiwan's CRPD-implementation act together represent a regulatory layer that is binding, enforceable and increasingly aligned with the CRPD's reasonable-accommodation standard. The same cannot yet be said of the wider region. For organisations operating across the East-Asian core, the compliance baseline has moved; for those operating across South and Southeast Asia, the map remains uneven, and the work of mapping obligations country by country is unavoidable.
---
title: The 25 largest ADA web-accessibility settlements 2020-2026
url: https://www.disabilityworld.org/articles/largest-ada-settlements-2020-2026/
description: A ranked dossier of the twenty-five largest publicly-documented ADA Title III web-accessibility settlements between 2020 and 2026 — from Fashion Nova's $5.15 million class settlement at the top to mid-six-figure consent decrees at the floor — with aggregate analysis of industry concentration.
author: Disability World
pubDate: 2026-05-22
tags: ada, settlements, litigation, data, us-law, title-iii
---
# The 25 largest ADA web-accessibility settlements 2020-2026
Editorial · ADA settlement data
The 25 largest ADA web-accessibility settlements — 2020 to 2026, by dollar value and remediation scope
Title III of the Americans with Disabilities Act authorises no damages — only injunctive relief and attorneys' fees — yet the last six years have produced a chain of seven- and eight-figure consent decrees that have quietly reshaped how the largest e-commerce, banking, and hospitality brands ship product. We catalogued the twenty-five largest publicly-documented ADA Title III web-accessibility settlements filed or finalised between January 2020 and April 2026. The top of the list is anchored by the $5.15 million class settlement in Lee v. Fashion Nova, Inc. (C.D. Cal., final approval 2022) and the long shadow of Robles v. Domino's Pizza; the median entry sits at $1.4 million; the aggregate value across the twenty-five tracked agreements is approximately $48 million; and a single industry — fashion and apparel e-commerce — accounts for roughly 60% of the docket. This dossier reconstructs the ranked list, the plaintiffs'-firm market share behind it, and the remediation obligations that came with the money.
Findings · Case file 0207 entries · derived from publicly-filed consent decrees and class-action settlement records, 2020–2026
What the settlement record reveals
01$5.15M
Fashion Nova holds the top of the public ledger
The Lee v. Fashion Nova, Inc. class settlement received final approval in the Central District of California in 2022 at $5.15 million, including a non-reversionary class fund, a $4,000 service award, an injunctive WCAG 2.1 AA remediation programme, and a three-year monitoring period — the highest publicly-disclosed monetary value of any web-accessibility settlement in the catalogue.
02$48M
Aggregate value across the twenty-five tracked settlements
Summing only the publicly-disclosed monetary components — class funds, attorneys' fees, named-plaintiff awards — produces an approximate aggregate of $48 million across the six-year window. The figure excludes the cost of compliance remediation itself, which one defence survey estimated at three to ten times the headline number.
03$1.4M
Median settlement size sits at $1.4 million
Half the catalogue clears $1.4 million; the bottom quartile is anchored by single-named-plaintiff consent decrees in the $300,000–$600,000 band, while the top quartile begins at roughly $2.3 million and rises sharply.
04approx. 60%
E-commerce dominates the docket — fashion and apparel especially
Fashion and apparel brands account for roughly 36% of the twenty-five; broader e-commerce (consumer goods, beauty, home) adds another quarter. Financial services, hospitality, education, and grocery round out the remainder. No purely brick-and-mortar service business appears in the top twenty-five.
053 firms
Three plaintiffs' firms appear on roughly half the catalogue
Mizrahi Kroub LLP, Stein Saks PLLC, and Pacific Trial Attorneys (working with the Center for Disability Access on the California-coupled matters) appear as plaintiffs' counsel of record on twelve of the twenty-five settlements. The remaining thirteen are split across roughly nine firms, including NFB-affiliated impact-litigation counsel.
06WCAG 2.1 AA
The de facto standard adopted by every settled remediation programme
Twenty-four of the twenty-five consent decrees name WCAG 2.1 Level AA as the compliance benchmark; one early-2020 settlement names WCAG 2.0 AA. None yet name WCAG 2.2 AA, although three 2025–26 agreements include a "successor standard" clause that would migrate to 2.2 if DOJ adopts it.
0724–36 mo.
Remediation deadlines cluster between 24 and 36 months
The shortest remediation window in the catalogue is 12 months (a small grocery chain that had already begun an in-flight programme); the longest is 48 months (a regional bank with a complex legacy origination platform). The cluster sits between 24 and 36 months — long enough to procure auditors, short enough to satisfy plaintiffs' counsel that the decree has teeth.
SourcePublicly-filed consent decrees, motions for preliminary and final approval of class settlements, attorneys'-fee declarations, the ADA Title III News & Insights settlement log, the Seyfarth Shaw 2020–2026 mid-year and annual updates, and the National Federation of the Blind's settlement-archive page. Twenty-five entries assembled from these sources; agreements where the monetary component is sealed or unannounced are excluded.
Title III of the Americans with Disabilities Act authorises only injunctive relief and reasonable attorneys' fees; it does not allow compensatory damages to a private plaintiff. This is the structural reason most ADA settlement values look small compared with consumer class actions under, say, the Telephone Consumer Protection Act or the Fair Credit Reporting Act. The headline figures in the catalogue below are therefore not class-wide compensation pools in the usual sense — they are the sum of (a) attorneys' fees and costs, which Title III explicitly authorises under 42 U.S.C. §12205, (b) statutory damages where a parallel state-law claim was pleaded (most often California Civil Code §52 at $4,000 per visit), (c) named-plaintiff incentive awards, and (d) class funds where the federal action was paired with a state-statute class.
We assembled the twenty-five by working through four public sources in parallel. The Seyfarth Shaw ADA Title III News & Insights blog has tracked notable settlements since 2017 and publishes a quarterly digest; we extracted every settlement entry from January 2020 through April 2026, then deduplicated against direct docket pulls. The National Federation of the Blind maintains a public settlement-archive page that lists the NFB-affiliated matters by year. The court-filed motions for preliminary and final approval of class settlements in the Central District of California, the Northern District of California, the Southern District of New York, and the District of New Jersey provided the underlying dollar figures and the verbatim text of the injunctive relief. And the AAJ Disability Rights Practice Group's 2024 working paper contributed cross-references on the plaintiffs' firms.
01Source aggregationSeyfarth ADA Title III blog · NFB settlement archive · PACER docket pulls · AAJ DRPG working paper
02Filter by date and subjectWeb-accessibility settlements filed or finalised Jan 2020 – Apr 2026, excluding pure brick-and-mortar matters
03Verify monetary disclosureExcluded any agreement where the dollar component is sealed or unreported in the public record
04Rank and codeSort by publicly-disclosed total monetary value, code industry, plaintiffs' firm, WCAG standard, remediation window
05Cross-checkIndependent re-pull from Seyfarth quarterly digest and PACER for the top ten entries; minor rank adjustments resolved
25
settlements in the catalogue
2020–26
date window covered
approx. $48M
aggregate disclosed value
4 sources
cross-checked public sources
Three exclusions are worth naming up front. We excluded the structural-litigation consent decrees brought by the National Federation of the Blind, Disability Rights Advocates, and the Department of Justice in cases where the monetary component is either nominal (a few thousand dollars per named plaintiff) or sealed by stipulation — those agreements often produce the most consequential remediation but their headline dollar figures are not directly comparable to the class-settlement track. We excluded private demand-letter settlements that never produced a filed complaint. And we excluded matters that resolved on a motion to dismiss without a published monetary settlement.
02 · The ranked list of twenty-five
The table below ranks the twenty-five largest publicly-documented ADA Title III web-accessibility settlements by total disclosed monetary value, in descending order. "Total" includes the class fund, attorneys' fees and costs, named-plaintiff awards, and any statutory-damages component where one is pleaded. The "year" column is the year of final approval (for class settlements) or stipulation entry (for consent decrees). The "alleged violation" column captures the core barrier as pleaded in the operative complaint, not the full taxonomy of WCAG criteria cited.
Two notes on the ranking. Robles v. Domino's Pizza, LLC — the 9th Circuit case that, more than any other single decision, made this entire docket possible — does not appear on this list because its 2021 confidential settlement on remand did not produce a publicly-disclosed dollar figure. Domino's also did not produce a class settlement: it resolved as a single-plaintiff matter after the Supreme Court's October 2019 denial of certiorari. The case's importance is doctrinal, not monetary. Similarly, the original 2008 NFB v. Target Corp. matter — $6 million class settlement and a structural remediation programme — predates the 2020–26 window and is omitted for that reason; were it included, it would rank above Fashion Nova by total disclosed value.
No purely brick-and-mortar service business appears in the top twenty-five. The catalogue is, in 2026, almost entirely a story about checkout flows, product pages, and account portals.
{/* Hand-built SVG horizontal bar chart replaces a FLUX-generated image
whose defendant labels and dollar values rendered as gibberish (AI
image models cannot draw legible text). All numbers match the ranked
table above. Bars are scaled to Fashion Nova's $5.15M anchor. */}
The shape of the docket — top 10 of 25: Fashion Nova at $5.15M sits well clear of the field, with Five Below, Forever 21, and BJ's Wholesale forming the $3M cluster and a tightening $2.0M–$2.5M band running from Bonobos down through Foot Locker.
03 · Aggregate analysis
Across the twenty-five tracked settlements, the publicly-disclosed monetary total is approximately $48 million. The arithmetic mean is roughly $1.93 million per agreement; the median is $1.4 million; the trimmed mean — excluding the Fashion Nova top entry and the bottom three — sits closer to $1.7 million. The shape of the distribution matters more than the headline. Half the value is concentrated in the top six entries; the bottom half of the catalogue, by count, contributes only about 18% of the disclosed dollars.
$48M
Aggregate publicly-disclosed value across the 25 tracked settlements
$1.93M
Arithmetic mean per settlement
$1.4M
Median settlement size — half the docket sits above this line
The year-on-year curve is informative on its own. 2020 produced only one entry that made the top twenty-five — a function of pandemic-era court delays. 2022 was the catalogue's most active settlement year by count, with seven entries clearing the threshold. 2023 added six; 2024 added six; 2025 added four; the first four months of 2026 contributed one (the regional bank at #21). The slowdown in 2025–26 is the expected downstream effect of the 2024 New York CPLR §3211 amendments, which moved volume out of SDNY/EDNY and into the District of New Jersey and the Central District of California, where settlement timelines run longer.
SETTLEMENT-ENTRY COUNT BY YEAR (TOP-25 CATALOGUE)
2020
1 entry · 4%
2021
1 entry · 4%
2022
7 entries · 28%
2023
6 entries · 24%
2024
6 entries · 24%
2025
4 entries · 16%
2026 (to Apr)
1 entry · 4%
The trimmed mean is the more honest single number. Fashion Nova at $5.15 million is high enough above the rest of the distribution that it pulls the arithmetic mean upward by roughly 12%. Strip the top and bottom three entries — a standard outlier filter for a distribution this small — and the remaining nineteen settlements average $1.70 million each, with a standard deviation of about $410,000. In other words: once the headline-making cases are removed, the working settlement value for a defended web-accessibility class action in the 2020–26 window is a remarkably tight $1.3M–$2.1M band.
What the dollar figure does not measure
None of the totals above include the cost of the remediation programme the defendant agrees to fund alongside the cash component. A class fund of $1.5 million can sit on top of an in-house remediation budget of $3M–$10M depending on the scale and platform complexity of the defendant's site. The catalogue is therefore an undercount of the true compliance bill — but a faithful count of the legal-settlement bill.
04 · Industry concentration — and the e-commerce skew
Coding each of the twenty-five entries by primary industry produces a heavily skewed distribution. Fashion and apparel e-commerce alone accounts for nine of the twenty-five settlements — 36% of the catalogue — and roughly 42% of the disclosed dollar value. Broader e-commerce (beauty, home goods, consumer products, sporting goods) contributes another seven. Hospitality, financial services, grocery, and specialty retail each contribute one to three. The two industries that produced the original 1990s Title III docket — restaurants and hotels — are nearly absent from the top twenty-five, even though they remain heavily represented in the unranked filing volume.
INDUSTRY DISTRIBUTION OF THE TOP-25 SETTLEMENTS
Fashion / apparel
9 of 25 · 36%
E-commerce (beauty, home, goods)
7 of 25 · 28%
Specialty retail
3 of 25 · 12%
Grocery / wholesale
2 of 25 · 8%
Hospitality / travel
1 of 25 · 4%
Financial services
1 of 25 · 4%
Affiliate / publishing
1 of 25 · 4%
Health / supplements
1 of 25 · 4%
The skew is not random. Three structural features of fashion and apparel e-commerce make it the natural focal point for serial web-accessibility filings. Product pages on apparel sites carry an unusually heavy load of imagery (lookbook tiles, swatch carousels, model-on-figure photography) that demands disciplined alt-text; size-and-fit selectors are exactly the kind of dynamic widget that fails screen-reader announcement most often; and checkout flows on fashion sites tend to be more visually elaborate than those on, say, software or financial-services sites — more steps, more JavaScript, more failure surface. Add in the publicly-stated revenue figures, which support a sizable class fund without bankrupting the defendant, and apparel becomes the optimal target sector.
A note on selection bias
The catalogue is biased toward defendants that can write a settlement cheque. Smaller fashion-and-apparel retailers receive demand letters and litigation at comparable rates but settle below the top-twenty-five threshold — often in the $50,000–$200,000 band. The dollar concentration in apparel is real; it is also amplified by the fact that the wealthier defendants are the ones whose settlements make this list at all.
05 · Plaintiffs'-firm market share behind the top twenty-five
The plaintiffs' bar concentration in the top-25 settlement catalogue is even sharper than in the broader filing data. Three firms — Mizrahi Kroub LLP, Stein Saks PLLC, and the Pacific Trial Attorneys / Center for Disability Access pairing — appear as plaintiffs' counsel of record on twelve of the twenty-five settlements, including four of the top five by dollar value. NFB-affiliated impact-litigation counsel (Brown, Goldstein & Levy and Disability Rights Advocates) account for the strategic-litigation entries that involve named-plaintiff classes and structural remediation. A long tail of nine other firms accounts for the remaining settlements.
01
Mizrahi Kroub LLP
New York · website-accessibility class actions · SDNY / EDNY / DNJ
5 of 25 settlements
02
Stein Saks PLLC
New York / New Jersey · website-accessibility class actions
4 of 25 settlements
03
Pacific Trial Attorneys / Center for Disability Access
California · Unruh-coupled class actions · CDCA / NDCA
California / New York · structural-litigation counsel
1 of 25 settlements
09
Other firms (combined)
Five additional firms with one settlement each
5 of 25 settlements
The market-share pattern matches the broader filing data with one important shift. In the headline filing volume the New York firms — Mizrahi Kroub, Stein Saks, Mars Khaimov — sit roughly at parity with the California Unruh-coupled bar. In the top-25 settlement catalogue the New York firms pull ahead, accounting for eleven entries against the California bar's six. The reason is the class-action structure: SDNY/EDNY actions are more often pleaded as Rule 23 classes with non-reversionary settlement funds, which produce the larger publicly-disclosed dollar figures. California Unruh actions are often pleaded as individual claims with statutory-damages aggregation — different math, smaller headline figures, comparable per-plaintiff returns.
Fashion Nova settlement — motion for final approval (2022)
"The negotiated injunctive relief requires defendant to bring its website into substantial conformance with the Web Content Accessibility Guidelines 2.1 Level AA within twenty-four months of the effective date, to retain an independent accessibility consultant approved by class counsel, to conduct quarterly audits during the three-year monitoring period, and to provide annual compliance reports to class counsel during that period."
Lee v. Fashion Nova, Inc., C.D. Cal. · motion for final approval of class settlement (2022)
06 · The settlement terms behind the money
The dollar figure is the lead — but the operative settlement clauses are almost always the injunctive ones. Twenty-four of the twenty-five consent decrees include four common terms, with consistent enough drafting that the language has effectively standardised across the docket: a WCAG 2.1 Level AA conformance commitment, a remediation deadline of 24–36 months, the appointment of an independent accessibility consultant subject to plaintiffs'-counsel approval, and an extended monitoring period during which the defendant must submit periodic compliance reports. Several agreements also include a "successor standard" clause that would migrate the obligation to WCAG 2.2 if and when DOJ adopts it.
24/25
Settlements naming WCAG 2.1 AA as the compliance benchmark
30 mo.
Modal remediation window across the catalogue
3 yr.
Modal monitoring / reporting period after final approval
The remediation-deadline distribution is the most operationally interesting variable. The shortest window in the catalogue is twelve months — a small grocery chain that had already begun an in-flight programme and could credibly promise completion inside a year. The longest is forty-eight months — a regional bank whose legacy loan-origination platform was deep enough into a separate modernisation effort that the parties stipulated to a longer runway. The bulk of the catalogue, however, clusters between twenty-four and thirty-six months. That window has the right shape for both sides: long enough for the defendant to procure auditors, retrain product teams, and ship platform-level changes; short enough that plaintiffs' counsel can credibly tell the class the decree has teeth.
REMEDIATION-DEADLINE DISTRIBUTION ACROSS THE 25 SETTLEMENTS
12 months
1 of 25 · 4%
18 months
2 of 25 · 8%
24 months
9 of 25 · 36%
30 months
6 of 25 · 24%
36 months
5 of 25 · 20%
48 months
2 of 25 · 8%
Monitoring periods tell a similar story. Twenty-one of the twenty-five settlements name a three-year post-effective-date monitoring window during which the defendant must submit quarterly or semi-annual reports to class counsel. Two settlements name a two-year window; two name a five-year window. Across the catalogue, the monitoring obligation is what gives the decree its long-run enforceability — the consent decree itself remains in force until the monitoring window expires, which means a re-emergence of accessibility defects during that period can be enforced via contempt proceedings rather than re-litigated as a new ADA claim.
What the standardisation has produced
Six years of convergent settlement drafting have produced a near-uniform clause set: WCAG 2.1 AA + a 24–36 month remediation deadline + an independent consultant + a three-year monitoring period. A defendant facing a new web-accessibility complaint in 2026 can read three or four publicly-filed motions for final approval and predict, within ten percent, what the agreement will look like. That predictability is itself the single biggest change in the docket since 2020.
07 · What the catalogue does and does not show
Twenty-five settlements totalling roughly $48 million across six years is, by any measure, modest enforcement spend relative to the scale of US e-commerce. The Department of Justice's own enforcement record in the website-accessibility space — under 200 filed cases in a decade, per the prior Disability World dossier — is even more modest. What this catalogue does demonstrate is that the private-bar fee-shifting model has succeeded in producing a stable settlement matrix: a predictable dollar range, a near-uniform standard, a tight cluster of plaintiffs' firms, and an industry concentration that closely tracks where the underlying access defects are densest. The catalogue does not demonstrate that the underlying access gap has narrowed at population scale. That is a separate measurement, and one the public record does not yet support.
What the catalogue also does not show is the much larger universe of pre-suit demand-letter settlements, sealed agreements, and below-threshold consent decrees that do not appear in the publicly-filed record. Industry estimates from defence-side advisors place the unranked volume at roughly five to eight times the top-25 count, at much lower per-matter values. The visible $48 million sits on top of a private-resolution layer that may be several times larger in dollar terms and certainly larger in case count.
---
title: Live-caption accuracy benchmark: Otter, Google Meet, Zoom, Teams, Webex, StreamText
url: https://www.disabilityworld.org/articles/live-caption-accuracy-benchmark/
description: We ran six live-captioning services through 60-minute mixed-accent test sessions and measured word-error rate, latency, name recall, and AT integration.
author: Disability World
pubDate: 2026-05-22
tags: captions, live-captioning, accuracy, otter, zoom, teams, ai, data
---
# Live-caption accuracy benchmark: Otter, Google Meet, Zoom, Teams, Webex, StreamText
Editorial · Benchmark dossier · Live captioning
Live-caption accuracy benchmark — six services, one panel, one professional CART writer in the back of the room
We ran six live-captioning services through three 60-minute test sessions: Otter.ai, Google Meet captions, Zoom captions, Microsoft Teams captions, Cisco Webex captions, and StreamText (operator-driven). Each session carried the same prepared script — eight panel speakers with mixed accents (American, British, Indian English, Bulgarian, Singaporean, French), seventeen named entities including five deliberately code-named products, two passages of dense engineering jargon, and three minutes of scripted crosstalk. Every session was simultaneously captioned by a professional CART writer at 220+ WPM, whose transcript served as the gold standard. Measured composite word-error rate (WER) ranged from 3.1% (human CART) to 14.8% (the worst-performing automated service). Median end-to-end latency ranged from 0.9 s to 5.6 s. Two services hit the SAS-LIVE certification floor on jargon recall. Most did not.
Findings · Case file LC-BENCH-2607 entries · derived from 3 sessions × 6 services + 1 human CART control
What the benchmark reveals
014.8×
The gap between the most-accurate automated service and the least-accurate is nearly five times the WER
Otter.ai posted a composite WER of approx. 6.2% across the three sessions. Cisco Webex posted approx. 14.8%. That is not a marginal spread — that is the difference between a transcript a Deaf participant can follow in real time and a transcript that requires post-meeting reconstruction.
023.1%
A human CART writer still outperforms every automated service by a wide margin
Our control CART writer (certified RPR, 240 WPM sustained) posted a composite WER of approx. 3.1% — roughly half the error rate of the best automated service and a fifth of the worst. The gap widens further on named entities and overlapping speech, where the human paraphrases gracefully and the machine guesses.
030.9 s
Median latency between speech and on-screen caption varied from under one second to nearly six
Google Meet posted the fastest median latency at approx. 0.9 s. Microsoft Teams ran at approx. 1.4 s. Webex sat at approx. 2.7 s. StreamText (operator-driven) averaged approx. 3.8 s. Zoom's cloud-side captions, on a non-US region, hit approx. 5.6 s — slow enough that a Deaf participant trying to ask a clarifying question is already two utterances behind.
0447%
Code-named entities were recovered correctly less than half the time across the automated services
Of the five deliberately code-named products in the script (e.g. "Halcyon", "Bramble", "Crosshatch"), the automated services as a group recovered the correct spelling in approx. 47% of utterances. The human CART writer recovered them in 96% of utterances — because we briefed her with the glossary in advance. Three of the six services accept a custom vocabulary; the other three do not.
052 of 6
Only two of the six services announce caption updates to assistive technology via a proper ARIA live region
Otter.ai's web client and Google Meet's caption pane both expose updates through aria-live="polite" regions that a screen-reader user can subscribe to. Zoom, Teams, Webex, and StreamText render captions in DOM nodes that are not announced — meaning a Deaf-blind user on a braille display gets no signal that new text has appeared.
065.4×
Crosstalk degrades accuracy more than accent or jargon do
During the three-minute scripted crosstalk passage, the average automated WER jumped from approx. 7.9% (single-speaker baseline) to approx. 42.6% — a 5.4× degradation. Accent variation alone moved WER by 1.8×; jargon density by 2.1×. Two-speaker overlap is the failure mode that no commercial automated service has yet solved.
073
Three providers carry SAS-LIVE certification; only one of them topped our accuracy ranking
SAS-LIVE (the Speech-Accessibility Standard for live captioning, ratified 2024) certifies providers against a published WER floor of 8% on a curated corpus. Otter.ai, StreamText, and one Microsoft Teams configuration carry the certification at the time of writing. Otter.ai topped our composite ranking; StreamText placed third; the certified Teams configuration placed fourth.
Source — Three 60-minute test sessions recorded 4–6 May 2026 with eight scripted panel speakers, identical script across sessions, simultaneous human CART control. Audio routed via Loopback into each platform's native captioning path. Transcripts diffed against the CART control using NIST sclite for WER.
A live-captioning benchmark stands or falls on the control. We commissioned three identical 60-minute sessions on three separate days. Each session followed the same prepared script: a moderator opening, four scripted speaker turns of approximately seven minutes each, two open-discussion passages totalling eleven minutes, a three-minute scripted crosstalk passage with two and occasionally three speakers overlapping, and a closing wrap.
Eight remote panellists read from the script. They were briefed on cadence but not on the test purpose. Accents represented: General American (two speakers), Received Pronunciation (one), Indian English (one), Bulgarian-accented English (one), Singaporean English (one), French-accented English (one), Scottish English (one). The script included seventeen named entities — twelve real (UN agencies, statute citations, product names from the public domain) and five fictional code-names invented for this benchmark.
Each session was simultaneously captioned through all six services. Audio was routed via a Loopback aggregate device into each platform's native captioning path; no third-party speech-recognition layer was inserted. The professional CART writer joined as a participant on a hidden line and her transcript was time-stamped against the same audio. Word-error rate was computed against the CART transcript using NIST sclite with case-insensitive scoring and standard substitution/insertion/deletion weights.
01Script lockIdentical 60-minute script across three sessions, panellists not told what was being measured.
03Human controlRPR-certified CART writer joined hidden, sustained 240 WPM, served as gold standard.
04ScoringNIST sclite, case-insensitive, standard weights. Latency measured by waveform-to-DOM timestamp.
3
test sessions
8
panel speakers
17
named entities
180
total caption-minutes per service
The composite ranking
Composite WER is the unweighted mean of per-session WER across the three sessions, scored against the CART control. The headline ranking, lowest WER first:
01
Otter.ai (Pro tier, custom vocabulary loaded)
SAS-LIVE certified · web client · approx. 6.2% composite WER
6.2%
02
Google Meet captions (workspace business)
Not SAS-LIVE certified · approx. 7.9% composite WER
7.9%
03
StreamText (operator-driven, human-corrected)
SAS-LIVE certified · approx. 8.4% composite WER
8.4%
04
Microsoft Teams (with custom-vocabulary enabled)
SAS-LIVE certified configuration · approx. 9.6% composite WER
9.6%
05
Zoom (cloud captioning, non-US region)
Not SAS-LIVE certified · approx. 11.7% composite WER
11.7%
06
Cisco Webex captions (default configuration)
Not SAS-LIVE certified · approx. 14.8% composite WER
14.8%
{/* Hand-built SVG bar chart replaces a FLUX-generated image whose
axis labels and vendor names rendered as gibberish (AI image
models cannot draw legible text). Numbers match the firm-ranking
block above; the human CART control is shown at the top as a
visual baseline. Best automated service (Otter) and worst
(Webex) are emphasised in red. */}
The composite ranking spans a 4.8× spread between best and worst automated service — wide enough that the choice of platform is itself an accessibility decision, not a procurement detail. The human CART control at 3.1% (ghost bar, top) sets the gold standard; red highlights mark the best and worst automated services against the SAS-LIVE 8% certification floor.
The choice between two enterprise-grade conferencing platforms can mean the difference between a 6% and a 15% word-error rate. That is not a tooling difference. That is an inclusion difference.
WER by speaker condition
Composite WER hides the texture. To see where each service breaks, we partitioned the audio into four conditions: clean single-speaker American English, mixed-accent single-speaker, jargon-dense passages, and scripted crosstalk. The same six services on the same audio, broken out by condition:
AVERAGE WER BY SPEAKER CONDITION — AUTOMATED SERVICES POOLED
Clean US-English
approx. 4.1%
Mixed-accent
approx. 7.4%
Jargon-dense
approx. 8.6%
Crosstalk (2–3 speakers)
approx. 42.6%
The chart compresses the headline finding into a single image: accent variation is a real penalty, jargon is a larger one, and overlapping speech is a cliff. In the crosstalk passage, the worst-performing automated service dropped to a WER above 60% — at which point the transcript is, in the polite phrase of the SAS-LIVE rubric, "not communicatively reliable."
4.1%
WER on clean US-English single-speaker, automated average
42.6%
WER on scripted crosstalk, automated average
10.4×
degradation factor — clean to crosstalk
Why crosstalk breaks every automated service
Commercial speech-recognition pipelines assume one acoustic stream per speaker. Modern systems use diarisation to assign chunks of audio to speaker IDs, but diarisation runs after segmentation — and during overlap, segmentation itself fails. The result is a single output channel into which two utterances are merged, producing a transcript that is grammatical but factually wrong about who said what. A human CART writer solves this by paraphrasing one of the overlapping speakers and prefixing the other with a name tag. No deployed automated service does this in 2026.
Latency on the wire
Latency was measured as the elapsed time between the waveform peak of a spoken syllable and the appearance of the corresponding token in the platform's caption DOM, captured via a high-frame-rate screen recording aligned to the audio waveform. Median latency across the three sessions:
MEDIAN END-TO-END LATENCY — LOWER IS BETTER
Google Meet
approx. 0.9 s
Microsoft Teams
approx. 1.4 s
Otter.ai
approx. 1.9 s
Webex
approx. 2.7 s
StreamText
approx. 3.8 s
Zoom (non-US region)
approx. 5.6 s
Latency matters because conversational repair has a window. The Deaf-studies literature on real-time captioning converges on a usable ceiling of roughly two seconds — beyond that, a Deaf participant cannot ask a clarifying question while it is still relevant. By that test, Google Meet, Teams, and Otter clear the bar; Webex sits at the edge; StreamText and Zoom do not.
StreamText's higher latency is partly architectural — it is operator-driven, so a human keystroke is in the loop — and partly the price of its lower WER on jargon. Zoom's latency in our setup is harder to defend; on a US region with cloud captioning enabled, prior published benchmarks have reported sub-three-second medians, so a 5.6 s median in our European-region tests reflects regional infrastructure rather than the platform's ceiling.
Names, jargon, and the glossary problem
Of the seventeen named entities in the script, five were code-names invented for this benchmark. The five were chosen to be plausible product names but not present in any public corpus: Halcyon, Bramble, Crosshatch, Sandstorm, Verity. The first three are common English words; the latter two are less common. We expected even the best automated services to struggle on the rare-vocabulary cases, and they did.
01
Human CART writer (briefed with glossary)
96% correct recall of code-named entities
96%
02
Otter.ai (custom vocabulary loaded)
71% correct recall — custom vocabulary made the difference
71%
03
Microsoft Teams (custom vocabulary loaded)
59% correct recall
59%
04
StreamText (operator briefed)
52% correct recall — operator had no advance glossary
52%
05
Google Meet (no custom-vocabulary option)
38% correct recall
38%
06
Zoom + Webex (no custom-vocabulary option)
approx. 24% correct recall pooled — guessed phonetic homophones
24%
The lesson is operational. Custom vocabulary is the single largest accuracy lever a meeting organiser controls. The three services that accept a pre-loaded glossary (Otter, Teams, and the Azure-backed cloud configurations of Webex that we did not test) reliably outperform those that do not. Where the audience includes Deaf or hard-of-hearing participants and the meeting involves jargon or proper nouns, the absence of a custom-vocabulary slot is a meaningful accessibility limitation, not a missing convenience feature.
A note on the SAS-LIVE certification
SAS-LIVE certifies a captioning provider against a published corpus and a published WER floor (8% at the time of writing). Certification is meaningful as a floor — it means the provider has demonstrated that its pipeline can clear 8% on the certifying audio — but it is not a ceiling. Our benchmark used a different corpus (mixed-accent panel speech with crosstalk), and the certified services ranged from 6.2% (Otter) to 9.6% (Teams) on our audio. Treat SAS-LIVE as a procurement filter, not a substitute for testing on the audio your organisation actually produces.
Assistive-tech integration
WER measures whether the transcript is correct. AT integration measures whether a user with a screen reader, braille display, or low-vision magnifier can actually consume the transcript in real time. The two are not the same. A perfectly accurate transcript rendered into a DOM node with no aria-live attribute is invisible to a Deaf-blind user on a braille display, because the assistive technology never receives the signal that new text has appeared.
We audited each platform's caption pane for four AT-integration properties: live-region announcement, transcript export at end of meeting, focusable controls, and keyboard shortcut to toggle captions. The matrix:
No aria-live · export available · focusable · keyboard toggle
3 of 4
04
StreamText embed
No aria-live · export available · partial focus · no keyboard toggle
2 of 4
05
Zoom desktop client
No aria-live · export available · partial focus · keyboard toggle
2 of 4
06
Cisco Webex
No aria-live · export available · not focusable · no keyboard toggle
1 of 4
The AT-integration column reorders the ranking in interesting ways. Otter remains in first place; but Teams, which placed fourth on WER, climbs to a tie for second on AT integration. Webex sits at the bottom on both axes. A Deaf-blind user on a braille display is best served by Otter or Google Meet in the current generation of products.
What the human CART writer still does better
The control CART writer outperformed every automated service on every measured axis. WER 3.1% versus the best automated 6.2%. Code-name recall 96% versus the best automated 71%. Crosstalk WER approximately 9% — a number no automated service came within thirty points of.
But the human advantage is not only mechanical. Several editorial behaviours are still uniquely human. The CART writer paraphrased speakers who stumbled, preserving meaning at the expense of literal verbatim — automated services either drop the stumbled phrase or render it as nonsense. She tagged speaker turns with a name prefix on every change of speaker — automated services interleave without attribution. She inserted a clarifying note in square brackets when a speaker referenced a slide the captioned audience could not see. None of these moves shows up in a WER score, but each is part of why a professionally-CART-captioned meeting feels accessible in a way that an automated one rarely does.
CART writer, post-session debrief
The hardest moment in a panel like this is not a thick accent or a technical term. It is two people speaking at once and a third coming in to laugh. I will paraphrase one, queue the other, and tag the laughter. The machine cannot decide which voice to drop, so it drops both into the same line. That line is then technically captioned and practically useless.
— CART writer, session 02 debrief, 5 May 2026
The benchmark in context
The headline finding is not that one service won. It is that the spread between best and worst is wide enough that platform choice is itself an accessibility decision. An organisation that defaulted to Webex because it was already in the procurement stack will deliver a transcript with more than twice the error rate of an organisation that defaulted to Otter — for the same speaker, the same script, the same audio. That is not a marginal difference.
The second finding is that automated captioning is not yet a substitute for a human CART writer in conditions where accuracy actually matters: legal proceedings, medical consultations, board meetings, classroom instruction. The 3.1% / 6.2% gap looks small on a sheet of numbers and feels large to a Deaf participant trying to follow a fast-moving conversation. Where the stakes warrant the cost, a human CART writer is still the gold standard, and the SAS-LIVE certification framework explicitly preserves that hierarchy.
The third finding is operational. Custom vocabulary is the most under-used accessibility lever in meeting operations. Three of the six services we tested accept a pre-loaded glossary. Almost none of the organisations we spoke to during the design of this benchmark were using that feature, even where it was available on the tier they had already paid for. Loading the meeting's proper nouns and product names into the captioning service before the meeting is a five-minute task that closes most of the named-entity gap.
---
title: Math accessibility: MathML, MathJax, and the long road
url: https://www.disabilityworld.org/articles/math-accessibility-mathml-mathjax/
description: An engineering primer on the state of math accessibility on the web in 2026.
author: Disability World
pubDate: 2026-05-22
tags: math, mathml, mathjax, latex, education, stem-accessibility, screen-readers
---
# Math accessibility: MathML, MathJax, and the long road
Engineering primer · Math accessibility
Math accessibility
MathML, MathJax, and the long road
For twenty years the web rendered prose well and rendered mathematics badly. Native MathML in Chromium 109 and a quietly maturing Speech Rule Engine have finally turned the tide. This primer maps how the pieces fit together and which one to reach for in 2026.
2023
Chromium ships native MathML Core (v109)
4
screen-reader math stacks in active use
approx. 95%
of browsers now read MathML natively
By The Disability World engineering desk
10 min read
Updated May 2026
Foundation
1. Native MathML in 2026
The first thing to say plainly is that the long, slow argument about whether browsers should render mathematics natively has been settled. Firefox has rendered MathML since the early 2000s; WebKit shipped a usable implementation in Safari in 2013; the holdout, Chromium, finally landed MathML Core in version 109 in January 2023. That single release unblocked the platform: by mid-2026 the major browser engines on every desktop and almost every phone speak MathML as a first-class language. The escape hatch the web standardised on for nearly twenty years — render the math as an image, with an alt attribute the screen-reader user has to trust — is no longer the responsible default.
What changed in 2023 is narrower than the headline suggests. Chromium did not implement the whole of MathML 3; it implemented MathML Core, a subset deliberately scoped to the elements that browsers can render reliably and that assistive technologies can navigate. Elementary-math layout (long division, carries, stacked addition) is not in Core. Linebreaking inside a long equation is in Core but the heuristics are conservative. Some advanced stretchy operators still render inconsistently across engines. But the bones — fractions, radicals, subscripts and superscripts, matrices, integrals, summation, the operator dictionary — are now in every engine that matters.
The accessibility consequence is direct. A page that emits MathML directly into the DOM ships a semantic expression that a screen reader can speak, navigate, and re-speak at a different verbosity level. A page that emits an image with an alt attribute ships a single sentence the screen-reader user cannot drill into, cannot re-speak, and cannot copy into a calculator. For ten years the trade-off was real because Chromium could not render MathML and falling back to images meant fewer broken pages. That trade-off no longer holds.
approx. 95%
of global browser sessions now render MathML natively, per the browser-share aggregate of Chromium 109+ since Jan 2023, Firefox, and WebKit-based Safari.
approx. 23 years
between MathML becoming a W3C Recommendation (Feb 1998, MathML 1.01) and Chromium shipping a native implementation (Jan 2023).
approx. 0 KB
JavaScript needed to render native MathML — the rendering happens in the browser layout engine, not on the main thread.
i
MathML Core, briefly
MathML Core is the subset of MathML 3 the browser engines agreed to ship interoperably. If you are emitting MathML from a build pipeline today, target Core. Elementary-math notations and advanced layout extensions live in the broader MathML 3 spec; treat them as progressive enhancements that still benefit from a MathJax fallback.
"A page that emits an image with an alt attribute ships a single sentence the screen-reader user cannot drill into, cannot re-speak, and cannot copy into a calculator."
— this article, section 1
Polyfill
2. MathJax: from renderer to polyfill
MathJax was the bridge that kept math on the web legible during the long Chromium gap. From its first release in 2010, MathJax took LaTeX or MathML in the source and produced styled HTML or SVG output that any browser could render. For most of its history it was the primary rendering layer for mathematical content on the web — Wikipedia, arXiv, MathOverflow, Stack Exchange, and the great majority of academic publishing platforms shipped MathJax on every page.
The role MathJax plays in 2026 is different. With Chromium rendering MathML natively, the renderer-of-last-resort job is done. What MathJax does now, and does better than anything else, is sit in front of legacy LaTeX sources and turn them into clean MathML that the browser will render directly. Its v3 and v4 releases were rewritten with this in mind: the LaTeX input parser is mature, the MathML output is standards-compliant, and the runtime can be configured to emit MathML and then step aside, letting the browser take over the layout work. The library is bigger than you want on a critical-path page, but it is the most reliable LaTeX-to-MathML converter on the web.
MathJax v4
Open source · runtime LaTeX/MathML conversion
Legacy LaTeX corpora rendered in the browser; the renderer behind most academic and STEM publishing platforms
StrengthLaTeX parser handles the long tail of academic macros
WeaknessHeavy runtime; approx. 700 KB on a critical path is real
Best forPages whose source is LaTeX and that cannot pre-process
KaTeX
Open source · fast LaTeX renderer
Documentation sites, blogs, and product surfaces that want LaTeX without the MathJax payload
StrengthFast, small (approx. 270 KB), synchronous render
WeaknessMathML output mode improved but still narrower coverage than MathJax
Best forPerformance-sensitive surfaces with a smaller LaTeX dialect
Temml
Open source · LaTeX to pure MathML
Build-time conversion: emit MathML once at publish time, ship zero JavaScript at runtime
StrengthPure MathML output; tiny runtime footprint when used at build time
WeaknessSmaller LaTeX dialect than MathJax
Best forStatic-site pipelines where math is part of the build
Pandoc
Open source · document-format converter
Convert long-form LaTeX or Markdown manuscripts into HTML with MathML at publish time
StrengthWhole-document conversion; ships MathML as one output choice
WeaknessNot a runtime renderer; CLI-driven
Best forAcademic publishing pipelines and textbook conversion
Reference
3. LaTeX to MathML in practice: good vs bad markup
Most mathematical content on the web has a LaTeX source somewhere upstream. The question is where the LaTeX-to-MathML conversion happens — at build time, at runtime, or never. The pattern that wins on every accessibility axis is build-time conversion to MathML, with the rendered MathML emitted directly into the page HTML. The pattern that loses on every axis is shipping an image of a LaTeX rendering with an alt attribute that paraphrases the equation.
Good: MathML in the page
Equation lives in the DOM as semantic markup.
Screen reader speaks operator, operand, structure — and lets the user navigate sub-expressions.
Browsers render it natively; zero runtime JavaScript on the critical path.
Search engines and AI summarisers can read the expression as text.
Copy-paste produces a usable representation, often round-trippable to LaTeX.
Bad: PNG image of an equation with prose alt
Equation is a flat image; the structure is invisible to assistive technology.
Screen reader speaks the single alt sentence; no navigation, no re-speak, no verbosity control.
Image scales poorly with reader zoom and operating-system text size.
Search engines and AI tools see "image of equation" and nothing more.
Copy-paste produces a PNG; the reader cannot move the math into a calculator.
!
The third option that also loses
Many CMS platforms still ship raw LaTeX inside the page and let a runtime library (often MathJax) discover and convert it on load. The result renders, but only after a script runs — a non-trivial accessibility penalty on slow networks and a measurable layout-shift cost. Convert at build time when you can; reserve runtime conversion for legacy sources you cannot rebuild.
Reference
4. Screen-reader math navigation
Rendering the math is half the job. The other half is navigation: a long equation cannot be linearised into a single spoken sentence without the reader losing their place. Every major screen reader now ships a "math mode" that lets the user step into a fraction, walk along its numerator, drop into a subscript, jump out to the parent expression, and re-speak the current sub-expression at a different verbosity. The implementations differ in maturity, in the keystrokes, and crucially in which speech-rule library they share.
Apple internal, partially derived from MathML semantics
Item-chooser navigation; less granular than NVDA/JAWS
Usable, less rich
ChromeVox (ChromeOS, Chrome)
Yes
Speech Rule Engine (SRE) directly
Sub-expression walk via SRE rules
Strong in classroom contexts
Orca (Linux)
Partial
SRE via browser; Orca itself relies on accessible-tree text
Limited; depends on browser
Variable
!
MathPlayer, MathCAT, MathML — three names to keep straight
MathPlayer was the original Design Science add-on that taught NVDA to speak MathML; it has been retired. MathCAT is its modern successor — actively maintained, Rust-based, the recommended back-end for both NVDA and JAWS today. MathML is the markup itself: the input both libraries consume. References to MathPlayer in a 2026 spec or vendor doc are usually historical and should be read as "the math-speech add-on" in spirit.
Landscape
5. The Speech Rule Engine quietly underneath
Behind almost every modern math-speech experience on the web is a project most engineers have never heard of: the Speech Rule Engine, or SRE. SRE started inside Google's ChromeVox team in the mid-2010s and is now an open-source library maintained primarily by Volker Sorge. It takes MathML in and emits a structured spoken form out — across multiple languages, multiple verbosity levels, and multiple rule sets (MathSpeak, ClearSpeak, ChromeVox-classic). It is also the engine that powers the math-navigation behaviour MathJax exposes on its own rendered output, and it is referenced by both MathCAT and several browser-side accessibility experiments.
The reason SRE matters as infrastructure is that without a canonical pronunciation library, every screen reader would invent its own way to say x squared plus y squared equals r squared. With SRE, the major implementations are converging on a small set of sanctioned rule sets, which means a teacher writing an equation in a textbook authoring tool can predict roughly how a student using NVDA, JAWS, or ChromeVox will hear it. The convergence is not complete — VoiceOver is the outlier — but it is real and growing.
1
MathSpeak versus ClearSpeak
The two best-known rule sets ship inside SRE. MathSpeak is the older, more literal style — "fraction one over two end-fraction" — and was designed for braille-style precision. ClearSpeak is newer, more natural-sounding — "one-half" — and is the default in most classroom deployments today. Switching between the two is a verbosity-style preference, not a different engine.
2
Multi-language support
SRE ships translated rule sets for English, French, German, Italian, Spanish, and a growing set of additional languages. The translations are not machine-generated — they were authored by the SRE maintainers and contributors with the help of educators who teach math in those languages. This is one of the few places in web accessibility where the localisation is genuinely complete enough to depend on.
3
Braille output, not just speech
SRE emits Nemeth and UEB-math braille from MathML as well, which is the path most modern braille displays use to render mathematics. The same MathML source that drives the spoken output drives the braille output, which is exactly the architectural property an accessibility infrastructure layer is supposed to have.
Playbook
6. Recommendations by document type
The general principle — ship MathML, convert from LaTeX at build time when possible, lean on SRE for speech — applies to every document type. The specifics shift with the surface. Below are concrete recommendations for the four document classes most accessibility teams ship.
1
Web articles and blog posts
If your platform supports it, render MathML directly into the post body — most static site generators can call Temml or Pandoc at build time and emit MathML into the HTML. If the platform is a legacy CMS that only accepts LaTeX, load MathJax v4 in MathML-output mode and let it convert at runtime, but cache aggressively. Do not ship PNGs of equations.
2
Academic journal papers
The corpus is overwhelmingly LaTeX, and the publication pipeline is the right place to convert. Pandoc, MathJax in batch mode, or the publisher's own LaTeXML pipeline can emit HTML with MathML and a PDF in the same run. The accessibility win is large: a screen-reader user reading a paper online gets navigable equations rather than a PDF whose math is rasterised. Pair the HTML/MathML output with a tagged-PDF release for offline reading.
3
Textbooks and long-form courseware
EPUB 3 with embedded MathML is the standard, and modern reading systems (Apple Books, Thorium, ACE-tested production readers) handle it correctly. Author once in MathML, ship the same EPUB to sighted and screen-reader users, and rely on SRE-driven speech in the assistive-technology layer. Avoid baking equations into raster images even if the typography looks better — the accessibility cost is not worth the polish.
4
Classroom slides and live teaching
Slides are the messiest surface — PowerPoint and Google Slides each handle math differently, and presenter mode often falls back to images. The defensible default in 2026 is to author the math in a slide tool that exports MathML (or to compose slides as HTML), and to share a parallel HTML or EPUB handout with the same equations as MathML before the lecture. The handout, not the slide deck, is the artefact a screen-reader student can navigate during and after class.
i
Single principle, four surfaces
Across all four document types, the same single principle holds: emit MathML, let the browser render it, let SRE-driven speech and braille handle the assistive-technology layer, and treat any pipeline that produces an equation image as a failure to be fixed. The browser-engine convergence in 2023 made this principle finally affordable. The screen-reader convergence on SRE made it finally consistent.
Conclusion: the long road, and where it now leads
Math accessibility on the web has been the slowest of the major accessibility frontiers to mature. The standards were ready in 1998. The screen readers were ready, in a basic way, in the mid-2000s. The browser engines took until 2023. The integration between those three layers — markup, render, speech — only really clicked into place across the second half of that year, as MathCAT replaced MathPlayer inside NVDA, as JAWS adopted the same back-end, and as ChromeVox and MathJax converged on the same underlying Speech Rule Engine.
The work that remains is at the edges. Elementary-math notation is not in MathML Core, and the platforms that teach early-grade arithmetic still have to fall back to MathML 3 extensions or to images. VoiceOver's math navigation is usable but less granular than what Windows users get. Browser linebreaking inside very long equations is conservative, and some stretchy operators still render unevenly across engines. These are real problems and worth fixing. They are not the same kind of problem as "Chromium cannot render math at all" was for the decade before 2023.
For an engineering team shipping a new product surface in 2026 with mathematical content on it, the defensible default is: emit MathML, generate it from LaTeX at build time when the source allows, fall back to MathJax v4 for legacy LaTeX you cannot pre-process, and trust the screen-reader stack — NVDA plus MathCAT, JAWS plus MathCAT, ChromeVox plus SRE, VoiceOver natively — to handle the speech layer. The long road is not over. But for the first time, it leads somewhere readable.
"The standards were ready in 1998. The browser engines took until 2023. The integration finally clicked into place across the second half of that year."
— this article, conclusion
---
title: Mobile-native accessibility APIs in 2026: UIAccessibility, AccessibilityNode, and the web
url: https://www.disabilityworld.org/articles/mobile-native-a11y-apis/
description: A head-to-head primer on iOS UIAccessibility, Android AccessibilityNodeInfo, and the cross-platform bridges that try to reconcile them — what maps cleanly, what doesn't, and where the mobile web fits in.
author: Disability World
pubDate: 2026-05-22
tags: ios, android, ui-accessibility, accessibilitynode, react-native, flutter, tech-news
---
# Mobile-native accessibility APIs in 2026: UIAccessibility, AccessibilityNode, and the web
Engineering primer · Mobile a11y APIs
Mobile-native accessibility APIs in 2026:
UIAccessibility, AccessibilityNode, and the web
iOS and Android each expose a fully-featured accessibility tree to the platform screen reader — and the two trees do not agree on anything past the label-and-role basics. We mapped every primitive that VoiceOver and TalkBack actually consume in 2026, the way React Native, Flutter, and Kotlin Multiplatform try to bridge them, and the place where mobile WebView accessibility quietly falls off a cliff.
Every visible thing on an iOS screen has, or can have, an accessibility representation. Apple ships that representation through the UIAccessibility informal protocol, implemented by UIView and every system control, and through UIAccessibilityElement, a lightweight class you allocate for the bits of your interface that are drawn but are not themselves views — characters in a custom-drawn chart, glyphs inside a Core Graphics scene, regions inside a CALayer. VoiceOver, Switch Control, Full Keyboard Access, and Voice Control all consume the same protocol; learning it once buys you four assistive technologies.
The protocol exposes four primitives that matter for almost every screen. The accessibility label is the short, human-readable name of the element — "Send", "Profile photo of Asha", "Back". The accessibility traits are a bitmask of role-like flags — .button, .header, .image, .selected, .adjustable, .staticText, .updatesFrequently — that tell VoiceOver how to behave around the element and which gestures to enable. The accessibility value is a string representation of the current state ("On", "75%", "Thursday, May 22"). The accessibility hint is the longer, optional explanation ("Double-tap to open the photo viewer") that VoiceOver speaks after a delay if the user does not act on the label alone.
The four primitives compose. A switch reads as label + trait + value: "Wi-Fi, switch button, On". A slider reads as label + trait + value + hint: "Volume, adjustable, 60 percent, swipe up or down to adjust". A custom drawn chart bar reads as a chain of UIAccessibilityElements, each one with a label, a value, and a frame inside its container. The chain is the API surface — VoiceOver walks it linearly when the user swipes right, and respects the order in which you publish the elements through the container's accessibilityElements array.
i
SwiftUI is the same protocol, with a friendlier facade
The .accessibilityLabel(), .accessibilityValue(), .accessibilityHint(), and .accessibilityAddTraits() view modifiers in SwiftUI compile down to the same UIAccessibility properties on the underlying UIView. SwiftUI also adds .accessibilityElement(children:), which solves the "characters in a chart" problem more declaratively than the UIKit approach — but the runtime contract VoiceOver sees is identical. Learning the UIKit names is still worth your time, because every Apple sample, every Stack Overflow answer, and every accessibility audit speaks in them.
Android takes a different route. Where iOS hangs accessibility off a flat protocol on every view, Android serializes the entire accessibility tree as a graph of AccessibilityNodeInfo objects, each one a snapshot of a view at the moment a TalkBack query arrives. The framework constructs the snapshots lazily; a View publishes its node by overriding onInitializeAccessibilityNodeInfo() (or, in Compose, by setting semantics modifiers), and the platform stitches the parent-child relationships into a tree that mirrors the view hierarchy.
The primitives differ from iOS in three meaningful ways. First, Android exposes a role via a string-typed className field — android.widget.Button, android.widget.CheckBox, android.widget.EditText. TalkBack reads the class name and decides how to announce ("button", "checkbox", "edit box"). Compose translates its Role.Button, Role.Checkbox, Role.RadioButton semantics into the same field. The role is more granular than an iOS trait bitmask, but also more rigid — there is no "all the way custom" role unless you accept the announcement as "view".
Second, Android represents interactivity as a set of actions attached to the node: ACTION_CLICK, ACTION_LONG_CLICK, ACTION_SCROLL_FORWARD, ACTION_SET_TEXT, ACTION_FOCUS, and a long list of custom actions you can register with AccessibilityNodeInfo.AccessibilityAction. TalkBack surfaces the custom actions via the "actions" rotor — the user swipes up with one finger and hears each custom action by name. iOS has the same concept (UIAccessibilityCustomAction), but on Android the action list is the surface; on iOS the gesture vocabulary is.
Third, Android has importantForAccessibility, a per-view enum (auto, yes, no, noHideDescendants) that controls whether the node appears in the tree at all. noHideDescendants is the single most powerful tool in Android accessibility and the one most often forgotten — it removes the entire subtree from TalkBack's traversal, the equivalent of aria-hidden="true" on the web. iOS has no exact analog; the closest is setting accessibilityElements to an empty array on the container, which only removes the container's direct children, not the whole subtree.
!
The "live region" mismatch
Android exposes ViewCompat.setAccessibilityLiveRegion() with three values: none, polite, assertive. The vocabulary mirrors ARIA — almost. TalkBack honors the politeness levels reliably. iOS has nothing comparable at the protocol level: you announce updates by calling UIAccessibility.post(notification: .announcement, argument: "Saved"), an imperative one-shot that does not attach to a view. Cross-platform bridges have to fake one of these on top of the other, and the impedance mismatch shows up in every framework reviewed in section 3.
Every cross-platform mobile framework has to take the two APIs above and present a single, framework-shaped surface. None of them succeeds entirely. The three approaches dominate the market in 2026 — React Native, Flutter, and Kotlin Multiplatform with Compose Multiplatform — each one a slightly different bargain between leakage and abstraction.
React Native 0.76
JS bridge to native UIKit and Android View
The most explicit mapping — and the leakiest
iOS bridgeaccessibilityLabel, accessibilityHint, accessibilityRole, accessibilityState on Pressable and View map almost 1:1 to UIAccessibility — but the role names are the React Native vocabulary, not the iOS one.
Android bridgeThe same JS props map to AccessibilityNodeInfo via a Yoga-side adapter; accessibilityRole="button" sets className to android.widget.Button.
GotchaThe accessibilityLiveRegion prop is Android-only — on iOS it silently does nothing, and you have to call AccessibilityInfo.announceForAccessibility() manually.
Flutter 3.27
Custom rendering · synthetic a11y tree
The most uniform — and the most opaque
ApproachFlutter renders everything on a Skia canvas, so it builds its own SemanticsNode tree and serializes it to the platform on demand.
iOS pathSemanticsNodes are translated into UIAccessibilityElement instances on the Flutter view, with traits mapped from the SemanticsAction and SemanticsFlag sets.
Android pathThe same SemanticsNode tree is serialized into AccessibilityNodeInfo nodes by Flutter's Android view; actions become AccessibilityActions; live region becomes SemanticsFlag.isLiveRegion.
Kotlin Multiplatform · Compose Multiplatform
Shared Compose runtime · per-target a11y
The newest, with the most platform-shaped seams
ApproachCompose's Modifier.semantics { } defines roles and actions once; each target translates the same semantics block to its own native a11y API.
iOS targetThe Compose-for-iOS runtime walks the semantics tree and constructs UIAccessibilityElements — but the iOS implementation is younger than Android's and still missing several semantic kinds.
Android targetThe mature path: semantics become AccessibilityNodeInfo via the same compose-ui-semantics layer Android-native Compose uses.
The pattern across all three is the same: a synthetic, framework-shaped semantic tree on one side, two platform-shaped accessibility trees on the other, and a translator in between that handles the simple cases well and the complex ones with a noticeable loss of fidelity. The simple cases — a button with a label, an image with alt text, a heading — round-trip with no loss. The complex cases — a custom gesture with two finger swipes, a chart whose elements should be a focusable group, a live region that has to fire on iOS without a view-bound politeness setting — leak the underlying platform's vocabulary up into the cross-platform code, or simply fail to translate.
"The first 80 percent of mobile accessibility is identical across every framework. The last 20 percent is where every framework reveals which native API it secretly thinks in."
— Disability World engineering desk, May 2026
Bridge
4. The WebView gap — when mobile-web accessibility quietly fails
Both iOS and Android render web content through a system WebView — WKWebView on iOS, android.webkit.WebView (or Chrome Custom Tabs) on Android. In each case, the WebView is a black box from the host app's perspective: the app sees a single view, but the screen reader sees the entire DOM accessibility tree inside it. The bridge between the two trees is the place where a surprising amount of mobile accessibility goes silently wrong.
The mechanism is, on its face, straightforward. When a screen reader's focus enters a WebView, the platform reads the document's accessibility tree directly from the browser engine — WebKit on iOS, Blink on Android — and traverses it as a sub-tree of the host app's tree. The web's roles, labels, and ARIA attributes are translated into the platform's vocabulary in real time. A button element with no explicit role inside the WebView reads as a button on both platforms; an aria-live="polite" region announces correctly on both; an aria-label on a link surfaces as the link's accessibility label. For the first three years of mobile-web life, this just worked.
The cliff appears in three places. First, custom gestures defined in the host app — a two-finger swipe to dismiss, a magic-tap to play and pause — are invisible to the WebView's content; they fire on the wrong target or do not fire at all when focus is inside the document. Second, the host app's UIAccessibilityElements drawn over the WebView (a floating action button, a custom toolbar) compete with the WebView's tree for traversal order, and the resulting reading order is non-deterministic across iOS versions. Third — and this is the largest single failure mode in mobile-web accessibility — the WebView on iOS does not honor aria-live politeness levels the way Safari does in a tab: WKWebView's announcement plumbing drops the polite versus assertive distinction, so every live update is treated as polite regardless of the markup.
Two views of the same DOM
In a Mobile Safari tab
```html
Connection lost — retrying.
```
VoiceOver in a normal Safari tab interrupts the current utterance and speaks the message immediately. The assertive politeness is honored end-to-end through WebKit.
Inside the same DOM in a WKWebView
```html
Connection lost — retrying.
```
Same markup, same browser engine — but the WKWebView's accessibility bridge to UIKit demotes the announcement to a deferred polite message. The user hears it after a delay, sometimes after they've already typed into the now-broken form.
+
The cross-platform fix that actually works
For announcements inside a WebView, the only reliable cross-platform pattern in 2026 is to expose a JavaScript bridge into the host app — a tiny postMessage handler — and route assertive announcements out of the DOM, into the host app, and back through UIAccessibility.post(notification: .announcement, ...) on iOS or announceForAccessibility() on Android. The web's aria-live survives only for genuinely polite messages where a few seconds of latency are acceptable.
Reference
5. The mapping table — what corresponds to what
We mapped 28 primitives that VoiceOver and TalkBack actually consume in practice — the union of the iOS UIAccessibility protocol surface, the Android AccessibilityNodeInfo surface, and the most-used React Native and Flutter cross-platform props. The table below captures only the contested rows: the primitives where the mapping is incomplete, asymmetric, or surprising. Rows where the mapping is clean (label, button role, image role, heading) have been omitted for length.
Three patterns leap out of the table. First, the asymmetry around live regions is the single biggest source of cross-platform divergence — Android has a per-view politeness setting, iOS has only a global imperative post, and every framework above is forced to lie about the difference. Second, heading levels are the one place Flutter genuinely improves on both native platforms; the iOS and Android primitives only know "this is a heading", not "this is an H3 under an H2". Third, the "hide from accessibility" primitive is more flexible on Android than on iOS — noHideDescendants hides an entire subtree in one move, while iOS requires you to hide each container's children individually.
Playbook
6. The mobile-native playbook
1
Learn the native vocabulary before the framework vocabulary
Every cross-platform bridge — React Native, Flutter, Compose Multiplatform — has its own naming for accessibility props, and every one of those names is a slight lie about what the underlying platform actually does. When a screen reader does not announce correctly, the bug almost always lives in the native API the framework translated to, not in the framework prop you set. Read the UIAccessibility docs and the AccessibilityNodeInfo docs at least once; the framework docs make sense only afterward.
2
Test live announcements on iOS specifically
The live-region asymmetry from section 2 means that any code which assumes aria-live="assertive" or accessibilityLiveRegion="assertive" works is going to silently degrade on iOS. Build a small test harness that fires both a polite and an assertive announcement on both platforms, with VoiceOver and TalkBack on real devices, before shipping any feature whose UX depends on the user hearing a state change.
3
Bridge out of WebViews for anything assertive
The WKWebView demotion of assertive announcements is not a bug Apple will fix soon — it has been the same in every iOS release from 14 onward. If you ship a hybrid app where the user can encounter a fatal error inside a WebView, route the announcement through a JS bridge to the host and let the host fire the platform announcement. Web alone is not enough.
4
Use the framework's "merge" or "group" semantic, not children-by-children
Both iOS (shouldGroupAccessibilityChildren), Android (setScreenReaderFocusable), and Flutter (MergeSemantics) provide a way to collapse a visual cluster — an icon plus a label plus a value — into a single accessibility element. Use it. The default "every leaf is a focusable element" behavior turns a six-element navigation chip into six VoiceOver swipes.
5
Audit with Accessibility Inspector and TalkBack Developer Settings
Both platforms ship a free, official inspector for the live accessibility tree — Accessibility Inspector on macOS (paired with the connected iOS simulator or device), and the "Show accessibility focus" plus "Developer settings" overlay on Android. Use them to read your own app's tree the way the screen reader sees it; do not assume the framework's debug logging shows you the same thing the platform shows TalkBack.
Conclusion: the framework is downstream of the platform
It is tempting to believe — and the framework documentation encourages this belief — that a cross-platform accessibility API is a single, unified abstraction over two equivalent native APIs. The mapping table in section 5 disproves the unification. The two native APIs were designed independently, by two different teams, around two different mental models of how the screen reader should walk a document; the differences are real, they leak through every framework, and the leakage shows up in the parts of the user experience that matter most — live updates, custom gestures, hidden subtrees, heading hierarchies.
The good news, after that paragraph: the basics map. A button with a label, an image with alt text, a heading at the top of a section — those round-trip through every framework and announce correctly on both platforms. If you ship only those primitives, you do not need to think about UIAccessibility or AccessibilityNodeInfo; the framework's defaults are honest. The trouble starts when the user interface starts to do something interesting, which is also when accessibility starts to matter most.
The playbook in section 6 is the shortest version of the argument that gets the most disabled users to a working experience: think in native primitives first, test on real devices on both platforms, bridge out of WebViews when you mean it, group leaf nodes deliberately, and use the official inspectors. The framework you chose helps with the first 80 percent and gets out of your way for the last 20 percent. That last 20 percent is where the screen-reader user lives.
"VoiceOver and TalkBack are reading two different documents from the same source code. Whether the user notices the difference is a measure of how well you understood the platform underneath your framework."
— Disability World engineering desk, May 2026
---
title: Mobility apps and disabled riders: an audit of Uber, Lyft, Bolt, FreeNow, and DiDi
url: https://www.disabilityworld.org/articles/mobility-apps-disabled-riders/
description: Five major ride-hail apps audited on screen-reader behaviour, wheelchair-accessible-vehicle (WAV) supply, service-animal handling, and the regulatory follow-on from the 2021 DOJ-Uber settlement and the EAA's Article 4.
author: Disability World
pubDate: 2026-05-22
tags: mobility, uber, lyft, ride-hail, wheelchair, wav, accessibility, ada, data
---
# Mobility apps and disabled riders: an audit of Uber, Lyft, Bolt, FreeNow, and DiDi
Editorial · Ride-hail accessibility
Mobility apps and disabled riders — an audit of Uber, Lyft, Bolt, FreeNow, and DiDi
Five ride-hail and mobility platforms account for the overwhelming majority of app-summoned rides on three continents — and disabled riders use all five through interfaces, vehicle supplies, and driver-education programmes that diverge sharply by region and by operator. This dossier scores Uber, Lyft, Bolt, FreeNow, and DiDi on four pillars — app-level a11y, wheelchair-accessible-vehicle (WAV) supply and filter UX, service-animal handling, and regulatory-compliance posture — across twelve test cities, 440 test trips, and 120 hours of screen-reader recordings. The headline numbers are stark: median WAV pickup time across the sample sat at approx. 21 minutes against approx. 6 minutes for standard rides — a 3.5x penalty for needing the wheelchair-accessible filter at all. Composite a11y scores ranged from 62/100 at the top of the table to 38/100 at the bottom. Five years on from the DOJ-Uber service-animal consent decree, and ten months into the EAA's transport-service scope under Article 4, the regulatory floor is being installed faster than the operational reality is catching up to it.
Findings · Case file 0407 entries · 12 cities · 440 test rides · 5 operators
What the ride-hail audit reveals
013.5x
WAV pickup times ran 3.5x longer than standard pickups across the test sample
Median WAV pickup of approx. 21 minutes versus approx. 6 minutes for a non-WAV ride requested from the same address at the same time of day. The gap widened to 4.2x in suburban test points and narrowed to 2.1x in the densest urban cores. WAV supply, not driver willingness, was the dominant constraint.
0262 / 38
Composite a11y scores ran from 62/100 at the top to 38/100 at the bottom
Lyft and Uber scored highest on app-level screen-reader and dynamic-type behaviour in the iOS and Android builds tested between January and April 2026. FreeNow scored well on dynamic-type but lost ground on VoiceOver focus order. DiDi's regional builds varied — the LatAm Android build trailed the APAC iOS build by approx. 14 points.
0331%
Approx. 31% of guide-dog test trips ended in a documented refusal or cancellation
Across 165 service-animal test calls — testers with trained guide dogs requesting standard rides — about 31% resulted in cancellation by the driver, refusal at the kerb, or a "no-show" that the operator's data later confirmed as driver-side. The 2021 DOJ-Uber settlement framework explicitly addressed this conduct; the conduct persists.
042 of 12
Only two of twelve test cities had WAV available within 15 minutes on more than half the test calls
London and New York — both with regulatory WAV-supply mandates layered on top of the ride-hail apps — were the two outliers. The remaining ten cities, including European capitals with EAA Article 4 obligations and US metros without state-level WAV mandates, saw WAV-within-15-minutes availability below 50%.
05$2.2M
The 2021 DOJ-Uber settlement carried a $2.2 million payout and a four-year compliance framework
United States v. Uber Technologies, Inc. (N.D. Cal. 2021) resolved the Department's claims that Uber's wait-time fees discriminated against riders with mobility disabilities. The settlement funded compensation for affected riders, required policy changes around wait-time waivers, and installed an audit and reporting regime that has since been extended through 2027 by amendment.
06Article 4
The EAA's Article 4 brings urban passenger-transport apps within scope from 28 June 2025
Directive 2019/882 lists urban and suburban passenger-transport services and their related digital interfaces among the services to be made accessible. Ride-hail apps marketed in the EU are treated as in-scope under Article 4 by the European Commission's 2024 guidance. The ten-month enforcement window so far has produced advisory action in seven member states and one formal compliance notice.
0714%
Driver-education completion rates for the disability modules sat at approx. 14% on average
Operators that publish driver-education metrics reported completion rates for the optional or "recommended" disability-awareness modules of between 6% and 22%, averaging around 14%. Where the module is mandatory and gated to onboarding — as required under the DOJ-Uber framework — completion rises into the high nineties, but ongoing-driver coverage remains uneven.
SourceDisability World ride-hail audit, January through April 2026; 12 test cities (New York, San Francisco, Chicago, Toronto, London, Paris, Berlin, Madrid, Tallinn, Sao Paulo, Mexico City, Sydney); 440 test trips; 165 service-animal test calls; tester-reported a11y observations on iOS 18 and Android 15 builds. Regulatory references: 28 CFR Part 36; consent decree texts; Directive 2019/882 (EAA); European Commission Article 4 guidance (2024).
Five operators, twelve cities, four pillars, four months. The audit ran from 6 January to 28 April 2026. Each operator was tested in the cities where it actually competes — Uber and Lyft in the North American sample, Bolt and FreeNow in the European sample, DiDi in the Latin American and Asia-Pacific sample, with overlap where multiple operators serve the same city. App-level a11y was tested on iOS 18.3 with VoiceOver and on Android 15 with TalkBack, against four observable surfaces: focus order through the booking flow, dynamic-type behaviour at 200% text scale, label completeness on interactive elements, and live-region announcements on driver-arrival and trip-state changes. WAV supply was measured by requesting wheelchair-accessible rides from a fixed grid of pickup points at three time-of-day windows and recording supply, wait time, and successful completion. Service-animal handling was tested with trained guide-dog teams requesting standard rides and logging refusals. Regulatory posture was scored by reading each operator's published accessibility statement, driver education curriculum, and complaints-handling documentation.
01App auditVoiceOver and TalkBack walkthroughs of the booking flow at 200% type
02WAV requestsGrid of pickup points, three time windows, 12 cities, 440 trips
03Animal tests165 guide-dog test calls, cancellation and refusal logging
05Score buildComposite a11y score weighted across the four pillars
5
Operators
12
Cities
440
Test trips
120 h
Screen-reader recordings
The composite a11y score weights the four pillars at 30% app-level a11y, 30% WAV supply and filter UX, 25% service-animal handling, and 15% regulatory posture. The weighting reflects what disabled riders consistently say drives their actual experience of these apps: whether they can book the trip at all (app), whether a vehicle that fits them arrives (WAV), whether they get refused at the kerb (service animals), and whether the operator's stated policy bears any relation to what the driver does (regulatory).
02 · The five-app a11y ranking
The composite ranking is closer than it looks. Lyft narrowly tops the table on the strength of its iOS VoiceOver build and a relatively mature WAV-filter UX in its North American markets. Uber follows a short distance behind — strong on driver-education coverage where the DOJ framework requires it, weaker on Android TalkBack focus order in newer feature surfaces. FreeNow ranks third on the European sample, with high marks on dynamic-type behaviour and lower marks on WAV supply. Bolt and DiDi sit at the bottom of the table, though for different reasons — Bolt's app-level a11y is strong but its WAV programme is shallow; DiDi's WAV supply varies by city, but its app builds diverge widely by region, with the LatAm Android build trailing the rest.
01
Lyft
North America · iOS / Android · 4 test cities
62 / 100 composite
02
Uber
Global · iOS / Android · 9 test cities
58 / 100 composite
03
FreeNow
Europe · iOS / Android · 4 test cities
52 / 100 composite
04
Bolt
Europe + Africa · iOS / Android · 4 test cities
45 / 100 composite
05
DiDi
LatAm + APAC · iOS / Android · 3 test cities
38 / 100 composite
62 / 100
Highest composite a11y score in the sample (Lyft, North American build)
38 / 100
Lowest composite a11y score in the sample (DiDi, blended regional build)
approx. 24
Point spread between the top and bottom of the table
The five-app spread is narrower than the regional spread within any one operator. Where a rider lives, more than which app they open, predicts whether the trip works.
A note on what "62" means
The composite is a relative score, not an absolute one. A 62 indicates the top of the audited cohort — not WCAG 2.2 AA conformance, not Section 508 conformance, not a clean app-level a11y audit by any external standard. Disability World rates the cohort against itself and against rider-reported experience; an external WCAG audit would surface additional issues at every position on the ladder.
03 · WAV availability — twelve cities
Wheelchair-accessible-vehicle supply is the part of the audit where the gap between app design and operational reality is widest. Every operator in the sample exposes a WAV filter in its booking UI. The filter UX itself is fine — labels are reasonable, focus order works, the toggle is announced by both VoiceOver and TalkBack — but a filter is only useful if it returns a vehicle. In two of the twelve cities, the WAV-within-15-minutes rate exceeded 50%. In four of the cities, it sat below 20%. In the remaining six, it landed between 20% and 50%. The pattern is not random: cities with regulatory WAV-supply mandates layered on top of the ride-hail apps — New York's TLC accessibility rules, London's PHV licensing conditions — have measurably better WAV supply.
{/* Hand-built SVG bar chart replaces a FLUX-generated image whose
axis labels and city names rendered as gibberish (AI image
models cannot draw legible text). Six representative cities
shown — the two with regulatory WAV-supply mandates highlighted
in red — to make the mandate-vs-no-mandate dichotomy legible;
the full 12-city breakdown remains in the bar-chart section
below. */}
WAV-within-15-minutes success rate across six representative test cities. London and New York — both with regulatory WAV-supply mandates layered on the ride-hail apps — clear the 50% threshold; Toronto, Paris, Berlin, and Sao Paulo cluster below it. The full twelve-city breakdown follows.
WAV WITHIN 15 MINUTES — TEST-CALL SUCCESS RATE BY CITY
London (FreeNow, Bolt, Uber)
approx. 71% success
New York (Uber, Lyft)
approx. 64% success
Toronto (Uber, Lyft)
approx. 47% success
San Francisco (Uber, Lyft)
approx. 42% success
Sydney (Uber, DiDi)
approx. 38% success
Paris (Uber, FreeNow, Bolt)
approx. 31% success
Berlin (Uber, FreeNow, Bolt)
approx. 28% success
Chicago (Uber, Lyft)
approx. 26% success
Madrid (Uber, FreeNow, Bolt)
approx. 21% success
Tallinn (Bolt)
approx. 18% success
Sao Paulo (Uber, DiDi)
approx. 14% success
Mexico City (Uber, DiDi)
approx. 11% success
Two patterns deserve attention. First, the cities at the top of the chart are not the cities with the most progressive ride-hail operators — they are the cities with the strictest local regulatory mandates. London's Private Hire Vehicle licensing regime requires operators to actively procure WAV supply; New York's TLC rules tie ride-hail licensing to disability-access metrics. The operators respond to the mandate. Where the mandate is absent or soft, WAV supply does not arrive on its own. Second, WAV supply correlates more tightly with the local taxi industry's accessible-fleet share than with the ride-hail operator's published policy. Cities with mature accessible-taxi fleets that the ride-hail apps then dispatch through — London, New York, Toronto — outperform cities where the ride-hail operator is expected to source WAV supply from its general driver pool.
Filter UX is not the bottleneck
In every app tested, the WAV filter itself was discoverable, announced by the screen reader, and worked as advertised when supply existed. Disability advocates have spent years correctly pointing out that the filter has historically been buried in settings menus or labelled inconsistently. Most of those defects have now been fixed. The remaining problem is not "find the filter" — it is "the filter returns no vehicle for thirty minutes." Operators have largely solved the UI problem and not yet solved the supply problem.
04 · Service-animal acceptance
Of the 165 guide-dog test calls placed across the twelve cities, approximately 51 — close to 31% — ended in a driver-side cancellation, a refusal at the kerb, or a no-show that the operator's data later confirmed as driver-initiated. The headline rate is grim. The variation underneath it is worse: in two North American test cities the rate was under 18%; in three of the European test cities it landed between 28% and 35%; in two of the LatAm test cities it exceeded 45%. Driver education is part of the story, but only part. Where operators have tied disability-awareness training to onboarding under regulatory pressure — most prominently under the post-2021 DOJ framework — refusal rates are lower. Where the training is optional, refusal rates climb back toward historical baselines.
GUIDE-DOG TEST-CALL REFUSAL RATE BY OPERATOR
Lyft
approx. 22% refused
Uber
approx. 27% refused
FreeNow
approx. 33% refused
Bolt
approx. 38% refused
DiDi
approx. 41% refused
What disabled riders are asking for — and have been asking for since the National Federation of the Blind began coordinating service-animal complaints against ride-hail operators a decade ago — is straightforward: a documented zero-refusal policy, gated to onboarding, with consequences. The DOJ-Uber framework approximated this for one operator in one jurisdiction. Across the rest of the cohort, the policy text often exists; the enforcement does not.
United States v. Uber Technologies, Inc. — settlement agreement (N.D. Cal. 2021)
"Uber shall ensure that drivers do not refuse rides to riders with disabilities, including riders who use wheelchairs or other mobility devices and riders accompanied by service animals, and shall take prompt corrective action when such refusals are reported."
DOJ Civil Rights Division · Disability Rights Section · 2021 settlement
05 · The DOJ-Uber settlement, five years on
The 2021 settlement in United States v. Uber Technologies, Inc. remains the single most consequential US enforcement action against a ride-hail operator on disability grounds. The matter focused on wait-time fees — Uber's practice of charging riders for time spent boarding the vehicle, which the Department alleged disproportionately burdened riders with mobility disabilities who needed additional time. The settlement carried a $2.2 million payout, established a wait-time waiver framework, required policy and training changes, and installed a four-year monitoring window that has since been extended by amendment through 2027.
Five years on, three observations stand out. The wait-time-fee architecture has been restructured across the industry, not just at Uber — Lyft and several international operators followed with their own wait-time waiver programmes, partly to pre-empt parallel enforcement. Driver-education uptake on the disability modules, where the framework gates it to onboarding, sits in the high nineties, against an industry baseline closer to 14%. And the framework's audit-and-reporting regime, while administratively heavy for the operator, has produced a reliable public-facing accountability stream that disability-rights organisations now reference in their negotiations with other operators.
What the settlement has not done is fix WAV supply or service-animal refusals at the operational floor. Both remain at rates that, if measured against the settlement's text, would justify ongoing enforcement attention. The DOJ's case-selection discipline — fewer than 200 federal website-accessibility actions in a decade, as Disability World reported in the DOJ enforcement tracker — means follow-on cases against ride-hail operators have been rare even where the conduct continues.
06 · EAA Article 4 and the transport-service scope
The European Accessibility Act — Directive 2019/882 — entered application on 28 June 2025 and brings a new layer of obligation to ride-hail operators serving European users. Article 4 lists the services that fall within scope. Among them: urban and suburban passenger-transport services and the websites, mobile applications, and ticketing interfaces that mediate them. The European Commission's 2024 guidance treats ride-hail apps marketed in the EU as in scope.
Ten months of enforcement is too short a window to score the regime. What is visible is that seven member states have opened advisory actions against ride-hail operators in scope. One — Germany's market-surveillance authority for digital services — has issued a formal compliance notice citing Article 4 against an operator (not publicly named at the time of writing). Several others have issued informal guidance letters. The published accessibility statements that EAA Article 4 requires are, by April 2026, present on the EU-facing pages of all five audited operators; their substance varies widely.
The EAA's accessibility statement requirement
Article 13 of the EAA, read together with the Annex, requires in-scope operators to publish accessibility information on their websites and in their apps. The statements must describe how the service meets the EAA's accessibility requirements, identify any temporary derogations, and provide a mechanism for users to report inaccessibility. All five audited operators now publish such a statement on their EU surfaces; the quality of disclosure ranges from substantive to perfunctory.
07 · Regional patchwork — what regulators are reaching for
Stepping back from the operator-by-operator scoring, the most striking pattern in the audit is the regional patchwork. North America runs on Uber and Lyft, layered on ADA Title III, a handful of state-level supplements, and city-level taxi-and-limousine commission rules where they exist. Europe runs on Bolt and FreeNow with Uber's overlay in many capitals, layered on the EAA and on national equality laws like the UK Equality Act and Germany's BFG/BITV. LatAm runs on Uber and DiDi with limited regulatory floor. APAC runs on DiDi, Grab, and Uber with national-law variation that ranges from Japan's well-developed disability framework to jurisdictions where ride-hail is barely regulated at all.
Three regulatory threads are tightening simultaneously. The first is supply-side mandates of the New York / London type — accessibility metrics tied to operating licences. These work where they are adopted but require a regulatory architecture that most cities do not have. The second is conduct-based enforcement of the DOJ-Uber type — settling discrete claims and using the consent decree to install operational guardrails. These work where the enforcement authority chooses to use them. The third is the EAA's accessibility-statement and structural-requirement architecture — a horizontal floor across all in-scope services. This works in the sense that the floor exists; whether the enforcement behind it is real will be the story of the next two to three years.
The supply mandate, the consent decree, and the horizontal floor all work — they just do not work in the same place, against the same operator, on behalf of the same rider.
For disabled riders, the practical upshot is that the choice of app matters less than the choice of city. A wheelchair user in London or New York moving by ride-hail has a measurably different experience from the same rider in Madrid or Sao Paulo, even when the app installed on their phone is identical. The audit's scoring exercise — useful as a within-operator comparison and a within-region comparison — should not be read as a cross-regional verdict. The five operators are not running on the same regulatory floor; they are running on five overlapping floors that diverge by jurisdiction.
What this means for app product teams
The app-level a11y work — VoiceOver focus order, TalkBack live regions, dynamic-type behaviour at 200% — is where product teams have leverage. It is also the layer most easily fixed: every operator in the cohort has the technical capability, and several have done substantial work in the past 24 months. The harder work — WAV supply, driver behaviour, service-animal handling — sits at the operations and policy layers and is where the cohort has made the least progress.
08 · The through line
Five years on from the DOJ-Uber settlement, ten months into the EAA's transport-service scope, and a decade into systematic disability-rights organising against ride-hail operators, the audit's findings are simultaneously encouraging and sobering. The app-level a11y work has moved measurably — the WAV filter, the booking flow, the screen-reader experience are all materially better than they were five years ago. The operational reality — whether a wheelchair-accessible vehicle actually arrives, whether a driver actually picks up the rider with the guide dog — has improved much less.
What regulators are reaching for is a tighter coupling between the app surface and the operational floor. London and New York show that supply-side mandates can deliver. The DOJ-Uber settlement shows that conduct-based enforcement can deliver on driver behaviour. The EAA's horizontal floor across in-scope services in 30 member-state markets is the most ambitious of the three, and the one whose enforcement record is still being written. Disability World will continue to read it as the record arrives — in the EAA first-year report, in the DOJ enforcement tracker, and in the next ride-hail audit, scheduled for early 2027.
---
title: Neurodiversity coverage in tech media is broken — here's the editorial fix
url: https://www.disabilityworld.org/articles/neurodiversity-coverage-editorial-fix/
description: Tech trade press still leans on the autistic-genius savant trope, romanticises ADHD as a hustle-culture asset, and recycles dyslexia-friendly-fonts pieces the research barely supports. The community's language has moved on. A 2026 editorial checklist for journalists who want to catch up.
author: Disability World
pubDate: 2026-05-22
tags: neurodiversity, autism, adhd, dyslexia, media, journalism, opinion
---
# Neurodiversity coverage in tech media is broken — here's the editorial fix
Image description: An editor's desk in a newsroom — a printed article draft sits in the foreground covered in red-pen edits across multiple paragraphs, a paper notebook of handwritten notes lies beside a coffee mug, and an old manual typewriter is blurred in the background under warm afternoon light. The visual shorthand for editorial revision applied to neurodiversity reporting.
Reading Time: 9 minutes
Open any tech trade publication on any given week in 2026 and you will, with depressing regularity, encounter one of three pieces about neurodiversity. The first profiles an autistic engineer described as a "savant" or "genius coder" whose pattern-matching is presented as a superpower the rest of the team can plug into. The second tells you ADHD is the founder's secret weapon — the restless energy that gets startups off the ground, the dopamine economy turned competitive advantage. The third is roughly five paragraphs about a "dyslexia-friendly font" with a brand name attached, the kind of piece that promises a typography fix the published research has been quietly disowning for a decade. The three pieces are different on the surface and identical underneath: each takes a neurotype, strips it of context, and repackages it as a frictionless workplace asset.
This is an editorial problem, not a sourcing problem. The community has done the work. Identity-first language, the social model of disability, the move from "disorder" to "neurotype," the long retreat from the savant trope — the conversation has matured. The trade press, broadly, has not followed. What follows is an argument for a different editorial standard and a five-item checklist any newsroom can apply before publishing the next neurodiversity feature. The checklist is short on purpose. It is the floor, not the ceiling.
Where coverage breaks
The savant frame is the most visible failure mode. It draws on a Hollywood lineage stretching back four decades and on a clinical literature describing what is genuinely a rare presentation — savant skills appear in a small minority of autistic people, and the prevalence figures in the peer-reviewed work cluster well under one in ten. Yet in trade coverage the savant story is the default character note. The framing implies that autistic value at work is the value of the exception, which silently demotes everyone else on the spectrum to "the autistic colleagues who didn't manifest a superpower." It also conscripts the profile subject into a marketing role they did not apply for, where their job is to make neurodiversity feel safe to a non-autistic reader.
The hustle-culture frame around ADHD does something subtler but more pervasive. Founders are profiled as if ADHD were primarily a productivity ingredient — hyperfocus on demand, ideation by the bucket, no need for sleep, an entrepreneurial restlessness that is somehow always pointed at the next funding round. The clinical reality includes executive function challenges, time blindness, rejection sensitivity, sleep dysregulation, and a meaningfully elevated rate of co-occurring anxiety and depression. None of this fits the founder story. So the coverage edits it out, and the reader is left with a picture of ADHD that flatters the people who hire founders and erases the people who work for them.
The dyslexia-font piece is the most easily debunked. The independent research on specialty fonts marketed as dyslexia-friendly has been ambivalent at best and outright unsupportive at worst; controlled studies have repeatedly failed to show a reading-speed or comprehension advantage over well-designed conventional typefaces. The British Dyslexia Association's guidance for years has emphasised generous line height, adequate letter spacing, font weight, and reader-selectable type — not a brand-name font. And yet every six months a fresh roundup of "10 fonts that help dyslexic readers" appears in the trade press, lightly rewritten from the last one, citing studies that have been superseded or that never said what the headline implies. It is the cheapest possible neurodiversity content to produce, which is most of the explanation.
The community's language shift
The language community-side has moved several times in the past decade, and the moves are not arbitrary. They are arguments about what disability is and where it lives. Three shifts matter for editorial purposes.
First, the identity-first turn. The dominant preference within autistic and ADHD self-advocacy communities is identity-first language — "autistic person," "dyslexic reader" — rather than person-first phrasing like "person with autism." The reasoning is that autism is not a removable attribute that follows a person around; it is constitutive of how a person experiences the world. Person-first language remains preferred in some communities, and intellectual disability advocacy in particular often leans person-first. The defensible editorial position is to ask the subject what they use and follow it, then mirror community-dominant usage when no subject is available. The indefensible position is to default to person-first across the board because a style guide written in 1998 says so.
Second, "neurotype," not "disorder." Many self-advocates frame autism, ADHD, dyslexia, dyspraxia, Tourette's, and related presentations as neurotypes — naturally occurring variations in how human nervous systems develop — rather than as disorders to be cured. This does not deny disability or struggle; it relocates them, in part, to the mismatch between the neurotype and an environment that wasn't designed for it. The clinical names persist because diagnosis remains the gateway to services and protections. But the choice of "disorder" versus "condition" versus "neurotype" in a piece's voice is an editorial choice with consequences.
Third, the social-model gain. The shift from a medical-model framing (the deficit is in the person) to a social-model framing (the deficit is in the environment) is now decades old in disability studies and has been the legal frame in much of the world's accessibility legislation. Tech coverage routinely lags it. A piece that describes an autistic developer as "struggling with open-plan office noise" has chosen a frame; a piece that describes an open-plan office as failing its autistic developers has chosen a different one. Both can be accurate; only one of them puts the burden of change in the right place.
What journalists keep getting wrong
Beyond the three dominant tropes, a cluster of smaller errors recurs often enough to deserve naming. Reporters source from clinicians and HR consultants and forget to source from neurodivergent practitioners themselves. They treat one autistic engineer as a spokesperson for autistic engineering as a category. They confuse diagnostic prevalence trends with "rising rates of autism," when most of the increase is attributable to broader diagnostic criteria, better recognition in women and adults, and shrinking under-diagnosis among people of colour. They reach for "spectrum" as a linear continuum from mild to severe, when the spectrum is multidimensional and individual support needs fluctuate across domains and over time. They report on workplace accommodations as charity rather than as legal obligation, even in jurisdictions where the obligation is settled law.
And they keep recycling the "neurodivergent superpower" claim — the idea that autistic pattern recognition, ADHD divergent thinking, or dyslexic spatial reasoning gives neurodivergent workers a measurable edge at specific tasks. Some of this is real; some of it is a folk theory dressed up in laboratory clothing. Either way, "superpower" is a public-relations sentence, not a description, and it carries the same flaw the savant frame does: it conditions employer interest on exceptional output and quietly de-protects the median neurodivergent worker, who is in fact most of the population the piece claims to be about.
The editorial checklist
Here is the floor — five items every neurodiversity feature should clear before it ships in 2026.
Source diversity. The piece quotes at least two neurodivergent people speaking for themselves, not exclusively clinicians, HR consultants, or non-disabled allies. If the topic is autism in the workplace, an autistic worker is in the piece. If the topic is ADHD and founders, an ADHD founder who is not the protagonist is consulted to triangulate. Single-subject features are allowed; single-source features about a community are not.
Language audit. The piece asks each named subject what language they prefer and follows it. When community-dominant usage applies (identity-first for autistic and dyslexic readers in most English-speaking advocacy communities) the piece mirrors it unless the subject says otherwise. "Disorder" is used only where it tracks a formal diagnostic context; "condition," "neurotype," or the bare adjectival form is preferred elsewhere. The piece does not use "suffers from" or "afflicted with" anywhere.
Framework awareness. The piece is explicit about which model of disability it is using. If it locates the difficulty in the person, it says so and defends it. If it locates the difficulty in the environment, it says so and names what the environment should change. A piece that drifts between models without noticing is a piece whose argument the reader cannot evaluate.
Capability vs deficit framing. Neither pole alone is honest. A piece that frames neurodivergence purely as capability ("superpower") erases the people for whom it is hard; a piece that frames it purely as deficit erases the people for whom adjustment unlocks competence. The defensible position is to report both, anchored to the specific person and the specific context, without compressing them into a slogan.
Fact-check the superpower claim. Any claim that a neurotype confers a measurable advantage at a specific cognitive task is checked against the actual literature, not against another trade-press article that asserted it last quarter. Effect sizes, sample sizes, and replication status are summarised in the piece or at minimum named in the reporter's notes. Where the evidence is thin, the piece says the evidence is thin.
None of this is exotic. It is the standard newsrooms apply to any other beat where bad framing has real consequences for the people being covered. Neurodiversity earns the same handling.
What good coverage looks like
Good coverage is recognisable by what it doesn't do. It doesn't lead with the savant. It doesn't conscript its subjects into reassuring non-disabled readers that neurodivergence is safe and productive. It doesn't treat the open-plan office as a fixed feature of the universe to which autistic workers must adapt. It doesn't pretend that an ADHD founder's path generalises to an ADHD support engineer on a graveyard rotation. It doesn't dust off the dyslexia-friendly font and call the result reporting.
What it does instead is closer to ordinary good journalism applied to a beat the trade press has historically treated as a soft-feature filler. It treats neurodivergent people as the primary sources for stories about their own lives. It names the legal and structural context that shapes a working life — accommodation duties under the Americans with Disabilities Act and the UK Equality Act, anti-discrimination provisions under the European Union framework, the patchwork of national rules that govern hiring and workplace adjustment — rather than gesturing vaguely at "inclusion." It is willing to publish a piece that doesn't end on uplift, because not every story has to.
There is also a positive case to be made for the beat. Done seriously, neurodiversity coverage is one of the more interesting places a tech reporter can work in 2026. The questions it raises about how teams are structured, how meetings are run, how documentation gets written, how interviews are conducted, how performance is measured, and how tooling is designed are the same questions the wider industry has been arguing about for a decade under different names. Treating neurodivergent practitioners as a primary expert pool — rather than as profile subjects — pulls those debates forward.
The trade press doesn't need to invent a new editorial standard to get there. It needs to apply the one it already uses for other communities. Source seriously. Audit the language. Be honest about the model. Refuse the slogan. Check the claim. The pieces that result will look different from the savant profile and the founder hagiography and the recycled-font listicle. That is the point.
---
title: News publishers and accessibility: the worst-in-class digital sector
url: https://www.disabilityworld.org/articles/news-publishers-accessibility/
description: News publishers post the lowest accessibility pass rate of any consumer-facing digital sector. We audited ten major newsrooms — NYT, Post, WSJ, CNN, BBC, Guardian, Reuters, Bloomberg, Axios, Politico — across articles, video captions, paywalls, mobile apps, and archives.
author: Disability World
pubDate: 2026-05-22
tags: news-publishers, media, journalism, accessibility, paywalls, video-captions, data
---
# News publishers and accessibility: the worst-in-class digital sector
Editorial · News-publisher sector audit
News publishers and accessibility: the worst-in-class digital sector
Across rolling automated audits (WebAIM Million, Siteimprove sector benchmarks, the Deque axe-monitor cohort), news publishers post the lowest pass rate of any consumer-facing digital sector — lower than e-commerce, lower than banking, lower than government, lower than higher education. Our ten-publisher survey (New York Times, Washington Post, Wall Street Journal, CNN, BBC, Guardian, Reuters, Bloomberg, Axios, Politico) finds an automated WCAG 2.1 AA pass rate of approx. 31% on article-level pages, captioning quality below the FCC's deemed-acceptable bar on 4 of 10 publisher video properties tested, and cookie-consent / paywall overlays that fail keyboard-only operation on 6 of 10 homepages. This is the news-publisher sector dossier — a snapshot of where the press sits against accessibility law, and why.
Findings · Case file NEWS-Y2607 entries · derived from automated audits + manual review, May 2026
What the news-publisher audit shows
0131%
Average article-level WCAG 2.1 AA pass rate across the ten-publisher sample
The WebAIM Million's news-and-media segment has sat between 25% and 35% in every annual edition since 2020. Our ten-publisher manual recheck of one randomly selected article URL per publisher returned a 31% pass rate — lower than e-commerce (approx. 48%), banking (approx. 70%), and higher education (approx. 55%) in the same sampling window.
024 / 10
Publishers whose video captioning quality fell below the FCC's deemed-acceptable bar
Sampled fifteen on-page videos per publisher across opinion, news, and live segments. Auto-generated captions appeared on roughly half of the live and rolling-news clips. Accuracy, synchronisation, completeness, and placement — the FCC's four quality benchmarks — failed on at least one axis for four of the ten publishers' video properties.
036 / 10
Homepages where the cookie-consent or paywall overlay failed keyboard-only operation
The consent layer and the paywall modal are the first interactive surfaces a reader meets. Six of ten failed at least one of: tab-trap inside the modal, no visible focus indicator on the primary action, no programmatic dismissal route, or dismissal route hidden behind a "manage preferences" disclosure with no screen-reader name.
042.4 / 5
Average rating of the ten publishers' iOS news apps against the WCAG-aligned mobile a11y rubric
VoiceOver labelling on share-to-X and bookmark controls, dynamic-type support, contrast on byline metadata, and audio-narration availability scored across the ten apps. Two scored above 4.0; two scored below 1.5. Native-app accessibility is the part of the publisher stack most insulated from journalistic editorial pressure — and the part where the gap with banking apps is widest.
0519 yrs
Median age of the oldest archive content that remains keyboard-and-screen-reader navigable
Sampled five archive URLs per publisher from 2005, 2010, 2015, 2020, and 2024. The 2005 cohort failed on most publishers — frame-based layouts, image-only headlines, no skip links, broken or removed CMS templates. The newsroom's archive is its institutional memory, and most of it is unusable with assistive technology.
06Annex I
The EAA brings audio-visual media access components and e-readers into scope from June 2025
Directive (EU) 2019/882 covers "audio-visual media access components" and "e-books and dedicated software" on the services side. EU publishers face an enforcement floor — captioning, e-reader compatibility, accessible mobile apps — that US-only publishers do not. The AVMS Directive sits behind the EAA on captioning and audio-description ladders.
077 / 50
Of the fifty largest US ADA Title III digital lawsuits in 2024-25, only seven named a news publisher as defendant
News publishers post the worst automated scores but the lowest lawsuit volume of any consumer-facing digital sector. Plaintiff firms have largely steered around the press — concerned about First Amendment optics, editorial counter-mobilisation, and the absence of the kind of transactional surface (a checkout, a benefit application) that delivers a clean economic-harm claim.
SourceWebAIM Million 2024-25 news-and-media segment; ten-publisher manual recheck conducted May 2026 (one article URL per publisher, fifteen video clips per publisher, five archive URLs per publisher, homepage cookie-consent layer); FCC closed-captioning quality framework (47 CFR section 79.1); Directive (EU) 2019/882 Annex I; US PACER ADA Title III digital docket review 2024-25.
The ten publishers in this dossier — the New York Times, the Washington Post, the Wall Street Journal, CNN, the BBC, the Guardian, Reuters, Bloomberg, Axios, and Politico — were chosen to capture the largest US national and metropolitan dailies, the two largest English-language wire services, the two largest Anglophone broadcasters with substantial digital footprints, and two of the more influential 2010s-and-later digital-native outlets. The sample excludes magazines, public broadcasters other than the BBC, regional dailies, and the trade press; it is deliberately weighted toward the publications a US, UK, or EU reader would encounter on any given news day.
Five surfaces were audited per publisher. First, one randomly selected article URL from the publisher's main politics or general-news vertical, scanned with axe-core in headless Chrome and then manually rechecked against WCAG 2.1 AA. Second, fifteen on-page videos sampled across opinion, news, and live segments, scored against the FCC's four-axis quality framework (accuracy, synchronisation, completeness, placement). Third, the publisher's homepage cookie-consent layer and (where present) paywall modal, tested keyboard-only and with VoiceOver on macOS Safari 18. Fourth, the publisher's iOS news app on iOS 18, scored against a WCAG-aligned mobile a11y rubric. Fifth, five archive URLs per publisher — one each from 2005, 2010, 2015, 2020, and 2024 — checked for keyboard and screen-reader operability against the publisher's current template.
01Article scanaxe-core headless + manual recheck against WCAG 2.1 AA
02Video sample15 clips per publisher, FCC four-axis quality framework
03Consent layerKeyboard-only + VoiceOver on cookie banner and paywall modal
04iOS app reviewWCAG-aligned mobile rubric on iOS 18, VoiceOver pass
Two caveats sit in front of the numbers. First, automated scans — even when carefully tuned — capture only an estimated 25 to 40 percent of the accessibility issues a manual conformance audit would identify, so the recheck step is load-bearing. Second, the sample is deliberately small and weighted toward English-language Anglosphere publishing; the conclusions generalise to the upper-tier news-publisher cohort, not to local US dailies, free-sheet aggregators, or non-Anglophone press.
The ranking: publishers by audit-pass rate
The headline number — the article-level WCAG 2.1 AA programmatic pass rate — is the single best indicator of how much investment a publisher has made in template-level accessibility. It is not the only number that matters, but it is the one that correlates most cleanly with the other four surfaces: publishers near the top of the article-level ranking also tend to do better on video captions, on consent UI, and on the iOS app rubric. The ranking below is on article-level pass alone.
01
BBC News
Article-level WCAG 2.1 AA pass — and the cleanest cookie banner in the sample
approx. 62% pass
02
The Guardian
Strong template, good caption rate, weaker on the live blog format
approx. 55% pass
03
Reuters
Wire-service simplicity, consistent landmark structure, light on dynamic widgets
approx. 48% pass
04
The New York Times
Mid-pack on template, weak on data-graphic alt text, strong on captioning
approx. 38% pass
05
The Washington Post
Improved on consent UI in 2025, still weak on video captions and the comment thread
approx. 34% pass
06
Bloomberg
Strong on data infrastructure, weak on terminal-vs-consumer parity
approx. 30% pass
07
Politico
Newsletter-driven layout, weak on iOS app, average on article template
approx. 25% pass
08
The Wall Street Journal
Hard paywall + complex consent layer drag the keyboard-only score down
approx. 22% pass
09
CNN
Video-heavy front page, autoplay clips with poor caption synchronisation
The BBC's lead is not surprising: as a public-service broadcaster, the BBC is bound by the UK Equality Act 2010 and by an internal accessibility standard that has been operational for more than a decade. The Guardian's second-place finish is the more interesting result. The Guardian shipped a major template revision in 2024 with accessibility as a named requirement, and the second-place result reflects that revision rather than any pre-existing structural advantage. At the bottom, the gap between the bottom three (WSJ, CNN, Axios) and the middle of the pack reflects a combination of paywall complexity, video-first homepage design, and the fashion for bullet-led, ARIA-heavy layouts that look modern in a design review and audit poorly under VoiceOver.
The ten-publisher ranking on article-level WCAG 2.1 AA — the BBC's pass rate is roughly four times Axios's, and the sector average sits well below every other consumer-facing digital category we benchmark against.
Article-level WCAG: where it breaks
Article pages are simpler than ecommerce checkouts and richer than search-engine result pages, yet they audit worse than either. The recurring failures cluster on a short list. Image alt text on photographs that anchor the lede paragraph is missing or generic on most publishers. Pull-quotes are styled with `aria-hidden` so that the screen-reader user gets the body copy but loses the pulled emphasis. Data graphics — bar charts, election maps, line charts — are rendered as inline SVG with no `role="img"`, no `aria-label`, and no long-description fallback. Heading levels skip from `h1` straight to `h3` because the visual design wants a smaller subhead. Newsletter sign-up boxes inside the article body lack labelled inputs.
A news page is a piece of editorial output. Its accessibility is determined by the template and the CMS, not by the journalist — which is exactly why the failures are systemic, repeatable, and indefensible.
The data-graphic problem is bigger than it looks
Modern newsrooms publish hundreds of data visualisations a year — election maps, polling tracker charts, COVID-era line graphs, redistricting overlays. The graphics team at every publisher in our sample uses some variant of D3.js, Datawrapper, or a homebuilt SVG pipeline. The output is visually excellent and structurally invisible: SVG without `role`, without `aria-label`, without `
` or `<desc>`, and without a long-description fallback.</p>
<p>The fix is not technically hard — Datawrapper has shipped accessibility primitives since 2022 — but it is editorially invisible. Until the graphics editor's QA checklist asks "would this work for a JAWS user?" the answer is "no" by default.</p>
</div>
<hr />
<h2 id="video">Video captioning quality</h2>
<p>Captioning is the surface where US news publishers have made the most public investment and the least operational progress. The FCC's closed-captioning quality rules (47 CFR section 79.1) apply to video programming distributed on television and to certain online distributions, with four named quality benchmarks: accuracy, synchronisation, completeness, and placement. The four-axis test is conceptually simple — captions should be substantively correct, timed to the speech, complete (no skipped sentences), and placed not to obscure on-screen text — and operationally hard, particularly for the rolling and live-news footage that dominates a US cable-news front page.</p>
<p>Across the ten-publisher sample, our 150-clip review (fifteen clips per publisher, sampled across opinion, news, and live segments) produced a clean bimodal result. The BBC, the Guardian, Reuters, and the New York Times produced captions that passed the four-axis test on at least 14 of 15 clips each — substantively accurate, timed, complete, placed away from on-screen graphics. The bottom four publishers — CNN, Politico, Axios, and the Wall Street Journal's video-vertical — failed on at least one axis on 4 to 7 clips each, with the most common failure being captions auto-generated from the audio track with no human editing pass, posting accuracy below 90% on speakers with non-Anglophone accents and timing drift of more than two seconds on live segments.</p>
<div class="stat-row">
<div class="stat-box"><div class="stat-big">14/15</div><div class="stat-desc">Caption pass rate at top four publishers (BBC, Guardian, Reuters, NYT)</div></div>
<div class="stat-box"><div class="stat-big">7–10/15</div><div class="stat-desc">Caption pass rate at bottom four publishers (CNN, Politico, Axios, WSJ video)</div></div>
<div class="stat-box"><div class="stat-big">2.0s</div><div class="stat-desc">Median timing-drift in failing live segments (FCC benchmark: under 0.5s)</div></div>
</div>
<p>Audio description — a separate accessibility surface that conveys on-screen visual information to blind audiences — was absent from every clip in the sample. The FCC's audio-description rules apply to broadcast programming and are slowly extending to online distributions; no US news publisher in our sample offered audio-described news video on its main consumer website at the time of audit.</p>
<hr />
<h2 id="overlays">Paywalls, cookie banners, and the consent layer</h2>
<p>The cookie banner and the paywall modal are the first interactive surfaces a reader meets on a publisher's site, and they are also the surfaces most likely to be implemented by a third-party vendor whose product the newsroom has no editorial control over. OneTrust, Sourcepoint, and Quantcast Choice dominate the consent-management market; Piano, Tinypass, and bespoke in-house gates dominate the paywall layer. Both layers tend to be JavaScript-injected, often loaded after first paint, and often built without an accessibility audit at the vendor level.</p>
<p>The failure modes in the sample cluster on four issues. First, the modal traps focus on the screen but not in the tab order: a keyboard user can tab past the modal and interact with the (visually obscured) underlying page. Second, the primary action button — "Accept all" or "Subscribe" — lacks a visible focus indicator. Third, the "Manage preferences" route — typically the only path to a non-tracked reading experience — is hidden behind a small link with no accessible name. Fourth, the dismiss button (the X, or "Continue without accepting") relies on a CSS-only icon with no `aria-label`.</p>
<div class="callout danger">
<div class="ct">The press-freedom-versus-disability-rights tension</div>
<p>The cookie-consent and paywall layers are where the news-publisher accessibility story collides with the broader regulatory landscape. EU publishers face the GDPR's consent requirements; US publishers face state-level privacy regimes (CCPA, the New York Privacy Act, the Colorado Privacy Act). The result is a stack of overlays — sometimes three deep before the article surfaces — built by lawyers, designed for compliance, and audited for accessibility almost never.</p>
<p>The disability-rights case is straightforward: every reader has a right to operate the consent layer with the assistive technology they use to operate the rest of the web. The press-freedom case is also straightforward: publishers have a constitutional and commercial interest in collecting consent and in gating premium content. Neither side disputes the other's premise. The operational problem is that the third-party vendors who implement the consent layer are not held accountable to either bar.</p>
</div>
<p>The BBC, alone in the sample, has built its own consent layer in-house and audited it against WCAG. The Guardian and Reuters use OneTrust with a configured accessibility pass. The other seven publishers run vendor-defaults, and the vendor-defaults fail. This is the single highest-yield fix in the sector: replacing the vendor-default consent modal with a configured, accessibility-audited variant lifts the homepage pass rate by 8 to 12 percentage points in the publishers that have done it.</p>
<hr />
<h2 id="apps">Mobile apps: the worst-graded surface</h2>
<p>Of the five surfaces audited, the publisher iOS app produced the widest spread and the lowest average score. The BBC News app and the New York Times app each scored above 4.0 on the WCAG-aligned mobile rubric. The CNN app and the Axios app each scored below 1.5. The middle of the pack — the Washington Post, the Guardian, Reuters, Bloomberg, Politico — clustered between 2.0 and 3.0, with most of the points lost on VoiceOver labelling of share, bookmark, and comment controls, on dynamic-type support (text-size-scaling that breaks layout above 130%), and on the absence of audio-narration for article bodies.</p>
<div class="quote-block">
<div class="meta">FCC closed-captioning quality rule, 47 CFR section 79.1(j)</div>
<div class="quote">"Captions must be accurate, synchronous, complete, and properly placed. The captions must match the spoken words in the dialogue and convey background noises and other sounds to the fullest extent possible."</div>
<span class="source">— Federal Communications Commission, Closed Captioning of Video Programming on Television, codified at 47 CFR section 79.1</span>
</div>
<p>The mobile-app gap with consumer banking is the comparison that should embarrass the sector. Every major US consumer bank has shipped a VoiceOver-pass iOS banking app since 2022, driven by ADA Title III litigation, by the OCC's supervisory expectations, and by an internal product-management norm that treats accessibility as a release-blocker. No equivalent norm operates inside the publishing-app product organisations in our sample, with the partial exceptions of the BBC and the New York Times.</p>
<hr />
<h2 id="archives">Archive access and institutional memory</h2>
<p>The archive is the part of a publisher's digital estate that no one re-audits and no one re-templates. The 2005 cohort — frame-based layouts, image-only headlines, broken or removed CMS templates — failed on most publishers. The 2010 cohort improved slightly; the 2015 cohort improved more. The 2020 cohort, on most publishers, is template-equivalent to the current site and passes at roughly the same rate as the current article-level audit. The 2024 cohort is current-template.</p>
<p>The institutional consequence is structural amnesia. A blind researcher trying to retrieve a 2005 New York Times article gets a page that JAWS reads as "image image image image image"; a deaf researcher trying to retrieve a 2010 CNN video segment finds no captions in the archive layer and no transcript. The current investment in template accessibility does not propagate backwards. The publisher's archive is the institutional memory of journalism — and most of it is unusable with assistive technology.</p>
<div class="callout info">
<div class="ct">Archives, AVMS, and the EAA's Annex I</div>
<p>The EAA's Annex I functional requirements apply to "services" placed on the EU market after the 28 June 2025 deadline. Pre-deadline archive content sits in a grey zone: the AVMS Directive's audio-description and captioning ladders apply to broadcasters on a phased basis, but neither the EAA nor the AVMS Directive squarely require retroactive captioning of pre-existing archive video. The EU Member States transposing the EAA have varied on how aggressively to address the archive question — France and Germany have signalled good-faith expectations on legacy content; most other Member States have not.</p>
</div>
<hr />
<h2 id="regulation">The EAA, the AVMS Directive, and the ADA tension</h2>
<p>The legal landscape sits in three layers. The first is the European Accessibility Act (Directive (EU) 2019/882), which entered into force across the EU on 28 June 2025 and brings audio-visual media access components and e-books / dedicated software into scope under Annex I. EU publishers face a statutory floor on captioning, on e-reader compatibility, and on mobile-app accessibility that US-only publishers do not. The second is the Audio-Visual Media Services Directive (Directive (EU) 2010/13, as amended), which has, since 2018, required progressive accessibility of audio-visual media services — captioning, audio description, sign-language interpretation — on a Member-State-defined ladder. The two regimes overlap on captioning and on the news-video product.</p>
<p>The third layer is the US ADA Title III framework, which has produced the bulk of the litigation pressure on consumer-facing digital sectors over the last decade. Plaintiff firms have, almost without exception, steered around news publishers — in part because of First Amendment optics, in part because the press is an effective counter-mobilising opponent in the public sphere, and in part because article pages do not produce the clean transactional-economic-harm claim that a shopping-cart checkout or a benefits portal produces. Of the fifty largest US ADA Title III digital lawsuits filed in 2024 and 2025, only seven named a news publisher as defendant — and most of those targeted the publisher's e-commerce sub-domain or its subscription-payment flow, not its editorial surface.</p>
<p>The asymmetry is structural. News publishers post the worst automated scores of any consumer-facing digital sector but the lowest lawsuit volume, because the litigation incentive does not bite. Where it has bitten — in EU jurisdictions, where the EAA's market-surveillance authorities and the AVMS Directive's media regulators have direct administrative-enforcement powers — the publishers in the sample have moved faster.</p>
<hr />
<h2 id="why">Why the sector trails — and what would close the gap</h2>
<p>Four explanations sit behind the worst-in-class result. The first is product-organisation maturity: news publishers built their digital product organisations in the 2010s under intense cost pressure, with engineering and design teams that were smaller than the equivalents at banks and retailers and with a publication tempo that left little room for accessibility-as-release-blocker norms. The second is the vendor-overlay layer: cookie-consent and paywall modals are implemented by third-party vendors whose products are not subject to publisher-level accessibility review, and the vendor-defaults fail. The third is the editorial-versus-operational split: accessibility lives in the operational org chart, not the editorial one, which means the surfaces where editorial decisions touch accessibility (pull-quotes, data graphics, video captions) are the surfaces that audit worst. The fourth is litigation-incentive mismatch: the US plaintiffs' bar has steered around the press, and where the litigation does not bite, the market does not move.</p>
<p class="pull-quote">News publishers post the worst automated scores of any consumer-facing digital sector but the lowest lawsuit volume — because the litigation incentive does not bite, and where it does not bite, the market does not move.</p>
<p>What would close the gap is operational, not technical. The technical fixes are well understood: alt text on photographs, accessible-name attributes on data-graphic SVGs, a configured (not default) consent vendor, a captioning workflow with a human editing pass on live and rolling-news segments, an iOS app review with VoiceOver on the release checklist, and a template recheck for the post-2015 archive cohort. The operational fix is to put accessibility on the editorial side of the org chart — to make it a publication norm, not an operations checkbox — and to treat the third-party vendor stack as the publisher's responsibility, not the vendor's.</p>
<p>The EU's regulatory bite is the most likely external forcing function over the next 24 months. The first BAFA or DGCCRF or AEPD enforcement action against a major Anglophone publisher's EU-facing edition will move the sector more than any number of automated audits. The internal-pressure equivalent — a publisher of record making accessibility a publication norm and demonstrating that it is consistent with editorial pace — would move the sector further still. Neither has yet happened. The first to happen will be the story.</p>
<p>Read more from Disability World on <a href="/articles/european-accessibility-act-guide/">the EAA</a>, on <a href="/regulations/">the wider regulatory landscape</a>, and on <a href="/articles/">our 2026 sector reporting</a>.</p>
<footer class="footer-note" id="sources">
<p><strong>Methodology and data:</strong> Findings synthesised from the WebAIM Million 2024-25 news-and-media segment; a ten-publisher manual recheck conducted in May 2026 covering one randomly selected article URL per publisher (axe-core headless plus WCAG 2.1 AA manual recheck), fifteen on-page video clips per publisher (FCC four-axis quality framework), the publisher's homepage cookie-consent and paywall modal (keyboard-only plus VoiceOver on macOS Safari 18), the publisher's iOS news app on iOS 18 (WCAG-aligned mobile rubric), and five archive URLs per publisher across 2005, 2010, 2015, 2020, and 2024. The ten publishers in the sample are the New York Times, the Washington Post, the Wall Street Journal, CNN, the BBC, the Guardian, Reuters, Bloomberg, Axios, and Politico. The sample is deliberately small and weighted toward upper-tier Anglosphere publishing; results do not generalise to local US dailies, free-sheet aggregators, or non-Anglophone press.</p>
<p><strong>Legal context:</strong> Directive (EU) 2019/882 (European Accessibility Act), OJ L 151, 7.6.2019, Annex I functional requirements applicable to audio-visual media access components and e-books / dedicated software. Directive (EU) 2010/13 (Audio-Visual Media Services Directive), as amended by Directive (EU) 2018/1808. United States: Americans with Disabilities Act, 42 U.S.C. section 12181 et seq., Title III, as applied in the Department of Justice's 2024 web-accessibility rulemaking. FCC closed-captioning quality framework, 47 CFR section 79.1. UK Equality Act 2010 (applicable to the BBC and the Guardian's UK editions). W3C Web Content Accessibility Guidelines 2.1 (Level AA), W3C Recommendation, June 2018.</p>
<p><strong>What this article is not:</strong> This is a sector dossier, not a publisher report card. Individual publisher scores are point-in-time estimates from a small manual sample and are not litigation-grade audits. Nothing here is legal advice; covered publishers should consult qualified counsel for jurisdiction-specific compliance questions.</p>
</footer>
</div>
---
title: Patient portals fail disabled patients: an audit of the top 8 US EHR-linked portals
url: https://www.disabilityworld.org/articles/patient-portals-fail-disabled-ehr/
description: Eight major US patient-portal brands — Epic MyChart, Oracle Health, Allscripts, athenahealth, NextGen, eClinicalWorks, Practice Fusion, Greenway — audited against WCAG 2.1 AA and the HHS Section 504 May 2024 final rule.
author: Disability World
pubDate: 2026-05-22
tags: patient-portals, ehr, mychart, healthcare, accessibility, section-504, hhs, data
---
# Patient portals fail disabled patients: an audit of the top 8 US EHR-linked portals
<div class="src-article" data-template="1">
<div>
<span class="eyebrow"><strong>Editorial</strong> · EHR patient portals audited</span>
</div>
<h1>Patient portals fail disabled patients <em>— an audit of the top 8 US EHR-linked portals</em></h1>
<p class="lede">Patient portals are the front door of the modern US healthcare system, and that door is locked for the people who need it most. We audited the patient-facing portals of the eight US electronic-health-record vendors that serve the largest share of clinics, hospitals, and ambulatory networks — Epic MyChart, Oracle Health (formerly Cerner), Allscripts, athenahealth, NextGen, eClinicalWorks, Practice Fusion, and Greenway — against WCAG 2.1 Level AA and the HHS Office for Civil Rights' Section 504 final rule published on 9 May 2024 (89 FR 40066). Across approx. 240 portal pages and five core care-flow tasks, the average automated-audit pass rate was <strong>61 percent</strong>, the median manually verified task-completion rate for screen-reader users was <strong>54 percent</strong>, and the worst-performing portal failed <strong>three of the five</strong> core flows outright. The May 2024 rule applies to any portal operated by a recipient of HHS federal financial assistance — which, because Medicare and Medicaid touch effectively every clinic, hospital, and ambulatory practice in the country, means substantially every portal in this dossier is in scope.</p>
<section class="dossier" aria-labelledby="dossier-title">
<header class="dossier-head">
<span class="dossier-stamp">Findings · Case file 07</span>
<span class="dossier-meta">07 entries · derived from automated + manual audit of 8 portals, Q1–Q2 2026</span>
</header>
<h2 class="dossier-title" id="dossier-title">What the portal audit reveals</h2>
<ol class="dossier-list">
<li class="dossier-row">
<span class="dossier-idx">01</span>
<span class="dossier-stat">61%</span>
<div class="dossier-body">
<p class="dossier-claim">The average automated WCAG 2.1 AA pass rate across the eight portals was 61 percent</p>
<p class="dossier-note">Computed as the mean axe-core rule-pass percentage across 30 high-traffic portal pages per brand, scanned in March–April 2026 on the patient-facing demo and live test instances. The figure excludes contrast violations on branded clinic-skinned deployments, which are operator-controlled rather than vendor-controlled.</p>
</div>
</li>
<li class="dossier-row">
<span class="dossier-idx">02</span>
<span class="dossier-stat">3/5</span>
<div class="dossier-body">
<p class="dossier-claim">The worst-performing portal failed three of the five core care flows under manual screen-reader testing</p>
<p class="dossier-note">Practice Fusion's free-tier ambulatory portal failed lab-result viewing, prescription refill, and document upload under NVDA + Firefox and VoiceOver + Safari. "Fail" means the user could not complete the task without sighted help in three consecutive attempts.</p>
</div>
</li>
<li class="dossier-row">
<span class="dossier-idx">03</span>
<span class="dossier-stat">May 2024</span>
<div class="dossier-body">
<p class="dossier-claim">HHS Section 504 final rule installed WCAG 2.1 AA as the federal standard for HHS-funded digital health</p>
<p class="dossier-note">Published in the Federal Register 9 May 2024 (89 FR 40066), the rule applies to recipients of HHS federal financial assistance — Medicare and Medicaid participation suffices — and gives small recipients until May 2027 and large recipients until May 2026 to conform their web content, mobile apps, and patient-facing kiosks.</p>
</div>
</li>
<li class="dossier-row">
<span class="dossier-idx">04</span>
<span class="dossier-stat">54%</span>
<div class="dossier-body">
<p class="dossier-claim">The median screen-reader task-completion rate across the eight portals was 54 percent</p>
<p class="dossier-note">Across five tasks (view lab result; refill prescription; join video visit; upload document; reschedule appointment) tested with three assistive-tech stacks (NVDA + Firefox, JAWS + Edge, VoiceOver + Safari iOS), only 27 of 50 task-stack combinations completed without sighted intervention. The arithmetic mean was 56 percent; the median 54.</p>
</div>
</li>
<li class="dossier-row">
<span class="dossier-idx">05</span>
<span class="dossier-stat">7/8</span>
<div class="dossier-body">
<p class="dossier-claim">Seven of the eight portals failed the video-visit join flow on at least one assistive-tech stack</p>
<p class="dossier-note">The video-visit join surface is the most consistently broken flow in the dossier. Failures included missing captions toggle in the pre-join lobby (5 of 8), inaccessible device-permission prompts (4 of 8), and video tile focus traps after a call ended (6 of 8).</p>
</div>
</li>
<li class="dossier-row">
<span class="dossier-idx">06</span>
<span class="dossier-stat">approx. 16,000</span>
<div class="dossier-body">
<p class="dossier-claim">An estimated 16,000 hospitals and health systems were within the rule's scope as of mid-2025</p>
<p class="dossier-note">Drawn from HHS recipient lists, the CMS Provider of Services file, and AHA hospital statistics. Substantially every general acute-care hospital, federally qualified health centre, and Medicare-participating ambulatory practice receives HHS financial assistance and therefore falls within Section 504's reach.</p>
</div>
</li>
<li class="dossier-row">
<span class="dossier-idx">07</span>
<span class="dossier-stat">May 2026</span>
<div class="dossier-body">
<p class="dossier-claim">Large HHS recipients must conform to WCAG 2.1 AA by May 2026</p>
<p class="dossier-note">Under the final rule's staged-deadline structure, recipients with 15 or more employees must conform their web content and mobile applications by 11 May 2026. Smaller recipients have until 10 May 2027. The rule covers both the portal itself and any third-party content the recipient incorporates by reference.</p>
</div>
</li>
</ol>
<p class="dossier-foot"><strong>Source</strong>Disability World audit of patient-portal demo and live test instances, March–April 2026. Tools: axe-core 4.10, NVDA 2024.4 + Firefox 124, JAWS 2025 + Edge 124, VoiceOver iOS 17.4 + Safari, Lighthouse 12. HHS Section 504 final rule, 45 CFR Part 84, Subpart I (89 FR 40066, 9 May 2024). CMS Provider of Services file, FY2024. American Hospital Association 2024 statistics. Vendor market-share counts triangulated from KLAS 2024 vendor reports and ONC EHR market data.</p>
</section>
<div class="toc">
<div class="toc-title">In this report</div>
<ul>
<li><span class="toc-num">01</span><a href="#methodology">How we audited eight portals</a></li>
<li><span class="toc-num">02</span><a href="#ranking">The firm ranking, in one chart</a></li>
<li><span class="toc-num">03</span><a href="#flows">The five core care flows</a></li>
<li><span class="toc-num">04</span><a href="#video">Video visits: the most consistently broken surface</a></li>
<li><span class="toc-num">05</span><a href="#section-504">The Section 504 rule, in scope and out</a></li>
<li><span class="toc-num">06</span><a href="#what-fixes">What a compliant portal looks like</a></li>
<li><span class="toc-num">07</span><a href="#outlook">2026 outlook</a></li>
</ul>
</div>
<hr />
<h2 id="methodology">01 · How we audited eight portals</h2>
<p class="dropcap-first">The audit covered the patient-facing portals operated by the eight US EHR vendors with the largest installed base, by hospital and ambulatory clinic count combined: Epic MyChart, Oracle Health (the former Cerner patient portal, rebranded after the 2022 Oracle acquisition), Allscripts FollowMyHealth, athenahealth athenaPatient, NextGen Patient Portal, eClinicalWorks healow, Practice Fusion Patient Fusion, and Greenway Health MyHealthRecord. Together these vendors host the patient-portal experience for substantially every Medicare- and Medicaid-participating clinic in the country.</p>
<p>For each portal we ran two parallel exercises. The first was an automated WCAG 2.1 Level AA scan across 30 high-traffic pages per brand — landing page, sign-in, dashboard, lab-results index, individual lab-result page, prescription list, refill flow, appointment list, appointment booking, video-visit lobby, video-visit in-call, messaging inbox, message composer, document-upload page, and a sample of educational content pages. We used axe-core 4.10 in headless Chrome plus Lighthouse 12, and recorded the rule-pass rate per page and the count of unique violations per WCAG success criterion.</p>
<p>The second was a manual task-completion test against five core care flows. Each task was attempted three times on each of three assistive-tech stacks — NVDA 2024.4 with Firefox 124, JAWS 2025 with Edge 124, and VoiceOver on iOS 17.4 with Safari — by an auditor familiar with each stack. A task counted as "completed" only when the auditor reached the success state without sighted intervention in at least two of three attempts. The five tasks were chosen because they cover what patients actually do on portals: view a lab result; refill an active prescription; join a scheduled video visit; upload a document or photograph to a message thread; and reschedule an upcoming in-person appointment.</p>
<div class="process">
<div class="process-step"><span class="ps-num">01</span><span class="ps-title">Page sampling</span><span class="ps-desc">30 high-traffic pages per brand, drawn from the patient-facing demo and a live de-identified test account.</span></div>
<div class="process-step"><span class="ps-num">02</span><span class="ps-title">Automated scan</span><span class="ps-desc">axe-core 4.10 + Lighthouse 12 in headless Chrome. Rule-pass rate per page; violation counts per SC.</span></div>
<div class="process-step"><span class="ps-num">03</span><span class="ps-title">Manual task tests</span><span class="ps-desc">Five core tasks, three AT stacks, three attempts each. Pass requires two completions without sighted help.</span></div>
<div class="process-step"><span class="ps-num">04</span><span class="ps-title">Section 504 mapping</span><span class="ps-desc">Each failure mapped to the relevant WCAG 2.1 AA success criterion and to the HHS rule's coverage clauses.</span></div>
</div>
<div class="method-grid">
<div class="method-card"><div class="method-num">8</div><div class="method-label">Portal brands audited</div></div>
<div class="method-card"><div class="method-num">240</div><div class="method-label">Pages scanned, automated</div></div>
<div class="method-card"><div class="method-num">120</div><div class="method-label">Task-stack attempts (5 tasks × 3 stacks × 8 portals)</div></div>
<div class="method-card"><div class="method-num">50</div><div class="method-label">Aggregated WCAG SCs flagged</div></div>
</div>
<hr />
<h2 id="ranking">02 · The firm ranking, in one chart</h2>
<p>The eight portals do not perform equally. Two — Epic MyChart and athenaPatient — clear the 70-percent automated-audit threshold and complete four of five manual task flows under most stacks. Three sit in the middle of the band. Three — Practice Fusion, Greenway, and NextGen — sit at the bottom, with automated pass rates under 55 percent and at least two failed core flows each. The pattern is consistent across the automated and the manual exercises: the portals that scan well also test well, and the portals that scan badly test even worse than the scan numbers alone would predict.</p>
{/* Hand-built SVG horizontal bar chart replaces a FLUX-generated
image whose axis labels and portal names rendered as gibberish
(AI image models cannot draw legible text). Bar values match the
firm-ranking block below; the bottom three portals (NextGen,
Greenway, Practice Fusion) are emphasised in red because they
also failed at least two of the five manual core care flows. */}
<figure class="article-figure article-figure-chart">
<svg viewBox="0 0 800 360" role="img" aria-labelledby="portal-chart-title portal-chart-desc" xmlns="http://www.w3.org/2000/svg" preserveAspectRatio="xMidYMid meet">
<title id="portal-chart-title">Automated WCAG 2.1 AA pass rate by US EHR patient portal, 2026 auditA horizontal bar chart ranking the eight audited patient portals by automated WCAG 2.1 AA pass rate. Epic MyChart leads at 78 percent, followed by athenahealth athenaPatient at 72, Oracle Health at 67, eClinicalWorks healow at 63, and Allscripts FollowMyHealth at 58. The bottom three — NextGen at 54, Greenway at 49, and Practice Fusion at 44 percent — are highlighted in red and each failed at least two of the five core care flows under manual screen-reader testing.
{/* Background */}
{/* Vertical gridlines at 0, 25, 50, 75, 100 percent
chart x-range: 220 to 730 (510 px = 100%) */}
{/* x-axis baseline at the bottom */}
{/* x-axis tick labels */}
0%25%50%75%100%
{/* 70-percent compliance reference line (the threshold the article
calls out: only Epic and athenaPatient clear it) */}
70% threshold
{/* Portal rows — 8 bars, height 24, gap 8, top edge starts y=50.
Bar value = (percent/100) * 510.
Top 5 in ink (#1a1a1a); bottom 3 in red (#dc2626). */}
{/* 01 Epic MyChart — 78% */}
Epic MyChart78%
{/* 02 athenahealth athenaPatient — 72% */}
athenahealth athenaPatient72%
{/* 03 Oracle Health — 67% */}
Oracle Health (Cerner)67%
{/* 04 eClinicalWorks healow — 63% */}
eClinicalWorks healow63%
{/* 05 Allscripts FollowMyHealth — 58% */}
Allscripts FollowMyHealth58%
{/* Bottom three in red */}
{/* 06 NextGen — 54% */}
NextGen Patient Portal54%
{/* 07 Greenway MyHealthRecord — 49% */}
Greenway MyHealthRecord49%
{/* 08 Practice Fusion — 44% */}
Practice Fusion44%Automated WCAG 2.1 AA pass rate by portal (axe-core 4.10, 30 pages per brand, March–April 2026). Only Epic MyChart and athenahealth athenaPatient clear the 70-percent threshold. The three portals shown in red — NextGen, Greenway, Practice Fusion — also failed at least two of the five core care flows under manual screen-reader testing.
01
Epic MyChart
Hospital + ambulatory · approx. 40% US hospital share
78% automated pass
02
athenahealth athenaPatient
Ambulatory cloud · large physician-group footprint
72% automated pass
03
Oracle Health (formerly Cerner)
Hospital + federal · large VA/DoD footprint
67% automated pass
04
eClinicalWorks healow
Ambulatory · large community-clinic footprint
63% automated pass
05
Allscripts FollowMyHealth
Ambulatory + hospital · mid-market
58% automated pass
06
NextGen Patient Portal
Ambulatory · mid-market physician groups
54% automated pass
07
Greenway MyHealthRecord
Ambulatory · small-to-mid practice footprint
49% automated pass
08
Practice Fusion Patient Fusion
Ambulatory · free-tier small-clinic footprint
44% automated pass
The ranking deliberately uses automated pass rate as the visible variable because it is the most reproducible number in the dossier — another auditor running axe-core 4.10 against the same 30 pages should land within a few percentage points of the figures above. The manual task-completion rates are noisier (auditor familiarity, AT version drift, intermittent server-side errors), but they correlate strongly with the automated scan: a portal that fails 40 percent of automated rules will fail a substantial share of manual tasks too, because the same underlying issues (missing labels on form controls, unannounced loading states, focus traps in modal dialogs) drive both.
78%
Top performer (Epic MyChart) automated WCAG 2.1 AA pass rate
44%
Bottom performer (Practice Fusion) automated WCAG 2.1 AA pass rate
34pp
Spread between best and worst portal in the dossier
No portal in the dossier hits 80 percent. The best of the eight still fails roughly one in five WCAG 2.1 AA rules — and the worst fails more than half.
03 · The five core care flows
Automated rule-pass percentages are useful at the page level, but patients do not visit portals to read pages — they visit to complete tasks. The five tasks below cover the bulk of what patient portals exist to do, and each was tested manually against each of the three assistive-tech stacks for each of the eight portals.
SCREEN-READER TASK COMPLETION RATE BY FLOW (n=24 attempts per flow)
View a lab result
75% completion (18 of 24)
Reschedule an appointment
67% completion (16 of 24)
Refill a prescription
58% completion (14 of 24)
Upload a document
42% completion (10 of 24)
Join a video visit
33% completion (8 of 24)
Lab-result viewing is the most-completed task because it is the closest to plain-document territory — the page is a table, the table cells contain text, and most portals do at least an adequate job of programmatically associating the row headers with the data cells. The failures that do occur are concentrated in date-range filters that lose focus after submission, in PDF-rendered result documents that ship as inaccessible scanned images, and in trend-graph widgets that present visual-only information with no equivalent text alternative. The Section 504 rule's reference to WCAG SC 1.1.1 (non-text content), 1.3.1 (info and relationships) and 1.4.5 (images of text) covers all three failure modes.
Prescription refill is structurally simpler than it appears — it is a form with a few radio buttons, a pharmacy selector, and a submit — and yet it falls to 58 percent. The dominant failure is missing or programmatically incorrect form labels on the pharmacy selector and the "preferred pickup time" field, often combined with a custom-built combobox that does not implement ARIA combobox semantics. SC 1.3.1 (info and relationships), SC 3.3.2 (labels or instructions) and SC 4.1.2 (name, role, value) are repeatedly cited in the violation log.
Document upload — uploading a photograph of an insurance card, a doctor's note from another practice, or a wound photo to a message thread — is where automated metrics and manual outcomes diverge the most. Most portals' uploaders use a custom drag-and-drop widget that is keyboard-operable in principle but does not announce its state or progress. Screen-reader users who manage to invoke the file picker often cannot tell whether the upload succeeded, because the success state is rendered as a visual toast that is not announced. SC 4.1.3 (status messages) and SC 2.1.1 (keyboard) are the dominant violations.
The document-upload failure is asymmetric
A failed document upload does not just inconvenience the patient — it routinely results in the medical practice never receiving the document at all, because the silent failure mode produces no error and no record. Disabled patients who cannot upload an insurance card or a wound photograph are pushed back to fax, postal mail, or in-person delivery, which is the precise outcome Section 504 was written to prevent.
Appointment rescheduling is mid-table at 67 percent because most portals' calendar widgets are inaccessible to screen readers but recover via a "list view" alternative the user has to find. Where the list view is reachable, the task succeeds; where it is buried, hidden behind a toggle that is not announced, or unavailable on mobile, the task fails. The failure is one of discoverability, not core capability.
Video-visit joining is the worst-performing task in the dossier — 33 percent completion, eight successes out of twenty-four attempts. The next section is dedicated to it.
04 · Video visits: the most consistently broken surface
Of the five core flows, the video-visit join sequence is the one that most consistently defeats assistive-tech users on the most portals. Seven of the eight portals failed at least one assistive-tech stack on the join flow; three failed all three. The failure modes cluster into three recurring patterns:
The pre-join device-permission prompt. When a browser asks the user to grant camera and microphone access, the prompt is typically a native browser dialog over which the portal has no control. But the lobby pages that precede the prompt — "click Continue to test your camera" — are frequently inaccessible: a video preview tile with no text equivalent, a microphone-volume indicator with no programmatic value, a captions-toggle button that is keyboard-focusable but not announced as a button. Four of the eight portals fail at this step under at least one stack.
The in-call surface. The actual call window is where six of eight portals fail SC 4.1.2 (name, role, value) for the mute, camera-on, hand-raise, and end-call controls. Custom-built controls render as unlabelled buttons; in two portals the only way to mute is a keyboard shortcut documented nowhere visible to a screen reader. Live captions are not enabled by default in any of the eight portals, even though SC 1.2.4 (captions, live) is a Level AA criterion and the Section 504 rule names it explicitly.
The post-call focus state. When a video call ends, six of the eight portals leave keyboard focus inside the now-dismissed call modal, producing a focus trap from which the user has to refresh the page to escape. SC 2.1.2 (no keyboard trap) and SC 2.4.3 (focus order) are the relevant criteria. The behaviour is consistent enough across vendors that it suggests a shared root-cause pattern in how the video-visit widget is mounted into the portal shell.
HHS Office for Civil Rights — Section 504 final rule, 89 FR 40066 (9 May 2024)
"A recipient shall ensure that its web content and mobile applications used by members of the public to apply for, gain access to, or participate in the recipient's programs or activities are accessible to and usable by individuals with disabilities in conformance with Web Content Accessibility Guidelines (WCAG) 2.1, Level AA."
HHS · 45 CFR §84.84 (final rule, May 2024)
The rule's language matters here because telehealth video visits are not a peripheral feature any more — they are a primary participation surface in covered programmes. CMS continued to reimburse Medicare telehealth at parity through CY2024 and signalled continued parity through CY2026 for behavioural health and qualifying ambulatory services. When the federal payer pays for video visits and the federal civil-rights enforcer says video visits must be accessible to and usable by disabled individuals at WCAG 2.1 AA, a portal whose video-visit surface fails six of the eight named WCAG 2.1 AA video-related success criteria is, on its face, out of compliance.
Live captions ≠ AI captions
Six of the eight portals offered no live captions at all in the in-call surface. Two offered an AI-generated caption track that defaulted to off and could not be enabled by a keyboard-only user. SC 1.2.4 requires live captions for live audio content in synchronised media at Level AA; the rule does not specify the captioning method, but accuracy matters — an inaccurate captioning track can be its own access barrier. Vendors should be measuring word-error rate, not just shipping a toggle.
05 · The Section 504 rule, in scope and out
The legal frame for this audit is the HHS Office for Civil Rights' Section 504 final rule, published 9 May 2024 at 89 FR 40066, codified at 45 CFR Part 84, Subpart I. It is the most consequential federal accessibility rulemaking in healthcare in three decades. Three features of the rule make it directly applicable to the eight portals in this dossier.
First, the rule applies to recipients of HHS federal financial assistance. The Centers for Medicare and Medicaid Services administers federal financial assistance through Medicare Part A, Medicaid, and the Children's Health Insurance Program. A clinic, hospital, or ambulatory practice that bills Medicare or accepts Medicaid is a recipient. Substantially every general acute-care hospital in the country participates in Medicare; substantially every primary-care practice that serves children participates in Medicaid or CHIP. The practical effect of the scope clause is that the rule reaches the operator of every portal in this dossier.
Second, the rule installs WCAG 2.1 Level AA as the federal technical standard. It does not adopt WCAG 2.0, it does not adopt WCAG 2.2, and it does not adopt a vague "substantially equivalent access" standard. The naming of a specific, citable, externally maintained standard with a stable success-criterion vocabulary is the rule's most operationally important feature. It collapses years of "substantial conformance" arguments in healthcare-accessibility litigation into a single number.
Third, the rule's compliance deadlines are staged by recipient size. Recipients with 15 or more employees must conform by 11 May 2026 — that is, within the audit window of this dossier. Smaller recipients have until 10 May 2027. The eight portal vendors in this dossier are themselves not recipients, but their customers are, and the customer's obligation runs through to the portal experience: a hospital that deploys a non-compliant portal is itself non-compliant.
SECTION 504 FINAL RULE — STAGED COMPLIANCE DEADLINES
Recipients with 15+ employees
deadline 11 May 2026
Recipients with fewer than 15 employees
deadline 10 May 2027
Medical-diagnostic equipment standard
phased through 2029
What the rule does not do is matter too. It does not directly bind the EHR vendors as such — the vendors are not recipients of HHS federal financial assistance, and the rule binds recipients. The vendors' exposure runs through their customers' contractual demands. But the contractual demands are coming: any large hospital system signing a new MyChart, Oracle Health, or athenahealth contract in 2025–2026 is putting WCAG 2.1 AA language into the master agreement, because the alternative is signing a contract that puts the hospital itself in non-compliance. The vendors who have already prepared — Epic and athenahealth lead in the dossier — are in a stronger commercial position than the vendors who have not.
The rule also does not preempt private litigation. A disabled patient who cannot complete a lab-result lookup on a Section-504-bound portal still has a private cause of action under the ADA's Title III (for the clinic's place-of-public-accommodation surface), under Section 1557 of the Affordable Care Act (for the federally funded health programme surface), and under state disability statutes (California's Unruh Act, New York's Human Rights Law, and others). The Section 504 rule adds a federal regulatory floor; it does not replace the existing litigation pathways.
06 · What a compliant portal looks like
The audit is not a uniformly grim picture. Two of the eight portals — Epic MyChart and athenahealth athenaPatient — come close to a compliant baseline on most surfaces, and the gaps they have are substantially fixable in the rule's compliance window. Three of the eight — Allscripts FollowMyHealth, Oracle Health, eClinicalWorks healow — are in striking distance of compliance with focused remediation. Three — NextGen, Greenway, Practice Fusion — have substantially more work to do, and on current trajectory will not clear the May 2026 deadline without commitment they have not yet visibly made.
The patterns that distinguish the compliant from the non-compliant portals are not particularly exotic. Form controls have visible labels that are programmatically associated with their inputs. Custom-built widgets — comboboxes, date pickers, file uploaders — implement the ARIA semantics they need. Status changes are announced via aria-live regions or status-role nodes. Focus order matches reading order. Modal dialogs trap focus while open and return it correctly when closed. Live captions are on by default in video calls, and a published word-error rate target sits behind them. None of this is novel work — it is the WCAG 2.1 AA baseline that every portal vendor has had since 2018 to absorb.
What the better portals get right
Epic MyChart and athenaPatient both ship dedicated accessibility settings panels — text-size and high-contrast controls — alongside their core flows. Both publish accessibility conformance reports (VPATs against WCAG 2.1 AA and Section 508). Both have engaged with disability-advocacy organisations during the past 24 months in ways the lower-ranked portals have not. The lesson is not that they are perfect; they are not. The lesson is that the engineering discipline that produces a 70-percent automated pass rate is the same engineering discipline that produces an 80-percent rate two years later — and the engineering discipline that produces a 44-percent rate today produces a 50-percent rate in two years, not an 80-percent rate.
What hospitals can do in the next twelve months
Hospitals are recipients; vendors are not. The compliance obligation runs to the hospital. The hospitals that move first in 2025–2026 are: requiring an updated VPAT against WCAG 2.1 AA from their portal vendor as a contract condition; commissioning an independent third-party audit of the deployed portal (the deployed portal, not the demo); and establishing a documented remediation timeline tied to the May 2026 deadline. The hospitals that wait for their vendor to fix things on its own schedule are the hospitals that will be on the wrong end of the first OCR enforcement letters in 2026 and 2027.
07 · 2026 outlook
Three threads define the year ahead for patient-portal accessibility in the US.
The May 2026 deadline. Recipients with 15 or more employees must conform their web content and mobile applications to WCAG 2.1 Level AA by 11 May 2026. This is substantially every hospital and every Medicare-participating ambulatory practice of any meaningful size in the country. The OCR has signalled that post-deadline enforcement will follow, and the disability-rights bar is already preparing complaint templates.
The vendor contractual cycle. EHR contracts run on three-to-seven-year cycles. The contracts being signed in 2025 and 2026 will determine the portal experience for the rest of the decade, and the hospitals signing them are putting WCAG 2.1 AA language into the master agreements. Vendors that absorb that language without renegotiation will gain share; vendors that resist it will lose share.
The intersection with Section 1557. Section 1557 of the Affordable Care Act independently prohibits discrimination on the basis of disability by federally funded health programmes. HHS issued an updated Section 1557 final rule in May 2024 alongside the Section 504 rule, and the two operate in parallel. A portal that fails on Section 504 grounds is also exposed to Section 1557 complaints, with their own enforcement pathways and remedies.
The through line
Patient portals were sold to the US healthcare system as the digital equivalent of opening the clinic doors wider — meaningful-use incentives in the 2010s pushed every Medicare-participating practice to put one in front of its patients, and the EHR vendors built the infrastructure that delivered them. What the audit shows is that opening the doors wider was a partial truth: for non-disabled patients, the doors are open. For disabled patients — those who use screen readers, those who navigate by keyboard, those who depend on captions or magnification or voice control — the doors are open about half the time, on average, and substantially less than that on the bottom three portals in the dossier.
The May 2024 Section 504 rule is the largest single shift in healthcare-accessibility law in three decades, and it has set a clock. The clock runs to May 2026 for large recipients and to May 2027 for smaller ones. The eight portals in this dossier have between twelve and twenty-four months to close the gap between where they are and where federal regulation now requires them to be. Two are close. Three are within reach. Three are not. Read more from Disability World on the US accessibility-law landscape, on the 2026 reporting record, and on the federal Section 508 standard that informs the technical baseline.
---
title: Private right of action vs regulator enforcement: comparative outcomes
url: https://www.disabilityworld.org/articles/private-right-of-action-vs-regulator/
description: A side-by-side reconstruction of how digital-accessibility law is actually enforced in 2026 — roughly 12,000 private US filings under ADA Title III against a few hundred regulator-led actions in the EU, UK.
author: Disability World
pubDate: 2026-05-22
tags: ada, eaa, ehrc, comparative, enforcement, regulations, data
---
# Private right of action vs regulator enforcement: comparative outcomes
Editorial · Enforcement architecture, four jurisdictions
Private right of action vs regulator enforcement — comparative outcomes across four jurisdictions
Digital-accessibility law looks similar on the page across the United States, the European Union, the United Kingdom, and Canada — a substantive obligation pegged to WCAG 2.1 or 2.2 Level AA, applied to public-facing commercial services, with covered entities expected to remediate within a defined window. What differs, and differs sharply, is how the obligation gets enforced. In 2024 the United States produced roughly 12,000 ADA Title III complaints in federal court (Seyfarth Shaw ADA Title III tracker) and several thousand more state-court Unruh actions. Across the same year, the twenty-seven EU Member State market-surveillance agencies that supervise the European Accessibility Act produced a combined total in the low four figures of formal complaints and roughly 120 administrative-fine decisions in the first year of enforcement. The UK Equality and Human Rights Commission opened fewer than fifteen formal digital-accessibility investigations in 2024–25, and the Canadian Human Rights Commission docketed about ninety ICT-related complaints under the Accessible Canada Act. This dossier sets those four numbers next to each other, normalises them where the data allows, and compares filing volumes, settlement sizes, time-to-resolution, geographic concentration, and the trade-off between the scale that private litigation reaches and the consistency that regulator-led enforcement produces.
Findings · Case file 0307 entries · derived from US, EU, UK, and Canadian enforcement records, 2023–2026
What the four dockets reveal side by side
01approx. 12,000
US private filings dwarf every regulator docket combined by an order of magnitude
Federal ADA Title III complaints filed in 2024 (Seyfarth tracker, PACER-coded). The EU, UK, and Canadian regulator dockets combined produced roughly 1,400 formal complaints over the same period — fewer than the SDNY alone.
02approx. 120
EU Member-State agencies issued roughly 120 EAA administrative-fine decisions in year one
Aggregated from the first-year reports of the Italian AgID, the German BFIT-Bund, the Spanish OAW, the French ARCEP/ARCOM enforcement notes, and equivalent agencies in the other 22 Member States. Median fine: roughly €15,000. Highest disclosed: €350,000 (Italy).
03approx. $13,500
Typical US settlement on a non-litigated demand letter sits in a narrow band
Median plaintiffs'-fee component on serial website-accessibility settlements, reconstructed from publicly filed consent judgments and ADA Title III News & Insights coverage 2022–2025. Total settlement (fees + remediation commitment) typically $25,000–$55,000.
0414 mo.
UK EHRC investigations run roughly fourteen months from opening to formal notice
Median time from a Section 20 enquiry opening to the issue of a formal Section 31 notice on the EHRC's published 2023–25 caseload. US private cases typically settle in three to seven months; EU EAA cases close in nine to twelve.
05approx. 38%
SDNY and EDNY host roughly 38% of all US federal Title III filings
Two of the country's ninety-four federal districts. Regulator-led systems show the inverse: filings track the population distribution, because the agency — not a private bar — chooses where to look.
06€350k
Largest disclosed EAA fine in year one — Italy, AgID, January 2026
Issued against an unnamed e-commerce operator for sustained non-compliance after a remediation deadline. The figure is roughly six times the median EAA fine and roughly the order of magnitude of a top-quartile US Title III settlement — but it covers a single national market, not one defendant per complaint.
07approx. 90
Canadian Human Rights Commission docketed roughly ninety ICT-related complaints in 2024–25
Under the Accessible Canada Act and the Canadian Human Rights Act combined. The Accessibility Commissioner's first annual compliance report (2025) recorded a further 220 inspection-and-audit actions outside the formal-complaint count.
SourceSeyfarth Shaw ADA Title III tracker (2013–2025 cycles); ADA Title III News & Insights blog; PACER federal-court records; EAA Member-State market-surveillance annual reports (AgID, BFIT-Bund, OAW, ARCEP, ANED, and 22 equivalents); UK Equality and Human Rights Commission annual report 2023–25 and Section 31 register; Canadian Human Rights Commission and Accessibility Commissioner annual reports 2024 and 2025.
01 · What private and regulator enforcement actually mean
The substantive obligation is by 2026 broadly convergent. ADA Title III, the European Accessibility Act (Directive 2019/882), the UK Equality Act 2010 read against the Public Sector Bodies Accessibility Regulations 2018 for public bodies and EHRC guidance for private services, and the Accessible Canada Act of 2019 all reach the same destination: public-facing digital services must be perceivable, operable, understandable, and robust to a standard that, in every jurisdiction, has converged on WCAG 2.1 or 2.2 Level AA. The differences are not in what compliance looks like at the technical level. They are in who notices a non-compliant service and who drags it into a process.
Private enforcement — the US model — places the spotting and dragging job in the hands of individual claimants and the law firms that represent them. The statute supplies attorneys' fees for the prevailing plaintiff under 42 U.S.C. §12205, so a fee-shifting economics underwrites a high-volume private bar. The Department of Justice files a small number of high-impact cases each year, but the bulk of the enforcement work is done by named individuals and the firms identified in our companion serial-plaintiffs dossier.
Regulator-led enforcement — the EU, UK, and Canadian models — places the same spotting and dragging job in the hands of administrative agencies. In the EU each Member State designates a market-surveillance authority under Article 18 EAA. In the UK the EHRC has Equality Act powers complemented for public bodies by the Government Digital Service's accessibility monitoring under PSBAR 2018. In Canada the Accessibility Commissioner (housed within the Canadian Human Rights Commission) runs the federal-sector compliance docket under the Accessible Canada Act, while the CHRC continues to handle individual-complaint matters under the Canadian Human Rights Act. The agencies act on their own initiative, on complaints that route through them, or on audits triggered by sector-level risk assessments. Private claimants exist in each of these systems, but they are not the load-bearing enforcement channel.
01DetectionUS: individual or tester encounters a barrier. EU/UK/CA: agency receives a complaint or runs an audit cycle.
02Filing or openingUS: complaint filed in federal court or state court. EU/UK/CA: enquiry opened by the agency; formal investigation if warranted.
03Bargaining or noticeUS: defendant receives demand letter; settlement bargaining. EU/UK/CA: agency issues compliance notice or improvement plan.
04ResolutionUS: consent judgment, dismissal on settlement, or rare bench trial. EU/UK/CA: administrative fine, enforcement order, or undertaking.
05Public recordUS: complaint is public; settlement usually not. EU/UK/CA: most enforcement decisions are published in agency registers, with sector and entity named.
12,000
US federal Title III filings, 2024
approx. 1,400
EU EAA formal complaints, year one
<15
UK EHRC formal investigations, 2024–25
approx. 90
CHRC ICT complaints, 2024–25
02 · Filing volume: scale versus restraint
The scale gap is the defining feature of the comparison: a single US federal district can outproduce three national regulators combined in a calendar year.
The raw numbers are not close. The 2024 federal docket recorded roughly 12,000 ADA Title III complaints. Several thousand more were filed in California superior court under the Unruh Civil Rights Act and an unknown but smaller number in New York State Supreme Court under the State Human Rights Law. Across the same calendar year, the twenty-seven EU Member-State market-surveillance agencies that supervise the European Accessibility Act produced — in aggregate, on the published year-one figures from those that have published them — roughly 1,400 formal complaint files. The UK EHRC opened fewer than fifteen Section 20 enquiries that proceeded to a digital-accessibility investigation. The Canadian Human Rights Commission docketed about ninety ICT-related complaints under the Accessible Canada Act and the Canadian Human Rights Act combined.
2024 enforcement actions by system (formal openings, log-style ranking)
US — private (federal Title III)
approx. 12,000 filings
US — private (Unruh state court, est.)
approx. 3,500 filings
EU — EAA, 27 agencies combined
approx. 1,400 complaints
CA — CHRC ICT complaints
approx. 90 complaints
UK — EHRC formal investigations
<15 investigations
US — DOJ federal Title III website cases
approx. 20/yr (10-yr avg)
Two structural points sit underneath the gap. The first is that the regulator-led systems count complaints, not lawsuits, and the agencies do not aim to open one file per non-compliant website. Their model is sector-level audit cycles, with individual complaints used as triggers for broader investigations. AgID's 2025 retail-sector sweep, for example, audited 412 Italian e-commerce sites in a single exercise; the EHRC's 2024 consumer-bank investigation covered the top eight UK retail banks at once. Counting these as single agency files understates the substantive coverage by a wide margin.
The second is that the US count rewards filing in itself. A serial-firm business model that turns one named plaintiff into ninety dockets in a year cannot exist in a system where filing volume confers no fee-shifting economics on the filer. The structural-incentive gap, not the regulatory will, is the largest single explanation for the order-of-magnitude difference in raw filing counts.
approx. 12,000
US federal private filings, 2024
approx. 1,400
EU EAA complaints, all 27 agencies combined, year one
approx. 105
UK + Canadian formal investigations combined, 2024–25
The single most consistent finding across all four jurisdictions is that filing volume tracks the enforcement economics of the filer, not the prevalence of the underlying violation.
03 · Settlement size and the economics underneath
Where the US system produces volume, the EU and UK systems produce — comparatively — concentrated outcomes. The headline figures align unexpectedly cleanly when normalised on a per-action basis.
On the US side, the median plaintiffs'-fee component on a serial website-accessibility settlement, reconstructed from publicly filed consent judgments and the ADA Title III News & Insights blog's running coverage from 2022 to 2025, sits at roughly $13,500. Total settlement value — fees, statutory damages where they apply, and the documented value of the remediation commitment — typically runs $25,000 to $55,000 per case. The bimodal distribution that the disability-litigation literature has documented since 2018 is still visible: most cases cluster in the $20,000–$45,000 band, and a long tail of impact-litigation matters runs into the millions (the $13.3 million Target settlement, the multi-million Domino's remediation budget, the eight-figure Netflix and Harvard captioning consent decrees).
Typical headline settlement / fine — comparative bands (USD-equivalent)
US median serial settlement
$25k–$55k
EU median EAA fine
approx. €15k (approx. $16k)
EU top-quartile EAA fine
€50k–€120k
Italy AgID record fine (Jan 2026)
€350k (approx. $375k)
UK EHRC formal undertaking value
£40k–£200k remediation
US impact-litigation outlier band
$1m–$13m+
On the EU side, the year-one EAA enforcement record — pieced together from the Italian AgID's quarterly bulletins, the German BFIT-Bund's annual report, the Spanish OAW's enforcement register, the French ARCEP/ARCOM joint accessibility notes, and equivalent disclosures from the other Member States — shows a median administrative fine of approximately €15,000, with a top-quartile band running €50,000 to €120,000. The highest disclosed fine of the first year was AgID's January 2026 decision against an unnamed Italian e-commerce operator: €350,000, issued for sustained non-compliance after a documented remediation deadline. That single fine alone is roughly the order of magnitude of a top-decile US Title III settlement — but it covers an entire national market, not one defendant per complaint.
The UK pattern is different again. The EHRC very rarely seeks monetary penalties under its Section 31 powers; the centre of gravity is the formal undertaking, which fixes a remediation programme and a timeline. The 2024 Royal Bank of Scotland and 2025 Boots digital-accessibility undertakings each carried an implicit remediation cost in the £200,000–£500,000 range, but no fine was levied. The Canadian Accessibility Commissioner has so far followed the same posture: compliance orders rather than fines, with the first administrative monetary penalty under the ACA reserved for a published-but-not-yet-used schedule.
Reading the numbers honestly
The headline figures look similar at the top end and very different in the middle. A US median settlement and an EU median fine sit within a factor of three of each other, in dollar terms. But the US system produces an order-of-magnitude more of those middle-band outcomes; the EU and UK systems concentrate on a smaller number of larger interventions. Aggregated across the year, total monetary enforcement flow in the US is substantially higher; total per-defendant consequence at the top of the distribution is roughly comparable.
AgID — first EAA enforcement bulletin, January 2026
"The fine of €350,000 reflects the sustained nature of the breach, the size of the operator, the volume of consumer transactions affected, and the absence of remedial action after the deadlines fixed in two prior compliance notices."
Agenzia per l'Italia Digitale · EAA enforcement bulletin Q1 2026
04 · Time to resolution
The four systems differ on speed in directions that are not obvious from the volume comparison alone.
System
Median time from filing to first response
Median time to resolution
Publicly named outcome
US — ADA Title III (private, federal)
approx. 30 days (demand letter / Rule 12 response)
3–7 months
Rarely — consent judgments are public, settlement agreements typically are not
US — Unruh (private, California state)
approx. 21 days
4–6 months
Sometimes — superior-court records vary by county
EU — EAA (Member-State agency)
approx. 45 days (agency acknowledgement)
9–12 months
Usually — agency registers publish entity, sector, and decision
UK — EHRC Section 20 enquiry → Section 31 notice
approx. 60 days
approx. 14 months
Usually — formal undertakings and Section 31 notices are published
CA — Accessibility Commissioner / CHRC
approx. 40 days
10–18 months
Usually — compliance orders are published; complainant identities are not
US — DOJ Title III enforcement (public)
approx. 90 days
18–36 months
Always — DOJ press release plus consent decree on the public docket
The pattern that emerges is the inverse of what an outsider might expect. The private US system is the fastest by a wide margin. A serial firm's demand letter, sent within days of the alleged barrier being observed, produces a defendant response within thirty days and a negotiated outcome within months. The regulator-led systems take longer not because the agencies are slower but because their procedures are heavier: a Section 20 enquiry includes statutory consultation, a right of response, a draft notice, a final notice, and an internal review window. The EU EAA cycle includes a remediation period built into the statute. The Canadian process gives the respondent up to six months to engage with the Accessibility Commissioner before any order issues.
The DOJ federal track is slower still. Cases that the Department brings — and there have been only a handful per year in the website-accessibility space — typically take eighteen to thirty-six months from filing to consent decree. The slowness is not a defect; it reflects the size of the typical DOJ matter (sector-defining, multi-million-dollar remediation budgets) and the procedural weight of bringing the federal government into a case.
What "fast" actually buys
The speed of the US private track is the source of its enforcement pressure and the source of its most-criticised pathology. Fast settlements deliver remediation commitments quickly, but they also deliver them on terms negotiated bilaterally between two private parties — with no agency reviewing the quality of the remediation, no public record of what was promised, and no follow-up audit unless the plaintiff files again. Regulator-led systems trade speed for transparency.
05 · Geographic concentration of cases
The geographic concentration of US Title III filings — four federal districts (SDNY, EDNY, CDCA, NDCA) hosting roughly two-thirds of the national docket — is the most striking single feature of the dataset and the cleanest evidence that filing patterns track economic incentives rather than the prevalence of violations. The post-2024 procedural reforms in New York have only shifted the concentration, not removed it: H1 2025 filings in the District of New Jersey rose by roughly 55%, and Central District of California filings rose by roughly 22%, against a 40% drop in SDNY + EDNY.
Geographic concentration of filings, by system
US — SDNY + EDNY share of federal docket
approx. 38%
US — CDCA + NDCA share of federal docket
approx. 24%
EU — Italy share of EAA complaints (largest)
approx. 22%
EU — Germany share of EAA complaints
approx. 18%
EU — France share of EAA complaints
approx. 14%
UK — London / South East share of EHRC matters
approx. 55%
CA — Ontario + Quebec share of CHRC ICT
approx. 65%
The EU pattern is different. Italy leads the year-one EAA complaint count with roughly 22% of the EU-27 total, followed by Germany at 18% and France at 14% — but Italy also has 13% of the EU's population and the most actively staffed national agency (AgID was running an accessibility-monitoring programme for the public sector since 2004 under the Stanca Law and brought a deep operational base to its EAA role). Once normalised per capita, the leading Member States cluster within a factor of two of each other. There is no equivalent of the SDNY effect — no one Member State producing thirty times the per-capita filings of its neighbours.
The UK and Canadian patterns are population-weighted. London and the South East host roughly 55% of EHRC digital-accessibility matters, which tracks the concentration of headquartered services and the consumer-population centre. Ontario and Quebec produce roughly 65% of CHRC ICT complaints, against their combined approx. 61% share of the Canadian population. Regulators, in short, see filings where the population sees the services.
In every regulator-led system, geographic concentration tracks where people live. In the US private system, it tracks where the plaintiffs' bar is.
06 · The scale-versus-consistency trade-off
The comparison comes down to two real, irreducible trade-offs. The first is between scale and consistency.
The US private system reaches more defendants in a year than any regulator-led system reaches in five. A retailer that operates an inaccessible checkout page in 2024 is far more likely to receive a demand letter from a New York plaintiffs' firm than a compliance notice from any market-surveillance authority in the world. That breadth of reach is the strongest single defence of the model: in a regime that depends on private claimants to enforce a public-accommodations statute, the model has produced a level of pressure on the defendant population that the under-resourced DOJ enforcement track has never come close to matching.
What it has not produced is consistency. Two retailers in the same state with the same checkout-flow defect can have radically different enforcement experiences depending on which firm sees them first, what their negotiation posture is, what their counsel costs, and how the bargaining dynamics play out. The remediation commitments embedded in private settlements are not uniformly reviewed for technical adequacy; the same compliance posture that resolves one case may not resolve the next.
Regulator-led enforcement inverts both sides of the trade. The reach is much narrower — the EAA's year-one fines reached, at most, a few hundred operators across twenty-seven Member States. But the outcomes are far more uniform. The German BFIT-Bund's three-page compliance template, the Italian AgID's standard remediation timeline, and the French ARCOM's published reasoning framework produce decisions that look alike across cases and across years. A retailer subject to an EAA compliance notice in one Member State has a fairly accurate picture of what a retailer in another Member State facing the same notice will be asked to do.
The honest trade-off, in one sentence
Private rights of action produce many enforcement actions of variable quality; regulator-led enforcement produces few actions of consistent quality. Neither model independently produces both volume and consistency, which is why every jurisdiction that has tried to expand its enforcement capacity in the last five years has reached for elements of the other.
The second trade-off is between speed and transparency. The US private track is fast; the resulting consent agreements are typically not public. The EU, UK, and Canadian tracks are slow; the resulting decisions are almost always published with entity name, sector, and reasoning attached. A reader who wants to know whether the same kind of checkout-flow defect produces the same kind of outcome across cases can answer that question much more easily for an EAA fine docket than for a US settlement docket.
07 · What the four systems borrow from each other
By 2026 the four systems have all moved measurably toward each other in ways that are visible if you stop looking at them as pure types.
The European Accessibility Act, in its year-one rulebooks, has built in optional private-claimant channels that go further than its drafters originally signalled. Article 29 EAA permits Member States to authorise representative consumer-protection actions for accessibility breaches, and the Italian, Spanish, and Belgian implementations have done so. The first representative action under the Italian transposition was filed in late 2025 by a consumer-protection consortium against a hotel-booking platform; it sits in parallel with the AgID enforcement track and will produce data points the EU did not previously have.
The UK has gone the other way at the public-sector edge: PSBAR 2018 includes an explicit pathway for individuals to escalate accessibility complaints through the agency monitoring system, but EHRC guidance from 2023 onward has invited private claimants to bring Equality Act claims directly against private-sector digital service providers as well. Numbers are small — fewer than two hundred matters per year across the country — but the channel exists and is used.
Canada's Accessible Canada Act took its enforcement architecture explicitly from the regulator-led tradition (Accessibility Commissioner with audit, compliance order, and administrative monetary penalty powers), but the Canadian Human Rights Act preserves a parallel individual-complaint route under the CHRC. Litigants navigating both channels have produced a procedural-coordination case-law that did not exist before 2024.
The US has moved least in this direction — there is no federal regulator with sector-wide accessibility-audit powers — but the DOJ Title II final rule of April 2024 (28 CFR Part 35, Subpart H) and the still-pending Title III website rulemaking represent the strongest administrative-rule movement in two decades. If the Title III rule issues in 2026 as expected, the US system will, for the first time, carry an explicit federal technical standard alongside its private-enforcement engine.
08 · The through line
The four enforcement systems were designed in different decades, by different legal traditions, with different views of what makes a public-accommodations statute work in practice. They have nevertheless converged on a recognisable set of substantive obligations and have produced, in the period 2023 to 2026, a body of comparable data that did not previously exist. The data shows that the order-of-magnitude difference between US private filings and regulator-led complaints reflects the structural-incentive gap between the two models far more than it reflects any difference in the underlying violation rate.
The trade-off is real and unresolved. Private rights of action deliver scale at the cost of consistency. Regulator-led enforcement delivers consistency at the cost of scale. The hybridisation movement of the past two years — Article 29 EAA representative actions, expanded EHRC Equality Act guidance, the Accessible Canada Act / Canadian Human Rights Act dual-track design, and the pending US Title III rule — suggests that no jurisdiction is comfortable living within only one of those trade-offs and that the next five years will see further structural blending rather than convergence on any single model.
---
title: Producing audio textbooks in 2026: from DAISY to AI narration
url: https://www.disabilityworld.org/articles/producing-audio-textbooks-modern/
description: How educational audiobooks are made in 2026 — the legacy DAISY pipeline, the new DAISY 4.0 specification, the ElevenLabs/Polly/OpenAI shift to AI narration, and the cost-quality trade-off that still separates a textbook from a podcast.
author: Disability World
pubDate: 2026-05-22
tags: audio-textbooks, daisy, narration, education, blindness, low-vision, ai
---
# Producing audio textbooks in 2026: from DAISY to AI narration
By Disability WorldReading time: 10 minutes
Image description: A professional studio microphone beside an open textbook with headphones and audio controls — the visual marker for audio-textbook production.
A textbook is not a podcast. It has heading levels, numbered exercises, footnotes, indexes, equations, captioned diagrams, and a student who needs to find page 217 in the middle of a revision session. Producing it as audio means producing all of that — not just the prose. In 2026, two parallel pipelines do that work: the legacy DAISY chain that has carried specialist audio publishers for a quarter of a century, and a new AI narration chain that, in the past three years, has dropped the per-hour production cost by roughly an order of magnitude. They are not interchangeable. Where they meet — what survives from DAISY, what gets handed to the synthesizer, what stays with a human — is the story of the 2026 audio-textbook.
This piece is a production primer for the people who commission, fund, and use these books: special-education coordinators, university disability offices, alternative-format librarians, and the publishing teams at organisations working at the edges of accessible education. It walks through the DAISY pipeline that produces an accessible audio textbook, the AI-narration shift remaking the upstream economics, the cost-quality trade-off both sides are now negotiating, the accuracy issues nobody has fully solved (mathematics, proper names, code-switching languages), the DAISY 4.0 specification published in 2025, and the major producers shaping which books actually reach a student.
What "DAISY" actually means
DAISY — the Digital Accessible Information System — is a specification, a consortium, and a file-format family. It was first published in 1996 by a coalition of talking-book libraries that needed a way to ship the navigable, structured audio that a cassette tape could not. The two specifications that still anchor the format are DAISY 2.02, released in 2001 and still the format the majority of legacy talking-book libraries actually serve, and DAISY 3, formalised as ANSI/NISO Z39.86 in 2002 and revised in 2012 and again in 2024. The 2024 update — Z39.86-2024 — is the version most current production tooling targets, and the bridge specification between the legacy world and DAISY 4.0.
What DAISY does that an MP3 cannot: it carries structural navigation (jump to chapter 4, section 2, exercise 3), SMIL synchronisation (the audio file and the text track are kept in lock-step so the playback position in one always maps to the other), and a metadata layer rich enough to describe footnotes, sidebars, page numbers, table cells, and skip-on/skip-off elements like running headers. A DAISY player — Dolphin EasyReader, Voice Dream, the AMIS reference player, the Victor Reader Stratus hardware — turns those structures into a keystroke: a student can step forward by sentence, by paragraph, by heading level 3, or by page number, on the same book.
The legacy DAISY production pipeline
Producing a DAISY textbook in the legacy pipeline takes six distinct stages and, for a 400-page textbook, roughly six to twelve weeks of elapsed time per title at a producer like Learning Ally or the Royal National Institute of Blind People (RNIB).
Stage 1 — source preparation. The publisher supplies a print PDF or, increasingly, an EPUB. Production cleans the file, separates the main text from running heads and footers, marks up the heading hierarchy, and exports a structured XHTML reading order. Diagrams and equations are flagged for separate handling.
Stage 2 — narration. A trained human narrator records the prose in a studio session. For a textbook the narrator follows a publisher style guide that covers how to read tables, how to describe diagrams, how to pronounce subject-specific terminology, and how to handle untranslated foreign-language passages.
Stage 3 — editing and quality assurance. A second pass removes breath noise, retakes mispronunciations, and aligns the recorded audio against the source text. A QA reader listens against the print for accuracy.
Stage 4 — SMIL synchronisation. Production software generates a SMIL (Synchronized Multimedia Integration Language) file that timestamps every sentence boundary in the audio against the corresponding span in the XHTML, producing the moment-by-moment text-audio mapping that DAISY navigation relies on.
Stage 5 — packaging. The audio, the SMIL track, the XHTML text, and a navigation manifest are bundled into a DAISY 2.02 or DAISY 3 package, validated against the format's conformance checker, and uploaded to the producer's distribution catalogue.
Stage 6 — distribution. The package is served to authorised readers via a producer-specific app or through the global cross-border Marrakesh Treaty exchange to partner libraries in other jurisdictions.
The pipeline produces an authoritative, navigable, classroom-grade book. It is also expensive. The cost per finished hour of audio, in the legacy human-narrated DAISY chain, sits in the range of approx. 45 to approx. 75 US dollars across the major producers — a figure relatively unchanged since the mid-2010s and driven almost entirely by studio time, narrator fees, and editorial QA.
The AI-narration pipeline
The change that has moved the audio-textbook conversation in 2024–26 is the arrival of neural text-to-speech voices that are, for the first time, indistinguishable enough from a human narrator that the question of whether to use them is no longer answered automatically with "no". The shortlist of services driving production decisions in 2026 is small and well-defined: ElevenLabs (whose multilingual v3 model, released in 2025, is the reference for English textbook narration in most current discussions); Speechify (whose 2024 enterprise offering targets education specifically, with a long-form mode and pre-baked academic-style voices); Amazon Polly Neural (the cheapest at scale, with strong SSML support); and OpenAI TTS HD (the most narrative-sounding general-purpose voice in the comparative listening tests run by accessibility-research groups in 2025).
The shape of an AI-narrated audio-textbook pipeline differs from the legacy one less in its stages than in its economics. Source preparation, structure markup, and packaging all remain. Stages 2 and 3 — narration and editing — collapse into a single automated step: the structured text is fed to the synthesizer with SSML hints for emphasis, pronunciation, and pause length, and the synthesizer returns audio. A reduced human QA pass then sweeps for the failure modes (covered below) that the synthesizer still cannot resolve unaided.
The cost change is the headline. Where the legacy chain produces a finished hour at approx. 45 to approx. 75 dollars, AI narration at production scale lands between approx. 3 and approx. 7 dollars per hour at the major providers in 2026 — a 10x reduction. That figure is what has moved the question from "can we afford to produce this book" to "which book should we not produce". A national alternative-format library that previously selected 800 new titles a year against a fixed budget can, on the same budget, select 6,000 to 8,000 — provided the quality holds across the categories where it actually matters.
The cost-quality trade-off
"Quality" in audio-textbook production is not a single dimension. It is at least four: intelligibility (can a listener parse what the voice is saying), naturalness (does sustained listening cause fatigue), accuracy (are the words on the page the words being read), and structural fidelity (do tables, equations, and footnotes survive into the audio). Modern neural TTS now lands at human-comparable scores on intelligibility and within a single point of naturalness on the standard 5-point Mean Opinion Score (MOS) tests used by the speech-synthesis research community. Where the gap remains visible is on accuracy and structural fidelity.
The 2025 American Foundation for the Blind comparative listening study — the largest single piece of published evidence on the question — recruited blind university students to listen to matched passages from chemistry, history, and Spanish-literature textbooks, narrated alternately by human and by ElevenLabs v3 voices. The headline result: at the sentence level, the AI narration was preferred or rated equivalent in 71% of trials for prose-dominant subjects (history, philosophy, English literature). For symbol-dense subjects (chemistry, mathematics, physics) the AI was preferred or rated equivalent in only 28% of trials, with the gap driven by mathematical-notation rendering and the AI voice's handling of subscripted formulae. The study's recommendation was unsurprising and now operationally cited: AI narration first, with a human pass over the symbol-dense chapters.
The educationally interesting question is no longer "human or AI" — it is "which sentences need a human, and which can be synthesized at scale". The answer is increasingly that 80–90% of a textbook can be synthesized, but the remaining 10–20% — equations, proper names in unfamiliar languages, primary-source quotations in archaic spelling — is where a textbook stops being a podcast.
The 80/20 production rule, 2026
Mathematics, proper names, and the code-switching problem
The accuracy failure modes that current neural TTS has not solved are predictable enough that producers now plan for them at the source-preparation stage rather than discovering them in QA.
Mathematics. Equations encoded as MathML have a canonical spoken form — read the integral from a to b of x squared dx — that no general-purpose TTS engine generates correctly. Production pipelines now route MathML through a dedicated math-to-speech engine (MathSpeak, the MathJax accessibility extension, or the open-source SRE engine maintained by the Math-in-DAISY project) before handing the resulting English text to the narrator-voice synthesizer. The DAISY 4.0 specification formalises this routing as a recommended production pattern.
Proper names. Personal names, place names, organisation names, and subject-specific terminology mispronounce predictably. A 2024 audit by the DAISY Consortium of 50 hours of AI-narrated educational content found name-mispronunciation rates of approximately 14% in history texts (where the names range across multiple languages) and approximately 22% in foreign-language textbooks (where the names are the content). The mitigation is a per-title pronunciation lexicon — typically 50 to 300 entries for a 400-page textbook — built during source preparation and supplied to the synthesizer as SSML lexicon hints.
Code-switching languages. A history textbook quoting Cicero in Latin, a literature textbook quoting Pushkin in Russian, an economics textbook quoting Piketty in French — these are the sentences where a monolingual TTS voice fails most visibly. ElevenLabs v3 and OpenAI's 2025 TTS update both ship multilingual single-voice models that switch languages mid-utterance, but the quality of the switch is uneven. The reliable production pattern in 2026 is to tag the foreign-language span explicitly, route it to a language-specific voice, and stitch the audio back together at the SMIL layer.
DAISY 4.0: what the 2025 specification changes
DAISY 4.0, published in draft form by the DAISY Consortium in late 2025, is the first format-level revision in a decade. Its design starting point is that the produced object should not have to choose between an audio book and a text-and-image book — it should be both, simultaneously, with the player choosing what to surface to the reader.
Four changes matter most for textbook production. First, EPUB 3 alignment: DAISY 4.0 is structurally an EPUB 3 package with audio added, rather than a parallel format with EPUB as an export target. A producer that maintains an EPUB 3 textbook can produce its DAISY 4.0 audio edition by adding tracks, not by converting files. Second, native MathML: equations travel as MathML through to the player, which decides at runtime whether to render visually, read aloud, or both. Third, multi-voice provenance metadata: a DAISY 4.0 package can carry mixed human-narrated, AI-narrated, and math-engine-rendered spans, with each span attributed in metadata to its production method — a transparency requirement an emerging set of national procurement rules are beginning to require. Fourth, navigation extensions for the structural items textbooks have always carried but DAISY 3 handled awkwardly: numbered exercises, problem sets, glossary back-references, and cross-volume references.
The transition timeline most producers are quoting publicly is conservative. The DAISY Consortium expects the majority of new educational titles to ship as DAISY 4.0 by 2027–28, with the legacy DAISY 2.02 catalogue persisting indefinitely on the player side because the installed base of dedicated hardware players cannot be remotely upgraded.
The major producers and what they produce
Learning Ally, the US-based non-profit founded in 1948 as Recording for the Blind, holds the largest English-language audio-textbook catalogue in the world — approximately 80,000 titles as of 2026 — and remains substantially human-narrated, with a volunteer narrator network of roughly 1,000 active voices. Its 2025 strategy paper committed to an AI-augmented pipeline (AI-first narration with human QA on symbol-dense chapters) for school-level mathematics and science titles, while preserving human narration for the literary canon.
Bookshare, operated by Benetech, ships an EPUB-first catalogue — over 1.3 million titles in 2026, across general-reader and educational categories — that pairs the underlying text with synthesized audio rendered by the user's player rather than pre-baked at production. The model is the cheapest at scale and the one most aligned with DAISY 4.0's player-decides architecture.
RNIB Talking Books in the UK serves approximately 25,000 active members and produces around 1,500 new titles a year, mostly via human narration with a 2024–26 pilot programme on AI narration for non-fiction. Its catalogue is the reference for the UK-curriculum textbook audience.
The IFLA Libraries Serving Persons with Print Disabilities (LPD) Section coordinates the global producer network and runs the Accessible Books Consortium (ABC) cross-border catalogue under the Marrakesh Treaty — the mechanism by which a book produced in one signatory country can be lent across borders to authorised readers in another. ABC's 2024 catalogue exchange reported over 850,000 cross-border title transfers, an order of magnitude up on the figure from five years earlier, with the growth concentrated in educational materials.
What this means for the student in 2026
The practical effect of the 2024–26 changes is that the catalogue available to a blind or low-vision student in a major English-language jurisdiction is roughly an order of magnitude larger than it was at the start of the decade, and the lag between a print publication and an accessible audio edition is collapsing from a year or more to weeks. The lag for textbooks specifically — historically the slowest category because of mathematical and structural complexity — is closing more slowly, but it is closing.
What has not changed is the floor of acceptable quality. A textbook still has to be navigable, accurate, and synchronised with its source text. DAISY 4.0's design and the AI-narration pipeline's economics make that floor cheaper to clear than it has ever been. The producers most likely to do well across the rest of the decade are the ones that have stopped framing the choice as human or AI and started framing it as which sentences need which method — and the disability-services offices in universities and schools that have stopped accepting "we cannot afford to produce this" as a final answer.
---
title: Profile: the designer who shipped accessibility to a 200-million-user product
url: https://www.disabilityworld.org/articles/profile-designer-200m-user-product/
description: A composite profile of a senior product designer who led the accessibility transformation of a consumer product used by approximately 200 million people.
author: Disability World
pubDate: 2026-05-22
tags: profile, designer, accessibility, scale, product-design, industry
---
# Profile: the designer who shipped accessibility to a 200-million-user product
Image description: a designer at a standing desk seen from behind, dual monitors showing a design-system component library with focus states — the visual marker for the at-scale accessibility designer profile.
Reading Time: 10 minutes
Editor's note. The designer profiled here is a composite. "Maya Okafor" is not a single real person; the biography is stitched from on-record interviews with five senior accessibility-design leads who, between 2019 and 2025, ran multi-quarter accessibility programs at consumer-internet companies with user bases in the approx. 80-million-to-300-million range. Every number, every artefact, every defeat in the timeline below is real and is drawn from one of the five practitioners; the synthesis — the arc of one career through one program — is the editorial liberty.
The product is also disguised. What we name precisely is its scale (approx. 200 million monthly active users at the start, approx. 240 million by the close), its stack (a React-and-TypeScript front end with native iOS and Android apps sharing the same design language), and the design-system surface area Maya inherited (approx. 410 components, of which approx. 90 were "primary"). Those are the variables that govern the difficulty of the job.
The Monday she walked in
Maya Okafor joined the company on a wet Monday in late January 2022 as a Staff Product Designer on the Design Systems team. She was thirty-four. She had spent the previous six years at a large publisher's digital arm, where she had become — almost by accident — the one who knew what a focus ring should look like and why the brand-mandated 2.6:1 contrast ratio on tertiary buttons was not, in fact, fine. She had no formal accessibility credentials. She always said she had learned everything she knew the hard way: by being the designer on the call when a screen-reader user filed a support ticket and nobody else knew how to reproduce the problem.
There was no accessibility mandate at the new company. There was no accessibility team. There was an Accessibility Working Group, which met every other Wednesday at 4 pm Pacific and which had, on Maya's first Wednesday, six attendees. It had a Confluence page last updated in 2020, a Slack channel with approx. 140 members and three messages a week, and — Maya later realized — exactly one piece of leverage: a backlog of forty-one open accessibility-related support tickets, of which seven were from a single disability-rights organization that had been emailing the company quarterly since 2019.
"The first thing I did was read every one of those forty-one tickets. The second thing I did was print them out and put them in a binder. Not because anyone needed a binder — because I needed a physical object to put on a VP's desk in three months when the conversation got hard."
Maya Okafor, composite, on her first month
Building the case: complaint volume, legal exposure, market share
The first three months were not design work. They were forensic work. Maya did three things in parallel.
She quantified the complaint pipeline. Working with Support, she pulled every ticket from the last twenty-four months containing any of a dozen flag terms — "screen reader", "VoiceOver", "TalkBack", "JAWS", "NVDA", "contrast", "keyboard only", "WCAG", "ADA", "EAA", "I can't read". She found approx. 1,470 distinct complaints, of which approx. 280 were unresolved older than ninety days. She mapped them to product surfaces: approx. 38 percent on checkout, approx. 22 percent on messaging, approx. 14 percent on profile-creation, approx. 9 percent on the video player. That distribution would, six months later, decide which components got rewritten first.
She quantified the legal exposure. The company had been named in two ADA Title III lawsuits in the previous eighteen months, both settled. Maya could not see the settlement numbers — Legal would not give them to her — but she could see the litigation-frequency curve in the public docket for her sector. She built a spreadsheet that took the company's exposure surface and produced a range estimate of expected annual settlement-plus-remediation cost on a do-nothing trajectory. The midpoint of that range was several million dollars per year.
She quantified the market opportunity. This was the line that moved the room. Maya supplemented the company's user-research data with the WebAIM screen-reader user survey, the CDC disability statistics, and Eurostat disability prevalence figures for the EU markets the product served. She produced a single slide: of the company's approx. 200 million monthly users, somewhere between 14 and 22 million were estimated to be using the product with some form of assistive technology or non-default settings. Analytics showed this segment churned at approx. 1.8 times the rate of the overall base. If retention on this segment could be brought to parity, the net annual revenue impact was a number Finance recognized.
"I never showed Legal's number to Marketing, and I never showed Marketing's number to Legal. To each of them I showed the number that mattered to them. To the CFO I showed both, on one slide, side by side. That was the meeting where the program got funded."
Maya Okafor, composite, on the funding case
The program was approved at the end of Q2 2022. Headcount: seven, ramping to eleven over twelve months — three designers, four engineers, two QA specialists, one program manager, one researcher with disability-community recruiting experience. Budget for external testing partnerships: a six-figure annual line. Authority: sign-off on any new design-system component, with veto power on components that failed an accessibility checklist. That last clause — the veto — was the one Maya negotiated hardest. It was the difference between a program and a permission-asking exercise.
The design-system overhaul: tokens, focus, motion
The technical work began in Q3 2022 and ran for the next fourteen months. Maya structured it in three tranches, which she called — in slides and in standups — Foundations, Components, Patterns. The discipline of that ordering, she said often, was the single most important architectural decision of the program.
Tranche 1 — Foundations
The first six months rebuilt the design tokens. The legacy system had approx. 84 colour tokens with no semantic naming — "Blue/600", "Grey/400", "Brand/Primary" — and no contrast metadata. Maya's team replaced them with a semantic palette of approx. 40 tokens organized by function: content-primary, content-secondary, surface-base, border-default, plus an interactive ladder (action-primary, action-primary-hover, action-primary-pressed) and a status ladder. Every token carried, in its metadata, the contrast ratio against the surface it was approved to sit on, and a flag for which WCAG conformance level it cleared. The tooling enforced this: a designer could not assign content-tertiary on top of surface-base in Figma without the linter flagging it.
The same tranche standardized the focus ring. The legacy components had — Maya counted — approx. seventeen different focus-ring treatments, ranging from a 1-pixel dotted outline that disappeared on light backgrounds to a 2-pixel solid blue ring that broke the layout on tightly packed lists. The new ring was a single token: a 2-pixel outline offset by a 2-pixel transparent gap from the component edge so the ring read against any surface. Every interactive component picked it up by default; there was no opt-out.
Motion preferences were the third foundation. The legacy system honored prefers-reduced-motion in approx. one place — a single onboarding animation — and the native apps honored it nowhere. The new foundation made motion a token, with three values (none, reduced, full) plumbed through every animation primitive. A designer who tried to override the preference had to attach a written justification that the program lead reviewed.
Tranche 2 — Components
With the foundations stable, the team turned to the approx. 90 primary components. The list was ordered by the complaint-pipeline data Maya had pulled in month one: checkout first, then messaging, then profile, then video. Each component went through a standardized rebuild: keyboard navigation map, screen-reader semantics, focus order, contrast verification at every state, reduced-motion variant, RTL variant, and — Maya's insistence — a documented test fixture the QA team could run on every release.
The credit-card-input field, in its old form, had been a single <input> with auto-formatting JavaScript that broke screen-reader announcement of typed characters; the rebuild used four separate inputs with explicit labels, errors tied via aria-describedby, and inline validation announced through a polite live region. It took six weeks for one designer and one engineer. The checkout-related accessibility tickets fell by approx. 70 percent the following quarter — because most of the new tickets simply stopped being filed.
Tranche 3 — Patterns
The final tranche was the one Maya described as the easiest in execution and the hardest in coordination. The team documented composition patterns — how to build an accessible modal flow on top of the rebuilt components; how to compose a list of items with mixed media; how to structure a settings page so the navigation worked under voice control. The patterns went into the design-system documentation site as runnable code examples. The hard part was not writing them. The hard part was getting every product team to use them instead of inventing their own.
The engineering rollout
A redesigned design system is a library; it is not, by itself, a rollout. The hardest project-management work of the program — Maya was unequivocal on this — was the migration. The product had approx. forty squads, each owning two-to-five surfaces, each free in practice to consume the design system at whatever cadence its roadmap allowed. A naïve plan would have asked every squad to migrate within a quarter. That plan would have failed.
Maya's solution was a graduated mandate. The new components shipped as the default; the old ones remained behind a feature flag, but every release of a surface still using a legacy component would automatically open a P2 ticket on that squad's backlog. The ticket would auto-escalate to P1 after ninety days and to P0 after one hundred and eighty. Within four quarters, approx. 78 percent of legacy primary-component usage had migrated. Within six quarters, that figure was approx. 94 percent.
"The hard part wasn't the design system. The hard part was a forty-squad org chart and a budget cycle that wasn't built for it. The components were three months of work. The rollout was three years."
Maya Okafor, composite, on the migration
What the program cost — and what it returned
Maya was scrupulous about tracking. By the time the program closed its formal phase in Q4 2024, the spend totalled — across two and a half years, eleven dedicated headcount and external testing — somewhere in the high single-digit millions. The accessibility-related ticket inflow was down approx. 73 percent against the 2022 baseline, despite a user base that had grown by approx. 20 percent. The two ADA-related legal matters opened during the program window were both closed without proceeding to court, on terms the company described in its annual filings as immaterial. The product's retention on the assistive-technology user segment — the segment Maya had identified in the funding pitch — had narrowed from a 1.8x churn ratio against the overall base to approx. 1.15x. Finance booked the difference. Maya did not say what the number was.
She also booked things that don't appear in the spreadsheet. The native iOS app's VoiceOver rotor support, which had been famously broken for years, became — in an independent audit in early 2025 — one of the best-performing in its sector. The high-contrast theme Maya had pushed for over brand-team objection became the default in regions where local regulators began enforcing the EAA. The design-system documentation site, viewed approx. 4,000 times a month at the start of 2022, was averaging approx. 38,000 monthly page views by mid-2025. A practice had been built; it would outlive her tenure.
What she tells designers at smaller orgs
By 2025, Maya was doing fewer ICU shifts on her own product and more advisory work for designers at companies an order of magnitude smaller — twenty-person product teams, fifty-person product teams, the size of org where a single designer has to be the accessibility lead by default. She had a small set of things she said in every coffee meeting. They are worth listing.
One. The complaint pipeline is the lever. You do not need a million users to have a complaint pipeline; you need a Support inbox and a willingness to read it. Print the tickets. Put them in a binder. Take the binder to the meeting. The binder works.
Two. The funding case has three columns. Legal exposure, market opportunity, and complaint volume. You don't need exact numbers on any of the three. You need the same person to see all three in one place, because no single column carries the room.
Three. Foundations before components, components before patterns. A team that starts by rewriting components first will spend a year doing it and arrive with a beautiful component library on top of an unsemantic colour palette, and the next designer will rewrite everything again.
Four. Negotiate the veto. The single biggest leverage point in a multi-team product company is the ability to say "this new component does not ship until it passes the checklist." The veto, deployed twice in two years, is sufficient. It is the credibility of the veto, not its frequency, that does the work.
Five. Hire the researcher with disability-community recruiting experience. The single line in Maya's program budget she would defend hardest was the researcher's seat. Without disabled users in the loop, the work is theatre.
Six. The clock on the legacy components is non-negotiable. Migrations without clocks do not happen. Migrations with clocks happen at the pace the clock allows.
Seven. Take the win and leave. Maya stepped off the program in Q1 2025 and moved to advisory. The founder of an accessibility program is the wrong person to run it in steady state. The founder's job is to make the program exist. The steady-state job is to be boring. Different temperament; different person.
A note on the binder
Maya still has the binder. She brings it out at conferences sometimes, when a senior designer asks her — usually with embarrassment, often after a panel — what to do about their own forty-one open tickets. The binder is half an inch thick. The sticker on the cover says, in a sans-serif handwriting Maya bought in a craft shop in 2022, "Day One". The forty-one tickets in it have all been closed. The names of the people who filed them are redacted in black marker. She does not show the names. She shows the pages, and she says: this is what the work looks like, and this is where it starts.
---
title: Profile: how one EU public-sector procurement officer enforces EN 301 549
url: https://www.disabilityworld.org/articles/profile-eu-procurement-officer-en-301-549/
description: A composite portrait of an EU member-state procurement officer who turns EN 301 549 from a referenced standard into rejected tenders, demanded evidence and post-award remediation clauses. Drawn from interviews with seven officers across five member states; identifying details anonymised.
author: Disability World
pubDate: 2026-05-22
tags: profile, procurement, eu, en-301-549, public-sector, web-accessibility-directive
---
# Profile: how one EU public-sector procurement officer enforces EN 301 549
Image description: An EU public-sector procurement officer's desk showing a printed EN 301 549 standard document with coloured tab-bookmarks, an EU flag pin beside it — the visual marker for the procurement-officer profile.
Reading Time: 10 minutes
"M." is a composite. The person described in this profile does not exist as a single named individual. The character is built from seven recorded conversations with public-sector procurement officers in five EU member states — three in the Netherlands and Belgium, two in Spain and Portugal, two in Germany — all of whom currently run accessibility-conformance review inside contracting authorities of broadly similar size (200 to 900 staff, annual ICT procurement spend between approx. 8 and approx. 40 million euro). Names, agency identities and any details that would identify a specific tender have been changed. Where direct quotations appear, they are verbatim from one of the seven interviews, attributed to "M." rather than to the original speaker. The workflow described — the clause language, the evidence threshold, the rejection categories, the remediation regime — reflects the consensus practice of the group, not the practice of any one office.
The point of profiling a composite officer rather than a single named one is that the same patterns recur across very different member states. None of the seven officers we spoke to had received central training in how to enforce EN 301 549 when they took up their role. All seven had built their workflow from the ground up — copying clauses from the published model contracts of larger ministries, refining the rejection criteria over three or four tender cycles, learning by losing arguments to vendors and then writing better tenders the next time. M. is what that learning curve produces. This piece is what M.'s desk looks like in May 2026, four years after the European Accessibility Act entered force and seven years into the maturity of the Web Accessibility Directive.
The route in — how a procurement officer becomes an accessibility enforcer
M. did not start out as an accessibility specialist. The route in was sideways. M. trained as a public-administration generalist, joined a member-state ministry's procurement office in the late 2010s as a contract administrator, and inherited the accessibility file in 2021 because the colleague who had been handling it left for the private sector and the desk had to go to somebody. At the time the file consisted of a folder of unsigned policy memos and a single previous tender in which an "accessibility clause" had been pasted in from a template the colleague had downloaded from a German ministry's open-procurement portal. The clause referenced "the relevant European standard" without naming EN 301 549 and demanded "conformance with WCAG 2.0 Level AA" — a version of WCAG that had been superseded for seven years.
The first thing M. did, in early 2022, was rewrite the clause. The rewritten version named EN 301 549 V3.2.1 explicitly, named the applicable chapters (9 for web content, 11 for non-web software, 12 for documentation and support), specified WCAG 2.1 Level AA via the EN's Chapter 9 reference, and required the vendor to submit a conformance report at tender stage rather than after award. That clause has been refined four more times since then — once after a vendor argued the standard did not bind them because their product was "primarily a back-office tool", once after a different vendor submitted a self-certification report consisting of two pages of marketing copy, once after the EAA's transposition act in the relevant member state added explicit penalty references, and once in late 2025 in anticipation of EN 301 549 V4.0.0's incorporation of WCAG 2.2 AA.
"The mistake I made in the first tender was treating accessibility as a checkbox. The vendor put a tick in the box. We did not ask them to show working. The second tender, I changed one sentence. I said: a tick is not evidence. From that day everything changed."
M., procurement officer, EU member-state contracting authority
The tender language — what M.'s standard clause looks like in 2026
M.'s standard accessibility clause is now four paragraphs long and runs to about 380 words inside the tender's technical-requirements section. The first paragraph names the legal authority: the national transposition of the Web Accessibility Directive for public-sector websites and mobile applications, and the national transposition of the EAA for any product or service falling under its scope. The second paragraph names the technical standard — EN 301 549 V3.2.1, with a forward-looking provision that any product delivered after the publication of V4.0.0 in the Official Journal must be re-evaluated against the new version at the supplier's expense within twelve months. The third paragraph specifies what conformance evidence the vendor must submit. The fourth paragraph specifies the remediation regime that applies post-award if any conformance gap is later discovered.
The third paragraph is the operative one. It demands what M. and the other officers we spoke to call the "European ACR" — an Accessibility Conformance Report following the structure of the US VPAT template but referencing the EN 301 549 clause set rather than Section 508. ETSI publishes a template for this; some member states publish their own. M.'s tender requires the ACR to (a) name each applicable clause of EN 301 549 by number, (b) state for each clause whether the product Supports, Partially Supports, Does Not Support or Not Applicable, (c) provide a one-paragraph remark for every clause whose status is anything other than "Supports", and (d) attach the underlying audit report on which the ACR is based.
The last sub-clause is the one that does the work. A blank ACR with "Supports" entries across the board can be produced by any vendor in under an hour. An underlying audit report cannot. M.'s tender explicitly requires that the audit be a third-party audit by a body listed in the relevant national accreditation register, or — where the contract value is below the threshold at which third-party audit is proportionate — that the audit be performed by an in-house team whose evaluators hold a recognised qualification (in M.'s practice, IAAP CPACC or WAS) and whose audit methodology is documented and reproducible. Pure vendor self-certification without either a third party or a qualified internal auditor is flagged automatically as non-compliant.
The evidence threshold — what counts and what gets flagged
The single biggest change M. made between 2022 and 2026 was tightening the evidence threshold. In 2022, M. would accept any ACR submitted on the vendor's letterhead, provided the format matched the ETSI template. By 2024, after two awards in which the post-award accessibility audit found large gaps the tender ACR had not declared, M. had switched to a sliding scale: third-party audit accepted at face value; in-house audit accepted subject to spot-check; vendor self-certification accepted only if accompanied by a signed declaration that the underlying methodology can be produced on demand and that the signing officer is personally accountable for its accuracy under the contract's misrepresentation clause.
In practice, the spot-check is the lever. M. now spot-checks approx. one in three in-house audits — pulling three to five clauses at random from the submitted ACR and asking the vendor to produce, within five working days, the test scripts, the assistive-technology configuration used, the names of the testers, and the raw output. Vendors who can produce these in five days pass. Vendors who cannot, or who produce them in a form that contradicts the ACR's claimed status, are rejected.
There are now four named patterns that get a tender flagged in M.'s office. The first is "WCAG 2.0 leakage" — clauses citing WCAG 2.0 instead of 2.1, usually a sign of an old template that has not been refreshed. The second is "Supports without remark" — every clause marked Supports with no explanatory remark anywhere, which the spot-check almost always punctures. The third is "Section 508 substitution" — a vendor submitting a US VPAT against Section 508 in place of an EN 301 549 ACR, which is non-responsive on its face but still common from US-headquartered suppliers. The fourth is "Out-of-scope claim" — a vendor stating that EN 301 549 does not apply because the product is back-office software, or B2B, or used only by internal staff. In the public-sector context M. procures in, none of those exclusions hold; staff-facing systems are explicitly within scope of the national Web Accessibility Directive transposition.
"I do not reject a tender for honest gaps. I reject a tender for dishonest paperwork. A vendor who says 'Partially Supports' and explains why is having a conversation with me. A vendor who says 'Supports' on every line is hoping I do not read it."
M., procurement officer, EU member-state contracting authority
Rejections and remediation — the argument that splits the profession
The biggest argument inside the profession in 2026 is not whether to require EN 301 549 in tenders — that is settled — but what to do when a tender's ACR reveals gaps. There are two camps. The first camp, the rejecters, treats any material non-conformance disclosed in the tender ACR as grounds for exclusion from the procedure. The second camp, the remediators, treats the disclosed non-conformance as a baseline against which the awarded contract sets a remediation schedule, with milestones, penalties for missed milestones, and a withholding provision against final payment.
M. has moved across the line twice. In 2022 and 2023 M. rejected. In 2024, after a procurement procedure was lost because the two most operationally suitable bidders had both disclosed Chapter 11 gaps and were both excluded, leaving an award to a less suitable bidder with a cleaner ACR but worse product fit, M. shifted to remediation. In late 2025, after a remediated award produced eighteen months of missed milestones and an eventual partial termination, M. shifted partway back. The current practice in M.'s office is to reject on conformance grounds only when the disclosed gap is in a Chapter 9 (web) clause that is fundamental to the user task — keyboard operability, focus visibility, programmatic name — and to remediate when the gap is in a Chapter 11 software clause that has a credible technical path to fix.
The argument for rejecting is that the procurement procedure is the moment of maximum leverage. Once a contract is awarded, leverage shifts to the vendor; milestones slip, change requests come with extra cost, accessibility falls down the priority list as other defects compete for engineering time. The argument for remediating is that strict rejection narrows the field — sometimes to one bidder, sometimes to none — and that a contracting authority with a thin market cannot afford to reject everyone. Both arguments are right under different conditions. The skill of the procurement officer is in reading which set of conditions applies to the tender in front of them.
The post-award clauses — what makes remediation actually work
When M. remediates, the contract carries four specific clauses. The first names a remediation schedule — usually three milestones at three, six and twelve months post-award — keyed to specific EN 301 549 clauses. The second names a payment-withholding provision — a stated percentage of each invoice (M. uses approx. 15 percent) held back until the milestone for the period is signed off. The third names a re-audit obligation — the vendor pays for a fresh third-party audit at month twelve to verify the remediation. The fourth names a termination trigger — two consecutive missed milestones without cause permit the contracting authority to terminate for material breach.
M.'s observation is that the payment-withholding provision does almost all the work. Stated penalties — flat fines for missed milestones, escalation clauses — are slow to invoke and politically expensive. A withheld invoice line is mechanical. The vendor's finance team applies pressure to the engineering team on the next working day. The remediation gets done.
What smaller agencies should copy
Most of M.'s practice is not specific to large contracting authorities. The seven officers we spoke to all said the same things when asked what smaller agencies — municipal IT departments, regional health authorities, local-government procurement offices with one or two staff — should copy from their workflow. We pull those out as a list, in the order the officers themselves ranked them.
Name the standard by number and version. Do not write "the relevant European accessibility standard". Write "EN 301 549 V3.2.1, with V4.0.0 to apply on publication in the Official Journal". A named standard is enforceable; an unnamed one is rhetorical.
Demand the ACR at tender stage, not at award. The ACR is your filter. If you cannot see it before you award, you cannot exclude on it.
Demand the underlying audit, not just the ACR. The ACR is a summary; the audit is the evidence. Refuse self-certified ACRs with no audit attachment.
Spot-check one in three in-house audits. Even if you do not have time to spot-check everything, randomised spot-checking changes vendor behaviour across the entire bidder pool.
Reject on Chapter 9 fundamentals; remediate on Chapter 11 gaps. Keyboard, focus, programmatic name are non-negotiable. Software-platform clauses have remediation paths and can be scheduled in.
Use payment withholding, not flat fines. Withhold approx. 15 percent per invoice against the next milestone. Pay it out only when the milestone is signed off. This single clause is the most effective enforcement instrument in the contract.
Re-audit at month twelve at the vendor's cost. A second audit, done by a different third party from the one that produced the tender ACR, is the only way to know whether the remediation actually closed the gap.
Refuse Section 508 substitution. A US VPAT against Section 508 is not an EN 301 549 ACR. Send it back and ask for the European document.
Copy clauses from larger ministries. Every EU member state has a ministry that publishes its model accessibility clauses. Start there. Do not draft from scratch.
Document your rejections. Build a one-page memo for every tender you reject on accessibility grounds. After three or four memos you have an institutional position that the next vendor's lawyers cannot argue around.
The desk at the end of the day
M.'s desk, when we visited the composite version of it on a Thursday afternoon in early May, had a printed copy of EN 301 549 V3.2.1 with coloured tabs along the right edge — green for clauses M. had cited in a tender that month, yellow for clauses currently in dispute with a vendor, red for clauses where a previous award had failed and the dispute had escalated. The small EU flag pin on the lanyard beside the document was a souvenir from a 2023 Brussels training session on the EAA. The pin and the document together are the visual signature of a role that did not really exist in this form ten years ago.
M. ended the conversation with a line that summarises the entire workflow: enforcement of a standard like EN 301 549 is not, in the end, about the standard. It is about the discipline of asking the vendor for evidence at the moment when the vendor most wants to give it — when the contract is unsigned — and then refusing to be talked out of the evidence requirement when the answers prove difficult. The standard exists. The contracting authority's job is to make the standard cost something when it is ignored.
---
title: Profile: an in-house counsel's-eye view of 2024-2026 ADA litigation
url: https://www.disabilityworld.org/articles/profile-in-house-counsel-ada-litigation/
description: A composite portrait of a senior in-house counsel at a mid-size US ecommerce / SaaS company who has handled more than 50 ADA web-accessibility demand letters between 2024 and 2026 — the playbook she sees on the plaintiffs' side, the early-settlement window.
author: Disability World
pubDate: 2026-05-22
tags: profile, in-house-counsel, ada-litigation, settlement, corporate-defense, legal
---
# Profile: an in-house counsel's-eye view of 2024-2026 ADA litigation
Image description: A documentary close-up of a corner of an in-house counsel's desk — a stack of letter-sized documents fanned slightly, a pair of reading glasses resting on top, a brass nameplate visible in soft focus, warm afternoon light from an office window.
By Disability World editorial deskUpdated May 2026Reading time 10 minutes
Reading Time: 10 minutes
Editor's note: the subject of this profile is a composite. The biographical detail is drawn from four senior in-house counsel — two at US-headquartered ecommerce retailers and two at B2B SaaS companies — who, between them, have handled more than two hundred web-accessibility demand letters from 2022 onward. Names, employers, and identifying transactional facts have been combined and altered. The procedural and financial numbers in the quoted passages have been preserved as the sources reported them and cross-checked against publicly filed motions, the federal PACER docket, and California Judicial Council civil-filing data. Where the subject speaks in the first person, the words are paraphrases the contributors approved as faithful to their on-record statements. We have used the name "M.R." for the composite to avoid implying any single individual.
M.R. is forty-three, a 2007 graduate of a Midwestern law school, and the Vice President and General Counsel of a privately held US ecommerce-and-SaaS company that sells branded consumer goods directly and also licenses a checkout platform to several hundred smaller merchants. Annual revenue sits in the low nine figures. The legal team is four lawyers plus a paralegal. Until late 2023, M.R. had never read the WCAG 2.2 success criteria from end to end. Today she can recite the first eleven of them in order. The story of how that happened — and of the cheque she nearly wrote before she realised she should be writing a different one — is, in miniature, the story of where the US ADA Title III web-accessibility-litigation industry has arrived in 2026.
Demand letter no. 1
The first one arrived on a Thursday afternoon in March 2024, in a manila envelope, certified mail. The return address was a one-attorney plaintiff-side shop in the Eastern District of New York. The named plaintiff was a legally blind New York City resident with a documented filing history of approx. 80 prior accessibility complaints over four years. The body of the letter ran to nine pages. Roughly the first six were, M.R. quickly realised, boilerplate: a recitation of Title III and the Second Circuit's place-of-public-accommodation jurisprudence, an invocation of WCAG 2.1 AA as the operative technical standard, and a paragraph asserting that the plaintiff had attempted to use the company's storefront with the JAWS screen reader and had been unable to complete a purchase. The remaining three pages were the part that mattered: a list of specific failures, dated screenshots, and a settlement demand.
The failures named in the letter were unsurprising to anyone who had ever read an accessibility audit. Five product-detail-page images without alt text. A custom-built quantity-selector widget that JAWS announced as "button" with no value and no label. A modal dialogue whose close-control could not be reached by keyboard. A focus indicator that disappeared inside the checkout flow. A footer "accessibility statement" link that opened a 404 page. The evidence threshold was modest: the letter cited five concrete failures, each illustrated with a screenshot or a JAWS speech-output transcript. It did not allege a comprehensive site-wide failure. It did not need to. Under settled Title III doctrine, a single denial-of-access barrier on a public-accommodation website is, in principle, an ADA violation.
The settlement demand was $18,500. The letter did not characterise it as a settlement; it characterised it as a pre-litigation good-faith resolution offer that would extinguish all claims related to the named accessibility barriers and would underwrite the plaintiff's "monitoring fees" for twelve months. M.R. read the demand three times and then forwarded the envelope, scanned, to the company's outside litigation counsel.
"I remember thinking — eighteen thousand, five hundred dollars. That is a quarter of one engineer for a month. It is half of one trade-show booth. It is roughly what we spend on coffee in this office over a year. The instinct, on that first letter, was not to fight. The instinct was to make it disappear."
M.R., VP & General Counsel (composite)
Outside counsel returned the file the next morning with a one-line recommendation: pay, take the release, fix the five named issues, move on. The recommendation came with a memo. The memo explained the economics. A motion to dismiss a properly pleaded Title III complaint costs, in the Southern or Eastern District of New York, somewhere between $40,000 and $90,000 in fees before any merits ruling. Survival of the motion does not end the case — it begins discovery. A trial-track Title III case carries fee exposure in the high six figures and, in the event of an adverse judgment, the plaintiff's reasonable attorney's fees on top. The plaintiff's settlement demand was, by construction, less than a third of the cost of the first procedural skirmish. M.R. signed the cheque on a Friday. The release came back on Tuesday. The five issues were patched the following sprint.
The early-settlement window
Then the second letter arrived. And the third. By the end of the second quarter of 2024, M.R. had received seven demand letters from four different plaintiffs' firms. By the close of 2024, the running total was nineteen. The boilerplate text varied at the margins — different cited authorities, different opening recitations, occasionally a different operative WCAG version — but the structure was identical. Six pages of legal scaffolding. A list of five-to-eight specific named failures. A demand in a tight band between roughly $10,000 and $20,000, almost always converging on the high teens.
That band is the early-settlement window. It is calibrated by the plaintiffs' bar to the cost curve M.R.'s outside counsel had laid out: low enough that a sober general counsel will not litigate, high enough that the plaintiff's firm — which typically takes 33 to 40 per cent of the gross — earns a meaningful fee for what amounts to four-to-eight hours of paralegal time generating the letter and the screenshots. The window has been stable across 2023, 2024, and 2025. PACER data and Judicial Council filings show the modal early settlement amount converging on approx. $14,000 to $18,000 in the major filing districts; the band tightened, rather than rose, as more defendants paid promptly.
The evidence threshold is similarly calibrated. The named failures in a typical demand letter are not arbitrary — they are drawn from the small set of high-frequency violations that are cheapest for a plaintiff's investigator to surface with a fifteen-minute screen-reader walkthrough of a homepage and a product-detail page. Missing or incorrect image alt text, unlabelled form fields, inaccessible custom widgets, keyboard traps in modals, and broken focus management are the canonical five. A plaintiff's filing investigator does not need to audit the entire site. A handful of named violations, each evidenced by a screenshot or a transcript, is sufficient to plead the complaint and to anchor the settlement demand.
"By letter five, I understood the model. By letter nine, I had a spreadsheet — date received, named plaintiff, plaintiff's firm, named failures, demand, settlement, days-to-release. By letter fifteen, I could predict the demand within two thousand dollars from the letterhead alone."
M.R., VP & General Counsel (composite)
The aggregate spend, by mid-2025, was running at approx. $260,000 a year in settlements alone, not counting outside-counsel hours on intake, release negotiation, and the routine remediations the company was performing in response. The marginal demand letter cost the company approx. $16,000 to settle plus approx. $3,500 in outside-counsel fees to manage. The plaintiff's firm, on the other side, was netting approx. $5,500 to $7,000 per letter for what was — visibly, repeatedly, identically — a paralegal task. The asymmetry was not a misperception. It was the design.
The procedural-reform turn
Two things changed the math in 2024 and 2025. The first was that the procedural-reform pieces — the Supreme Court's December 2023 disposition in Acheson Hotels, LLC v. Laufer, the federal-court tester-standing uncertainty that followed, California's reinforced Civil Code §425.55 gate for high-frequency Unruh litigants, and New York's CPLR §3211 reforms tightening pre-answer motion practice — began to bite. The second was that M.R. started reading the procedural posture of the cases she was settling, instead of just the demand numbers.
CPLR §3211 has been on the books in New York for decades. What changed for accessibility defendants between 2023 and 2026 was the willingness of New York State Supreme Court justices to entertain pre-answer §3211(a)(7) motions in NYCHRL accessibility complaints — and, more importantly, the way the New York plaintiffs' bar adapted. As federal tester-standing motions began to bite in SDNY, the same plaintiffs' firms started filing under the New York City Human Rights Law in Supreme Court of New York County, where standing doctrine is meaningfully more generous and where compensatory damages are available. The migration of filings from federal court to state court was, for M.R., visible in the letterheads on her desk. The fourth-quarter 2024 letters arrived as drafts of state-court complaints, not federal ones.
California §425.55 was, in some respects, the more consequential of the two reforms — at least for the defendants on the receiving end of Unruh-anchored demand letters. The provision, in force since 2015 and meaningfully reinforced in 2022, requires that any "high-frequency litigant" — defined by the number of accessibility cases filed in the preceding twelve months — pay an additional $1,000 filing fee for any state-court Unruh claim and submit specific verified disclosures about their disability, their visit to the public accommodation, and their reason for filing. The federal-court counterpart, California Code of Civil Procedure §425.50, imposes parallel verified-pleading requirements. The combined effect is that California Unruh complaints filed by repeat-filing plaintiffs now carry an out-of-pocket procedural cost — both at the filing fee level and at the verified-pleading drafting level — that did not exist in 2015. The plaintiffs' firms responded by being more selective about which defendants they targeted and by raising their early-settlement asks in California venues by approx. 15 to 20 per cent, but the underlying volume began, slowly, to compress.
For an in-house counsel watching the trend lines, the inference was straightforward: the procedural gates do not eliminate the demand-letter industry, but they raise the cost of running it. The plaintiffs' bar that survived the gates was the segment that filed harder cases, named more defendants per letter, and demanded larger settlements. M.R.'s spreadsheet started to show fewer letters per quarter from late 2024 onward, but the median demand per letter began to drift upward — from approx. $16,500 in the first quarter of 2024 to approx. $22,000 by the fourth quarter of 2025.
The remediation pivot
The moment M.R. decided that the settlement strategy had run its course did not arrive as a strategic realisation. It arrived as a board question. In February 2025, the company's audit committee — three independent directors and the CEO — asked, in the ordinary course of the quarterly legal-spend review, why the litigation-reserve line for "accessibility settlements" was tracking at approx. $280,000 against a remediation-budget line of approx. $45,000. The CFO had constructed the question as a straightforward variance review. M.R. did not have an answer that survived two minutes of scrutiny.
"The board was not angry. The board was confused. One of the directors asked the obvious question: if you are paying two hundred and eighty thousand dollars a year to plaintiffs' firms, would the same money, deployed inside the engineering organisation, fix the problem? I had to say I did not know. That was the morning I started rebuilding."
M.R., VP & General Counsel (composite)
The rebuild took eighteen months and is still continuing. M.R. brought in an external accessibility audit firm to conduct a full WCAG 2.2 AA audit of the storefront, the checkout, the licensed checkout SDK shipped to merchant customers, and the admin console. The initial audit returned approx. 340 named issues across the four surfaces, classified by WCAG criterion and severity. Roughly 60 per cent of the issues were trivial-to-moderate fixes — alt text, ARIA labels, focus management, contrast adjustments — that could be batched into engineering sprints across three quarters. Roughly 30 per cent were custom-widget rewrites of the kind that show up repeatedly in demand letters: the quantity selector, the modal dialogue, the cart drawer, the address autocomplete. Roughly 10 per cent were architectural — the design-system component library, the form-validation pattern, the announcer-region strategy for asynchronous updates — and required senior-engineer time across two quarters.
The total investment, calendar-2025 plus the first half of 2026, was approx. $410,000: approx. $90,000 in external audit and consulting fees, approx. $260,000 in re-allocated internal engineering time, and approx. $60,000 in tooling, training, and an automated CI accessibility-regression pipeline. The settlement reserve for calendar 2025 came in at approx. $215,000 — a moderate decline from 2024, reflecting the long tail of pre-remediation issues still arriving in demand letters. The forecast for calendar 2026, with the bulk of the high-frequency issues remediated and the regression pipeline running on every pull request, is approx. $90,000 to $120,000.
The dual-tracking strategy — paying the early-settlement window while remediating — was deliberate. M.R. did not stop settling in 2025. The cost calculus on the marginal letter — $16,000 to $22,000 to make it disappear versus $40,000-plus to litigate the motion to dismiss — was unchanged. What changed was the underlying surface area. As the remediated surfaces went live, the named failures in incoming demand letters increasingly described pages that had already been fixed; the screenshots were stale. Outside counsel could respond with a substantive denial — backed by a current accessibility-audit report, a deployment log, and, in two cases, a video recording of the named page being navigated successfully with JAWS — without resorting to motion practice. Several of the late-2025 letters were withdrawn without payment after that initial substantive response.
The insurance piece sat alongside the remediation track and was, in M.R.'s telling, the most useful single move she made. The company carried general-liability coverage that did not address accessibility claims and a media-liability policy that had a narrow defence-only endorsement for ADA Title III matters. In the 2025 renewal cycle, M.R. negotiated a specific accessibility-liability rider that covered defence costs, indemnity for settlements within agreed limits, and — crucially — a "remediation incentive" credit against the premium when the company could demonstrate documented progress against a WCAG 2.2 AA roadmap. The rider cost approx. $38,000 in additional premium and recovered approx. $74,000 in defence costs across 2025 alone. The remediation-incentive credit became the lever that let M.R. justify the engineering re-allocation to the CFO without re-opening the budget cycle: every dollar she put into remediation reduced the following year's insurance premium by a documented fraction.
Lessons — what M.R. tells other in-house counsel
M.R. now takes informal calls from peer general counsel at other ecommerce and SaaS companies roughly twice a month. The companies calling are smaller than hers, in the early stages of the demand-letter cycle, and asking the same questions she was asking in mid-2024. The substance of what she tells them is consistent enough to write down.
First, settle the first letter; track every variable from the second. The economics of motion practice on a single demand letter favour settlement under any reasonable read of the cost curve. But the moment a second letter arrives — and it will arrive within ninety days, almost without fail, from a different plaintiff's firm citing different but adjacent failures — the company is in a demand-letter relationship, not a litigation incident. The relationship needs a spreadsheet. Date, plaintiff, plaintiff's firm, named failures by WCAG criterion, demand, settlement, days-to-release. Without the spreadsheet, the company is paying a sequence of unrelated bills. With the spreadsheet, the company is buying data.
Second, read the procedural posture, not just the demand. A letter that threatens federal-court filing in 2026 is making a different threat than a letter that threatens NYCHRL state-court filing or Unruh state-court filing. The defensibility of each posture, the removability of each posture, the cost curve of motion practice in each forum, and the standing-doctrine vulnerability of the plaintiff vary materially. The state-court migration is real, the procedural-reform statutes are biting differently in different venues, and a 2024 settlement script applied to a 2026 letter will overpay.
Third, do not budget remediation against this year's settlement spend. The remediation case is not "we will spend $400,000 this year to save $260,000 next year." That comparison loses on a one-year horizon. The case is "we will spend $400,000 once, to compress the demand-letter surface area, to make the marginal letter denyable rather than payable, and to reduce the insurance premium and the engineering-time-on-incident cost in every year that follows." The CFO conversation needs a three-year model, not a one-year variance.
Fourth, dual-track the insurance and the remediation. Coverage that does not include an accessibility-specific rider is no coverage. A rider that does not include a documented-remediation premium credit leaves money on the table. The 2025 and 2026 renewal markets are willing to write the rider on reasonable terms for defendants who can show a WCAG 2.2 AA roadmap and a regression pipeline. They are unwilling to write it for defendants who cannot.
Fifth, do not delegate the technical reading to outside counsel. Outside litigation counsel is, in the median case, not WCAG-literate. They will read a demand letter as a procedural document and will not register the difference between a named failure that the company has remediated and a named failure that the company has not. The in-house counsel who reads the WCAG criteria in the letter alongside the company's current audit report is the one who can tell outside counsel which letters to settle and which letters to deny.
What the in-house view changes
The defendant-side narrative about US ADA web-accessibility litigation has, for most of the past decade, been written in the language of grievance — boilerplate letters, repeat-filing plaintiffs, an industry that exists to extract small settlements. That narrative is not wrong about the mechanics; the early-settlement window is a designed feature of the industry, not an accident. But it has been wrong about the response. The response that minimises legal spend over a three-year horizon is not litigation. It is remediation, dual-tracked with insurance, sequenced against the procedural-reform landscape, and managed in-house with a spreadsheet that treats each demand letter as a data point in a stable distribution.
What M.R.'s story illustrates is that the in-house counsel who reaches this view is not the in-house counsel who reads the most case law. It is the in-house counsel who reads her own settlement ledger, asks a board-level question about the variance, and accepts that an answer she does not yet have is the start of a different conversation. The demand-letter industry will outlast any one defendant's pivot. The defendants who pivot first will, in the aggregate, fund less of it.
This article will be followed, in this same series, by parallel views from a senior plaintiff's-side accessibility attorney and from a member of the New York State plaintiffs' bar working under the post-CPLR-§3211 procedural environment. The intent is not to balance the in-house narrative against an opposing one — it is to show how each side of the docket reads the same set of letters, settlements, and reforms differently, and where the readings agree.
---
title: Progressive web apps and accessibility: the state of the art in 2026
url: https://www.disabilityworld.org/articles/pwa-accessibility-state-of-the-art/
description: Where progressive web apps stand on accessibility in 2026 — install-prompt UX, adaptive icons, screen-reader handoff between web and native, the file_handlers / share_target / window_controls_overlay manifest surface, offline assistive-tech behavior, and the iOS Safari install path post-iOS 16.4.
author: Disability World
pubDate: 2026-05-22
tags: pwa, progressive-web-apps, service-worker, offline, manifest, mobile, tech-news
---
# Progressive web apps and accessibility: the state of the art in 2026
Engineering primer · PWA accessibility 2026
Progressive web apps and accessibility:
the state of the art in 2026
Six years after Apple finally shipped a working install path on iOS 16.4, the progressive web app has stopped being a curiosity and started being a procurement question. This primer is for engineering teams who need to know, in 2026, what a PWA actually owes its assistive-tech users — and where the platform still falls short of a real native app.
2023
iOS 16.4 — first usable PWA install path on Safari
11
manifest properties that affect AT behavior
approx. 35%
Lighthouse PWAs whose install button is unlabeled to AT
By Disability World — engineering desk
9 min read
Updated May 2026
Foundation
1. What "PWA accessibility" means in 2026
A progressive web app is, at runtime, three things layered on top of a normal website: a Web App Manifest, a service worker, and an installed-mode chrome that replaces the browser frame with the operating system's own task-switcher entry. Each of the three layers introduces its own accessibility obligations — and each fails its assistive-tech users in a different, separately-debuggable way.
In 2020 the entire conversation collapsed into "WCAG applies to PWAs," which was technically correct and operationally useless. In 2026 the conversation is split into the four surfaces that actually matter: the install-prompt UX, the manifest properties that drive OS-level affordances, the handoff between the browser's accessibility tree and the operating system's accessibility tree once the PWA is launched in standalone mode, and the assistive-tech behavior of the service-worker offline fallback. WCAG 2.2 governs the document; the platform-integration layer is governed by a much patchier mix of W3C drafts, vendor-specific behavior, and ARIA conventions inherited from the web.
i
Scope note
This primer covers the platform-integration surface of PWAs — install prompts, manifest properties, standalone-mode AT behavior, offline fallback. It assumes the underlying document already meets WCAG 2.2 AA. A PWA wrapper on top of an inaccessible page is still an inaccessible page.
Surface
2. The install prompt
The install prompt is the most user-facing PWA surface and, in 2026, still the worst-engineered one. On Chromium, the prompt is gated by `beforeinstallprompt`, which fires only after a heuristic engagement threshold and which sites typically wire into a custom "Install app" button. That custom button is where accessibility goes wrong: roughly one in three Lighthouse-scoring PWAs renders the install affordance as a `<div>` or a styled `<span>` with no role, no accessible name, and no keyboard handler — invisible to a screen reader, unreachable by Tab, and indistinguishable from decorative chrome.
The fix is unglamorous and mandatory: render the install affordance as a real `<button>`, set an accessible name that includes the verb ("Install Disability World on this device"), expose the same button to all input modalities, and announce success or failure via a live region after the user dismisses the OS-level confirmation sheet. The same applies to the related-applications and beforeinstallprompt dismissed states — both must produce an AT-perceivable status change.
Install-button anatomy: bad vs good
Worse — what we see in the wild
`<div onclick="install()">Install</div>` — not focusable, no role, screen reader sees only the word "Install" with no actionable affordance.
Button hidden until `beforeinstallprompt` fires — keyboard users land on a stale "Install" link that does nothing after the event.
No status announcement after dismissal — the AT user has no way to know whether the install succeeded.
Better — what 2026 ships
`<button type="button" aria-label="Install Disability World">...</button>` with explicit `aria-disabled` when the install isn't yet available.
`beforeinstallprompt` handler stashes the event; the button reflects the resulting state via `aria-disabled` toggled on event arrival.
A polite live region announces "Installed" or "Install dismissed" after `userChoice` resolves, so the AT user has a perceivable confirmation.
Reference
3. The manifest surface
The Web App Manifest grew quietly between 2022 and 2026, and many of its newer properties carry direct accessibility consequences. The matrix below maps the eleven manifest properties that interact with assistive technology to what each browser actually does with them today — across Chrome on Android, Safari on iOS, Edge on Windows, and Firefox on desktop. Properties such as `file_handlers`, `share_target`, and `window_controls_overlay` did not exist in any meaningful form in 2021; in 2026 they shape whether the PWA appears in the OS share sheet, opens files from the system file manager, and renders its own titlebar — all of which the screen reader user must be able to perceive and operate.
When a PWA opts into `window_controls_overlay`, it takes over the OS titlebar — including the region where, on a native app, the screen reader would announce the window title automatically. Apps that adopt this property must explicitly render their own focusable, AT-labeled titlebar control inside the safe-area inset, or screen-reader users lose the only on-screen anchor for "where am I in this app."
Behavior
4. The web ↔ native screen-reader handoff
The single hardest debugging problem in PWA accessibility, in 2026, is what happens when the user crosses the seam between standalone-mode PWA chrome and the operating system itself. On Android, TalkBack reads the manifest `name` when the user focuses the home-screen icon, then transitions to reading the in-app accessibility tree once the PWA launches; on iOS 16.4+, VoiceOver does the same for an installed PWA but with one important quirk — the first focusable element after launch is announced without the app-level context that a native iOS app would supply through its UIWindow title.
The PWA author has one tool to bridge this gap: on cold launch, focus a heading or main-region landmark that includes the app name in its accessible label, and set the document `<title>` to a string that the OS task switcher will read when the user swipes between apps. Without this, the screen-reader user loses the contextual cue that they have switched applications — a "where am I" failure that doesn't exist for native apps.
"In 2024 a Bluetooth-keyboard VoiceOver user told us, on a PWA we had certified to WCAG 2.2 AA, that they had no idea they had switched out of Safari and into our app. The document was accessible. The handoff was not."
— Disability World user-research diary, October 2024
Behavior
5. Offline + assistive-tech behavior
When the service worker serves an offline fallback page, two AT-specific failure modes appear: the focus that was inside the now-unloaded page is silently dropped onto the document body, and the offline page itself rarely uses a live region to tell the screen-reader user what just happened. The result is a user who hears one announcement of the offline-page title (if they're lucky) and otherwise experiences a total loss of context.
The fix is to treat the offline transition as a state change, announce it through a polite `aria-live` region, restore focus to a known landmark on the offline page, and surface a "Retry" control as a real button rather than the "Reload" link most service-worker boilerplates ship. The same applies to the foreground-sync recovery path: when connectivity returns and the service worker drains the queue, that too is a state change the AT user must be told about.
+
Service-worker checklist
Polite live region announces "You are offline" on transition. Focus is moved to the offline-page main heading. A clearly-labeled `<button>Retry</button>` is the first interactive element. On reconnection, a second polite announcement says "Connection restored" and focus is restored to whatever the user was last interacting with.
Comparison
6. iOS Safari vs Android vs native
The "should we ship a PWA or a native app" question now has an accessibility dimension as well as a feature-completeness one. Below, we compare the same hypothetical news-reader app delivered four ways — as a PWA on Android, as a PWA on iOS 16.4+, as a native iOS app, and as a native Android app — across the five surfaces a screen-reader user touches first.
PWA · Android
PWA · iOS 16.4+
Native · iOS
Native · Android
Install affordance discoverable by AT
If the dev did it right
Add-to-Home-Screen menu — discoverable
App Store — fully accessible
Play Store — fully accessible
App name + description on launcher icon
Yes
Yes (`name` + `apple-mobile-web-app-title`)
Yes (UIKit Info.plist)
Yes (Android manifest)
Adaptive icons (themed / monochrome)
Yes (maskable)
No
Yes
Yes
App-switcher context announced
Yes
Partial
Yes (UIWindow title)
Yes
OS share sheet entry
Yes (`share_target`)
No
Yes (UIActivity)
Yes (Intent filter)
Long-press shortcuts
Yes (`shortcuts`)
No
Yes (UIApplicationShortcutItem)
Yes
Push-notification accessible content
Yes
Yes (since iOS 16.4)
Yes
Yes
Custom rotor / quick nav
N/A
N/A
Yes
Yes
i
The iOS gap in 2026
iOS 16.4 unlocked the install path, push notifications, and the badging API for PWAs, and iOS 17 closed the gap further on the basic launch surface. But `file_handlers`, `share_target`, `shortcuts`, and `window_controls_overlay` remain unsupported. For an AT user who relies on the OS share sheet to move content between apps, a PWA on iOS is still a meaningfully smaller surface than a PWA on Android or a native iOS app.
Conclusion: the 2026 playbook
Ship the install affordance as a real `<button>` with an accessible name. Wire a polite live region to the `userChoice` outcome. Fill out `name`, `short_name`, `description`, `lang`, and `dir` in the manifest, and ship maskable icons for Android. If you opt into `window_controls_overlay`, render and label your own titlebar; if you opt into `file_handlers` or `share_target`, treat the resulting launch as a state change and announce it on entry.
Restore focus to a labeled landmark every time the screen-reader user crosses the seam — first launch, app-switcher return, offline transition, share-target launch, reconnection. Treat each crossing as a discrete event that owes the user a perceivable announcement and a known focus anchor. None of this is hard; almost none of it is shipped consistently.
A PWA in 2026 can be very nearly indistinguishable from a native app for an assistive-tech user — on Android. On iOS it is closer than it was, and still has a real gap. The gap is closing roughly one manifest property per year. For procurement teams choosing between a PWA and a native app, the accessibility question is no longer "can a PWA be accessible?" — it can. The question is whether the team building it has read the eleven manifest rows above and accepted that each one is part of their deliverable.
"A PWA wrapper does not absolve a team of the platform-integration work. It adds eleven new accessibility surfaces and asks the team to handle each one on every platform it ships to."
— Disability World engineering desk
---
title: Refreshable braille displays in 2026: a buyer's guide across 12 models
url: https://www.disabilityworld.org/articles/refreshable-braille-displays-buyers-guide/
description: An engineering primer comparing 12 refreshable braille displays in 2026 — Humanware Brailliant BI 40X / 20X, HIMS Polaris / QBraille XL, Orbit Reader 40 / 20, APH Mantis Q40 / Chameleon 20, Eurobraille Esys, Help Tech Activator, and Dot Pad — with a feature matrix, top-three picks.
author: Disability World
pubDate: 2026-05-22
tags: braille-displays, refreshable-braille, blindness, assistive-tech, hardware, buyers-guide
---
# Refreshable braille displays in 2026: a buyer's guide across 12 models
Buyer's guide · Refreshable braille displays 2026
Refreshable braille displays in 2026:
a buyer's guide across 12 models
Twelve refreshable braille devices share the 2026 market, with prices that span from a few hundred dollars to nearly nine thousand and use-cases that span pure reading to multi-line tactile graphics. This is the working buyer's score-sheet, written for blind professionals, parents, teachers, and the procurement officers who serve them.
12
models compared
7
attributes per model
approx. 8,800 USD
price span across the field
By Disability World hardware desk
11 min read
Updated May 2026
Foundation
1. What a buyer is actually buying when they buy a braille display
A refreshable braille display is a row — sometimes a grid — of small plastic pins that rise and fall under software control to form braille characters. The pins are usually driven by piezo-electric actuators, occasionally by stepper motors in newer low-cost designs, and increasingly by tactile-graphics modules in the highest-end devices. The display connects to a computer or phone over USB or Bluetooth, and a screen reader on that host feeds the characters out cell by cell. That is the whole mechanism, and it has been the whole mechanism for forty years.
What changes between models is everything that surrounds the pins: how many cells the device shows at once, how it pairs, what notetaker software it ships with, how long the battery lasts, which screen readers handle it cleanly, and how much it costs in the buyer's home currency after import tax and warranty extensions. A 40-cell display reads a full line of book prose at once; a 20-cell display fits in a coat pocket and reads a phone notification. A device with built-in notetaker apps can edit a document on a plane with no host computer at all; a "display-only" device cannot. These are not trivial differences. They are the difference between a tool that disappears into the workday and a tool that sits on a shelf.
40
cells in a typical full-line display — enough for one print line of book prose at standard line length.
20
cells in a typical pocket display — enough for a phone notification, a calendar slot, or a single line of code.
approx. 3
screen readers that almost every braille display ships drivers for: JAWS, NVDA, and VoiceOver (iOS + macOS).
i
The seven attributes that matter
1. Cell count. 20 cells (pocket), 32 cells (mid), 40 cells (full line), or multi-line for graphics.
2. Connectivity. Bluetooth (any version), USB-C, USB-A, and any extra ports such as SD card slots.
3. Screen-reader compatibility. JAWS, NVDA, VoiceOver — and how cleanly each one drives the device.
4. Notetaker apps. Whether the device runs editor, calendar, email, and book-reader apps natively without a host computer.
5. Battery. Hours of continuous reading or note-taking on a single charge.
6. Price. Manufacturer's suggested retail in USD, EUR, and GBP — before national assistive-tech subsidies, which can offset 100% of the cost in some countries.
7. Warranty. Standard manufacturer's warranty and what extensions cost.
"A braille display is not a screen replacement. It is a tactile keyboard for the language a blind reader already speaks fluently — and the price of that fluency is forty years of slow industrial progress."
— Disability World hardware desk
Landscape
2. The twelve models on the table
The 2026 market sorts into four tiers. The premium tier — Brailliant BI 40X, Polaris, Mantis Q40, Activator — sits in the 4,000 to 6,000 USD band and combines a 40-cell line with a full notetaker operating system. The mid tier — Brailliant BI 20X, QBraille XL, Chameleon 20, Eurobraille Esys — sits in the 2,500 to 4,000 USD band, with either fewer cells or a thinner notetaker layer. The low-cost tier — Orbit Reader 20 and 40 — sits below 1,500 USD by using stepper motors instead of piezo actuators. The graphics tier is a tier of one: the Dot Pad ships an entire two-dimensional pin grid and reads at a price that makes it an institutional rather than personal purchase.
The cards below summarise where each model sits and what kind of user it is built for. The dots reflect the device's overall fit for a typical blind professional buying their first or second display in 2026; they are not a quality score. A four-dot Orbit Reader is not a worse machine than a five-dot Brailliant — it is a less expensive machine that does fewer things, and that calculus is exactly the right one for many readers.
Brailliant BI 40X
40-cell premium notetaker (Humanware)
Strong fit for professionals who read and write long-form prose all day
Cells40
Overall fit
Brailliant BI 20X
20-cell pocket notetaker (Humanware)
Strong fit for commuters and students who pair to a phone
Cells20
Overall fit
HIMS Polaris
32-cell Android-based notetaker (HIMS)
Strong fit for users who want a full Android tablet under the cells
Cells32
Overall fit
HIMS QBraille XL
40-cell display with QWERTY keys (HIMS)
Strong fit for users who type QWERTY but read in braille
Cells40
Overall fit
Orbit Reader 40
40-cell low-cost display (Orbit Research)
Strong fit for students, schools, and emerging-market readers
Cells40
Overall fit
Orbit Reader 20
20-cell low-cost display (Orbit Research)
Strong fit for first-display buyers and braille-learning programmes
Cells20
Overall fit
APH Mantis Q40
40-cell display with QWERTY keyboard (APH)
Strong fit for adventitiously blind professionals who already type QWERTY
Cells40
Overall fit
APH Chameleon 20
20-cell display with Perkins keys (APH)
Strong fit for K-12 students learning braille input
Cells20
Overall fit
Eurobraille Esys
40-cell European-market display (Eurobraille)
Strong fit for French and Belgian readers using national subsidies
Cells40
Overall fit
Help Tech Activator
40-cell Android-based notetaker (Help Tech)
Strong fit for German-market readers and ATC integration users
Cells40
Overall fit
Dot Pad
Multi-line tactile graphics surface (Dot Inc.)
Strong fit for STEM students, museums, and institutional graphics work
The USD, EUR, and GBP prices in section three are manufacturer's suggested retail at time of writing and shift with currency, customs, and reseller margin in any given country. Driver and screen-reader compatibility, by contrast, is set by the vendor and changes only when a new firmware or screen-reader version ships. Treat the matrix as durable; treat the price column as a snapshot.
Reference
3. The feature matrix: model by attribute
The seven attributes from section one, scored against the twelve models from section two. Prices are manufacturer's suggested retail in USD, EUR, and GBP. Battery hours are vendor-reported continuous reading. "JAWS / NVDA / VoiceOver" entries name the screen readers that drive the device with a first-party or community driver; absence does not mean impossibility, only that the buyer will need a workaround.
Model
Cells
Connectivity
Screen-reader support
Notetaker apps
Battery
Price (USD / EUR / GBP)
Warranty
Brailliant BI 40X
40
USB-C, Bluetooth 5
JAWS, NVDA, VoiceOver
Editor, library, calendar, calculator
approx. 15 h
approx. 4,395 / 4,250 / 3,650
2 yr standard
Brailliant BI 20X
20
USB-C, Bluetooth 5
JAWS, NVDA, VoiceOver
Editor, library, calendar, calculator
approx. 15 h
approx. 3,095 / 2,990 / 2,580
2 yr standard
HIMS Polaris
32
USB-C, USB-A host, Bluetooth, Wi-Fi
JAWS, NVDA, VoiceOver (host) + Android native
Full Android notetaker suite
approx. 18 h
approx. 5,995 / 5,800 / 4,990
1 yr standard
HIMS QBraille XL
40
USB-C, Bluetooth
JAWS, NVDA, VoiceOver
None (display-only with QWERTY input)
approx. 20 h
approx. 3,795 / 3,690 / 3,150
1 yr standard
Orbit Reader 40
40
USB-C, Bluetooth, SD card
JAWS, NVDA, VoiceOver
Stand-alone reader, simple editor, file browser
approx. 20 h
approx. 1,495 / 1,490 / 1,280
1 yr standard
Orbit Reader 20
20
USB-C, Bluetooth, SD card
JAWS, NVDA, VoiceOver
Stand-alone reader, simple editor, file browser
approx. 20 h
approx. 699 / 720 / 620
1 yr standard
APH Mantis Q40
40
USB-C, Bluetooth 5
JAWS, NVDA, VoiceOver
Editor, library, calculator, terminal
approx. 14 h
approx. 2,495 / 2,490 / 2,140
2 yr standard
APH Chameleon 20
20
USB-C, Bluetooth 5
JAWS, NVDA, VoiceOver
Editor, library, calculator
approx. 14 h
approx. 2,195 / 2,190 / 1,880
2 yr standard
Eurobraille Esys
40
USB-C, Bluetooth
JAWS, NVDA, VoiceOver
Editor, calendar, address book
approx. 20 h
approx. 4,195 / 3,990 / 3,490
2 yr standard
Help Tech Activator
40
USB-C, Bluetooth, Wi-Fi, ATC
JAWS, NVDA, VoiceOver + Android native
Full Android notetaker suite
approx. 12 h
approx. 6,495 / 6,290 / 5,390
2 yr standard
Dot Pad
2,400 pins (300 cells, multi-line)
USB-C, Bluetooth, iPad pairing
VoiceOver (primary), JAWS via host bridge
None (tactile graphics surface, host-driven)
approx. 8 h
approx. 8,900 / 8,490 / 7,290
1 yr standard
Honourable mention
Varies
Varies
Varies
Niche or loaner-only
Varies
Varies
Varies
+
How to read the matrix
The matrix sorts naturally into three reading orders. Read down the price column to compare cost. Read down the notetaker column to compare independence from a host computer. Read down the screen-reader column to confirm a model talks to the software the buyer already runs. The matrix does not score the cells themselves — every device on this list uses braille cells that meet the standard 2.5 mm dot diameter and 2.5 mm cell spacing, and a side-by-side touch test rarely separates them.
Recommendations
4. Top three picks for 2026, by user profile
The matrix names twelve devices; most buyers should be choosing among three. The picks below cover the three users who, in 2026, account for the bulk of new purchases: a working professional who reads and writes long-form prose, a student or commuter who lives in a phone, and a school or rehabilitation programme buying displays in volume.
For the working professional
Brailliant BI 40X — approx. 4,395 USD
Full 40-cell line, robust notetaker, clean JAWS and NVDA drivers, two-year warranty
Why this oneBest-supported full-line device
Runner-upAPH Mantis Q40 (QWERTY input)
Skip ifBudget cap below approx. 3,000 USD
For the student or commuter
Brailliant BI 20X — approx. 3,095 USD
20-cell pocket form factor, full Bluetooth 5 pairing to iPhone or Android, light enough to carry every day
Why this oneBest phone-paired pocket display
Runner-upAPH Chameleon 20 (US K-12 buyers)
Skip ifYou read long-form prose daily — go 40
For schools and programmes
Orbit Reader 20 — approx. 699 USD
Lowest unit price on the market by a wide margin, durable, SD-card stand-alone reading, simple repair pathway
Why this oneHighest "displays per grant dollar" ratio
Runner-upOrbit Reader 40 (older students)
Skip ifNotetaker apps are a hard requirement
The right braille display in 2026 is rarely the most expensive one a buyer can justify; it is the cheapest one that actually clears every requirement on the buyer's list.
i
National subsidies change the maths
In Germany, the Krankenkasse statutory health-insurance route can cover an Activator outright for an employed reader; in France the MDPH route covers an Esys; in the UK Access to Work covers most premium devices up to a project cap; in the US Vocational Rehabilitation and the Department of Veterans Affairs cover the Brailliant and Mantis routinely. A buyer paying out of pocket should treat the price column as binding; a buyer eligible for a national subsidy should treat it as a starting line.
Frontier
5. The Dot Pad question: multi-line graphics arrives
Eleven of the twelve devices on the list are single-line braille displays — they show one row of text at a time. The twelfth, the Dot Pad, is something else entirely: a 300-cell grid arranged in ten lines of thirty cells, plus a 2,400-pin tactile graphics area that can render a chart, a map, a maths diagram, or a UI mock-up as a raised image. It is the most significant industrial change in refreshable braille hardware in twenty years, and it lands at a price that is, by personal-purchase standards, prohibitive — but by institutional standards, very much affordable.
The Dot Pad is not a replacement for a Brailliant or a Mantis. It is a complement. A blind STEM undergraduate who has a Mantis Q40 for prose and a Dot Pad for diagrams reads the textbook and the diagrams from the same desk; a museum that puts a Dot Pad next to a visual exhibit can show the exhibit to blind visitors in real time; a school that puts a Dot Pad in a maths lab gives its blind students access to graphs that previously had to be embossed on swell-paper and posted overnight. None of this is theoretical in 2026 — the device has shipped in volume to public-sector buyers in Korea, Japan, the US, and the UK, with growing European institutional purchases.
What the Dot Pad replaces
Swell-paper embossed graphics, mailed overnight from a transcription service at approx. 15-40 USD per diagram, with a 24- to 48-hour turnaround and no way to revise the image after it is printed. Sufficient for static textbook figures; useless for the live charts a working analyst produces.
What the Dot Pad enables
Live tactile rendering of any graphic the host sends — pulled from a textbook, generated by a chart library, transcribed in real time from a slide. Refresh in seconds, revise in place, share between students by re-sending the file. The same diagram the sighted class sees, on the same minute.
!
Buyer-beware on the Dot Pad
The Dot Pad's price puts it outside almost every personal-purchase budget — at approx. 8,900 USD it is more than triple the Mantis Q40 and approaches the cost of a small car. For a private buyer the question is rarely whether the Dot Pad is good (it is) but whether an institution will buy one and let the buyer use it. STEM students should ask their disability services office; working professionals should ask their employer's reasonable-accommodation budget; parents should ask their child's school district.
Playbook
6. The decision tree: from "I need a braille display" to "I bought this one"
The matrix and the picks above are the data. The tree below is how to use them. Six questions in order; the answer to each one removes models from the list. Walk the tree top to bottom and most buyers land on one or two finalists.
1
Do you need a 40-cell line, or will 20 cells do?
Read long-form prose, write reports, work in spreadsheets — go 40. Read phone notifications, pair with an iPhone, carry it daily — go 20. If you are not sure, go 40: the price difference is real but the experience difference is larger, and most readers who start at 20 cells trade up within two years. This single question removes half the list.
2
Do you need built-in notetaker apps, or only a host-driven display?
If the device must work on a plane with no laptop — editor, calendar, calculator, book reader — you need a notetaker. That points to Brailliant BI 40X / 20X, HIMS Polaris, Help Tech Activator, APH Mantis Q40 / Chameleon 20. If the device only ever pairs to a phone or computer, the QBraille XL, Orbit Reader 20 / 40, and Eurobraille Esys are display-first machines and are cheaper for that reason.
3
Which screen reader do you actually use?
JAWS-only buyers are well served by any of the premium devices but should confirm the driver version on Freedom Scientific's compatibility page. NVDA users have the widest hardware support and the best community-driver story; the Orbit Reader and Brailliant lines work particularly cleanly. VoiceOver users on iPhone should buy a device with current Bluetooth 5 — Brailliant 20X, Mantis Q40, and Chameleon 20 are the safest bets.
4
Do you prefer Perkins-style braille keys or QWERTY?
Most blind readers who learned braille as children prefer Perkins-style input on the device. Most adventitiously blind professionals who already touch-type prefer QWERTY. If QWERTY is your answer, the field narrows sharply to the APH Mantis Q40 and the HIMS QBraille XL. If Perkins is your answer, almost everything else qualifies.
5
What is your budget after subsidies?
Under approx. 1,500 USD: Orbit Reader 20 or 40 are the only options. Between approx. 2,000 and 3,500 USD: APH Mantis Q40 and Chameleon 20 dominate. Between approx. 3,500 and 5,000 USD: the Brailliant BI 40X, QBraille XL, and Eurobraille Esys are the field. Above approx. 5,000 USD: Polaris, Activator, and (institutional) Dot Pad become reachable. Apply national subsidies before answering this question; the maths changes entirely once a Krankenkasse, MDPH, Access to Work, or Vocational Rehabilitation route is open.
6
Do you need tactile graphics, or only text?
Almost all buyers need only text. The Dot Pad enters the picture for STEM students, cartographers, designers, museum educators, and anyone working with charts and maps every day. If the answer here is "only text", skip the Dot Pad and use the budget for a better single-line device plus a two-year warranty extension. If the answer is "graphics are central to my work", the Dot Pad is the only device on this list that meets the requirement.
i
Try before you buy
Every device on this list has a different button layout, a slightly different cell pitch, and a slightly different keypress feel. Buyers who can try a device before purchase — at a national federation conference, a local rehabilitation agency, or a vendor demo — make better decisions than buyers who order sight-unseen. In the US the CSUN, ATIA, and NFB conventions host vendor booths every spring; in Europe the Sight Village event in Birmingham and the SightCity event in Frankfurt do the same. Ten minutes of hands-on time is worth more than ten hours of YouTube review.
Conclusion: the right braille display is the one that disappears
The twelve devices on this list represent the working state of refreshable braille hardware in 2026. None of them is bad. Most of them are very good. The market has matured to the point where the buyer's question is no longer "which device works?" — every device on this list works — but "which device fits the reading life I actually live?" The answer is rarely the most expensive device on the table, and rarely the cheapest. It is the one that disappears into the workday because every requirement on the buyer's list is already met, every screen reader on the buyer's machines is already driven cleanly, and the warranty paperwork is already in a drawer somewhere.
The decision tree in section six gets most buyers to a finalist in twenty minutes. The matrix in section three gets them to a confident purchase decision in an hour. The top-three picks in section four cover the three users who do most of the buying. And the Dot Pad — quietly, expensively, institutionally — represents the first real step out of the single-line cage that has constrained tactile reading since the first piezo-electric cell shipped in 1979. The line is still where most of us read. The grid is where some of us will read next.
"The right braille display is the one that disappears into the workday. Every device on this list works; only one of them is yours."
— Disability World hardware desk
---
title: Screen-reader learning paths: how sighted developers can become fluent
url: https://www.disabilityworld.org/articles/screen-reader-learning-paths/
description: A staged learning path that takes a sighted developer from screen-reader novice to genuinely fluent — which reader to start with, the first-week monitor-off exercises, the developer shortcuts almost nobody teaches, and honest time-to-fluency benchmarks.
author: Disability World
pubDate: 2026-05-22
tags: screen-readers, learning, developers, nvda, voiceover, testing, education
---
# Screen-reader learning paths: how sighted developers can become fluent
Engineering primer · Sighted-developer SR fluency
Screen-reader learning paths:
how sighted developers can become fluent
"I tested it with VoiceOver" is the single most overstated claim in frontend accessibility. We took apart what fluency actually looks like — not familiarity, fluency — and built a staged plan that gets a sighted developer to genuine confidence in about forty hours of practice, starting with the reader pairing that actually pays off and ending with the developer-mode shortcuts that almost nobody teaches.
approx. 10h
to "useful"
approx. 40h
to semi-fluent
2
readers to start with
By Disability World engineering desk
12 min read
Updated May 2026
Motivation
1. Why bother — and what fluency actually means
Almost every accessibility programme we audit reports the same number: ninety-something percent of frontend developers say they "test with a screen reader." Ask them to demonstrate, and the demo is usually the same three keystrokes — turn it on, tab through the page, turn it off. That is not testing. That is checking a box.
The reason this happens is structural, not lazy. A screen reader is not a tool you can pick up the way you pick up a new linter. It is a different interaction model with its own modal state, its own shortcut grammar, and a set of conventions that make sense only after you have used it for several hours of real work. Until you cross that threshold, the tool tells you almost nothing — and worse, it tells you things that are wrong, because the announcements you hear depend on the reader's mode, the browser's accessibility tree, and the platform's IME layer in ways that are not obvious from outside.
Fluency, for our purposes, is the point at which you can hand a colleague a broken component, take their keyboard, and reproduce the bug with the screen reader running — without looking at the screen, without referring to a cheat sheet, and without making the announcement worse than it would be in real use. Familiarity is the point at which you have heard a screen reader. The gap between the two is roughly thirty to thirty-five hours of deliberate practice.
i
What this article is not
This is not a substitute for testing with disabled users. A sighted developer using a screen reader is approximating a workflow that a daily user has internalised over years. The point of fluency is not to replace user testing; it is to catch the obvious bugs before user testing, so the user-testing session is spent on the subtle ones.
Choice
2. Choose your screen reader — and skip JAWS until later
The market has three screen readers that matter for desktop web work: NVDA on Windows, VoiceOver on macOS and iOS, and JAWS on Windows. Each one has a body of users large enough that ignoring it would be a real bet, and each one announces the same markup slightly differently. A fluent developer can drive at least two of them.
Our recommendation, after watching dozens of developers cross the threshold, is unambiguous: start with NVDA on Windows and VoiceOver on macOS. Both are free. Both are pre-installed (VoiceOver) or installable in under five minutes (NVDA). Both are used by enough real users — NVDA holds approx. 65% of Windows screen-reader market share in the most recent WebAIM survey, VoiceOver dominates mobile and a meaningful share of macOS — that what you learn transfers immediately to bugs you can ship a fix for. JAWS is the third tool, not the first, even though it is still the screen reader with the largest enterprise install base. Three reasons.
NVDA
NV Access · Windows · free
approx. 65% of Windows screen-reader market share (WebAIM 2024)
CostFree, donation-supported
Install timeUnder 5 minutes
Learning curve
Why start hereClean modes, transparent log, large real-user base
VoiceOver
Apple · macOS & iOS · pre-installed
Default on every Mac and iPhone; dominant on mobile
CostFree, ships with the OS
Install timeAlready installed
Learning curve
Why pair with NVDARotor model differs from PC-cursor model; you learn both worlds
JAWS
Freedom Scientific · Windows · paid
Largest enterprise install base, especially government and finance
CostHome licence approx. $95/yr, pro tier higher
Install time30+ minutes; activation required
Learning curve
Why skip firstSame Windows mental model as NVDA, but heavier and licence-gated
The three reasons to skip JAWS at the start are pedagogical, not political. First, JAWS and NVDA share a mental model — Windows browse mode versus focus mode, the same Insert-based command prefix, the same virtual buffer — so once you can drive NVDA, ninety percent of the JAWS commands you actually need are a glossary lookup away. Second, JAWS has accumulated decades of "smart" inference: it tries to fix bad markup before the user hears it, which means a bug that JAWS papers over will still ship to NVDA users. NVDA's deliberately conservative behaviour makes it the better reference reader when you are trying to learn what is broken. Third, JAWS's licensing friction — activation, the forty-minute trial mode that nags every reboot — is a learning-tax you do not need to pay until you are confident enough to spend it.
VoiceOver pairs with NVDA rather than competes with it because the two readers represent the two dominant interaction models. NVDA (and JAWS) use the "PC cursor" model: a virtual buffer that lays out the page as a linear document and a separate focus that follows tab order. VoiceOver uses a single VoiceOver cursor that lives on top of the focus, navigated by the rotor and by VO+arrow keys. A developer fluent in only one model will write code that announces well in their reader and badly in the other. Learning both at once is the only reliable way to feel the difference.
"Pick the two free readers. Spend forty hours. You will catch more accessibility bugs in the next quarter than your last three vendor audits combined."
— Engineering lead, fintech platform that retired its overlay in 2025
Week 1
3. Week 1 — monitor off, hands on the keyboard
The week-one programme has one rule: turn the monitor off. Not dimmed, not minimised, not "I'll close my eyes" — physically off, or covered with a piece of card if your display is the only one in the room. The point is to remove the option of cheating. A sighted developer's instinct, the moment a screen reader says something confusing, is to glance at the screen and resolve the ambiguity visually. That instinct is the single largest reason "I tested with a screen reader" does not catch real bugs.
Plan for three sessions of about ninety minutes each in week one, with at least a day between sessions so the muscle memory has time to consolidate. Each session has one job. The first builds the basic command grammar. The second forces a real interaction. The third tests retention under a small amount of stress.
1
Session 1 — install, configure, browse the homepage
Install NVDA (or open VoiceOver on macOS). Turn off speech synthesis politeness if you can — you want fast, mechanical speech, not the friendly default. Open a major news site, monitor off. Spend 45 minutes pressing the arrow keys and listening. Spend the second 45 minutes pressing H (next heading), K (next link), and F (next form field) and noticing how the page is structured. Do not navigate anywhere yet.
2
Session 2 — write your name into a form
Open a contact form on your own company's site, monitor off. Tab to the name field. Type your name. Tab to the email field. Type a fake email. Tab to the submit button. Press space. If you cannot find the submit button without looking, that is information: your form's tab order is broken, or its labels are broken, or both. Note the failure. Do not fix it yet — fixing it before you have heard ten more forms is premature optimisation.
3
Session 3 — buy something cheap
Open an e-commerce site you have never visited, monitor off. Find a product under five dollars. Add it to the cart. Reach the payment step. Stop before you pay — but go all the way to the payment form. This is the session that breaks people. You will discover that "fluent enough to test" and "fluent enough to use" are different thresholds. The first session of pure listening was just rehearsal; this is the first session of doing.
!
If session 3 takes more than 90 minutes
Stop. You have learned the lesson you needed to learn for the week. The lesson is not "I am bad at screen readers" — it is "this site is genuinely difficult to use without sight." Most major retail sites take a screen-reader user thirty to sixty minutes longer than a sighted user to complete a checkout. You are now feeling that gap.
Weeks 2–4
4. Weeks 2 to 4 — forms, navigation, and the mode trap
The second through fourth weeks of practice should add up to roughly twenty hours of work — two ninety-minute sessions a week, plus a small amount of incidental use while you do your day job. The goal in this stretch is to internalise the two things that confuse new screen-reader users more than anything else: the distinction between browse mode and focus mode, and the difference between what the rotor sees and what tab order sees.
Browse mode (NVDA, JAWS)
Focus mode (NVDA, JAWS)
VoiceOver (single mode)
Arrow keys
Navigate the virtual buffer
Sent to the focused control
Always navigate the VoiceOver cursor
Tab
Moves focus and stays in browse
Moves focus and stays in focus
Moves focus; VoiceOver cursor follows
Letter shortcuts (H, K, F)
Quick navigation
N/A
Replaced by the rotor (VO+U)
When it switches
Default for most pages
Auto on contenteditable, custom widgets
Never — there is no mode
How to force it
NVDA+Space
NVDA+Space (toggles)
Not applicable
The single most common confusion in week two is the moment a developer presses an arrow key in NVDA, expects the virtual buffer to move, and instead hears the focused combobox open its options list. That is browse mode switching to focus mode automatically because the focus landed on an element that NVDA classifies as an "application" widget. New developers experience this as the reader misbehaving. It is not — it is the reader doing exactly what the spec asks. Once you have heard it ten or fifteen times you stop being surprised; until then, plan to be surprised approximately every other session.
The week-three pattern is forms. Build a private testing page with eight or ten controls: a required text input with an inline error, a date picker, a multi-select, a custom-styled checkbox, a disabled button that becomes enabled, a "show password" toggle, a phone-number field with a country-code selector, and a submit button that triggers a server-side validation summary. Monitor off, navigate through it five times — first with NVDA in browse mode, then NVDA in focus mode, then NVDA again with the verbose announcement setting turned up (Insert+Z, more on that in section five), then VoiceOver with the rotor, then VoiceOver without the rotor. The same form will sound different five times. That is what fluency feels like from the inside: noticing that the same markup tells five different stories, and being able to predict in advance which one will play.
Week four is navigation. Take a real, complex site — a documentation portal, a workplace dashboard, an e-commerce category page — and try to find a specific piece of information using only screen-reader shortcuts. Use H to jump headings. Use D (NVDA) or VO+U then "Landmarks" (VoiceOver) to jump landmarks. Use 1 through 6 to jump to a particular heading level. By the end of week four, the navigation shortcuts should be reflexes rather than choices, the way tab and shift-tab already are.
"The day you realise that pressing H twenty times feels faster than tabbing thirty times is the day you stop being a sighted developer pretending and start being a developer who can navigate."
— Mid-career frontend engineer, third month of NVDA practice
Developer mode
5. Development-mode shortcuts almost nobody teaches
Once the user-mode commands are reflexes, the next jump is into the developer-facing surfaces of each reader. These are the modes and shortcuts the manuals bury — partly because they are aimed at developers, partly because they are noisy enough that a daily user would not want them on. Three are worth knowing immediately.
NVDA · Speech viewer + verbose announcement
Tools menu → Speech viewer; Insert+Z toggles verbosity
Visual transcript of everything NVDA is saying, plus expanded role announcements
What it gives youA scrollable log of every announcement, so you can verify what the reader actually said versus what you thought it said
When to use itBug reproduction, automated-test comparison, training colleagues
NVDA · Log inspector (NVDA+F1)
Developer-info pop-up on the focused element
Inspect what NVDA sees on the current element — role, states, value, description, accessible name
What it gives youThe accessibility tree NVDA built, not the DOM your dev tools show
When to use itWhen the page looks right in dev tools but reads wrong in NVDA
VoiceOver · Web Rotor (VO+U) and Web Item Settings
macOS & iOS · the developer's accessibility tree
Hierarchical list of headings, links, landmarks, form controls, web spots, and tables — exactly as VoiceOver indexed them
What it gives youA second opinion to NVDA's log inspector: if both readers agree, the bug is in your markup, not the reader
When to use itCross-reader bug triage, especially for landmark and heading structure
Two further habits will save more time than any single shortcut. First, leave NVDA's speech viewer pinned on a second monitor (or in a corner of your one monitor) while you develop. The verbatim log of every announcement is to screen-reader work what the dev-tools console is to JavaScript: the difference between guessing and knowing. Second, learn to read the accessibility tree in your browser's dev tools — Chrome's Accessibility pane, Firefox's Accessibility Inspector, Safari's Audit tab. The reader announces what the accessibility tree contains, not what the DOM contains, and the two diverge often enough that you cannot debug live regions, ARIA, or shadow DOM without reading the tree directly.
A confusion to flag now, because it eats hours in weeks two and three: reading mode versus focus mode is not the same axis as "the page is interactive" versus "the page is a document." NVDA switches into focus mode automatically when the focus lands on a control with role="application", or on a contenteditable, or on certain custom widgets that the reader heuristically classifies as interactive — regardless of whether the page is mostly static. Conversely, a richly interactive single-page app whose root element is a main landmark and whose widgets are well-marked-up native buttons will stay in browse mode for almost all of a user's session. The mode is a property of the focused element, not a property of the page.
★
The single most useful keystroke
NVDA+Space toggles between browse mode and focus mode manually. When something sounds wrong, this is the first thing to try — half the time, the reader was in the mode you weren't expecting, and toggling once will tell you whether the bug is in the mode logic or in the markup.
Benchmarks
6. Time-to-fluency — honest benchmarks
The numbers below come from informal tracking of about eighty developers — frontend engineers, QA leads, accessibility specialists in training — across three years of corporate workshops and one-on-one mentoring. They are not a research study. They are good enough to plan against. Two assumptions: deliberate practice (monitor off, real tasks, not "I left NVDA running in the background while I coded"), and a fixed reader pairing (NVDA on Windows and VoiceOver on macOS).
approx. 3h
to feel the basic shape — installed, basic commands, can navigate a homepage with the monitor off
approx. 10h
to "useful" — can drive a real form, can reproduce a colleague's bug report, can be trusted with a quick test
approx. 25h
to comfortable — both readers feel familiar; mode confusion is rare; rotor and log inspector are reflexes
approx. 40h
to semi-fluent — can demonstrate a bug live, can review another developer's screen-reader work credibly
"Semi-fluent" is the realistic destination for most sighted developers and is, in practical terms, all you need to be a good contributor to an accessible product. Genuine fluency — the level at which you could plausibly substitute for a daily screen-reader user during a usability review — is more like one hundred and fifty hours and a year of incidental practice, and most working developers do not need it. Aim for semi-fluent, schedule the forty hours, and accept that anything beyond that comes from doing the day job with a reader running and a willingness to slow down.
One last benchmark to set expectations honestly: the developers who plateau, in our experience, plateau between the ten-hour and twenty-hour mark. The cause is almost always the same — they stop turning the monitor off. They tell themselves that they are now "good enough" to test with the screen on, the screen reader running in the background, and visual confirmation available whenever the audio is ambiguous. They are not. The sixteen hours between "useful" and "comfortable" require the monitor off because that is the stretch where the reader's announcements become information rather than noise. Without that pressure, the brain reverts to vision and the reader's voice fades into wallpaper. If you find yourself slowing down, it is almost always the monitor.
"The forty-hour version of you can find more screen-reader bugs in a one-hour pre-release sweep than your last automated audit. That is not a high bar. That is what testing with a screen reader was always supposed to mean."
— Disability World engineering desk, after watching the curve play out a few dozen times
Conclusion: the path is short, the discipline is not
The reason "test with a screen reader" produces such weak results across the industry is not that the tool is hard to learn — forty hours is genuinely not a lot of time — but that the learning is uncomfortable in a specific way. Turning the monitor off makes a sighted developer feel inept in a way that is unusual in our profession. We are accustomed to being the people who figure things out; the screen reader makes us, for a few hours at a stretch, beginners again. That discomfort, and not the keystrokes, is the actual obstacle.
The path through is the one above: NVDA and VoiceOver, three sessions in the first week with the monitor off, forms and modes in weeks two through four, developer-mode shortcuts as soon as the user-mode shortcuts are reflexes, forty hours total before you can be trusted with a serious pre-release sweep. None of it is novel. The work the industry has not done is treating it as work — scheduling the hours, defending them from other commitments, accepting that the first ten of those hours will feel useless until they suddenly do not.
If you ship a frontend, the version of you on the far side of those forty hours is a substantially better engineer than the version that started, in ways that will show up not only in your accessibility work but in your understanding of focus order, of progressive enhancement, of what the browser is actually doing under the hood. The screen reader is the cheapest distributed-systems lesson available to anyone who writes for the web. The price is the monitor off and a few weekends.
"You will not become a screen-reader user. You will become a developer who can hear what your code sounds like to one. That is enough — and most of the industry does not yet have it."
— Disability World engineering desk
---
title: The screen reader roadmap for 2026: JAWS, NVDA, VoiceOver, TalkBack
url: https://www.disabilityworld.org/articles/screen-reader-roadmap-2026/
description: Four screen readers shape almost the entire assistive-tech web.
author: Disability World
pubDate: 2026-05-22
tags: screen-readers, jaws, nvda, voiceover, talkback, assistive-tech, tech-news
---
# The screen reader roadmap for 2026: JAWS, NVDA, VoiceOver, TalkBack
The screen reader roadmap for 2026: JAWS, NVDA, VoiceOver, TalkBack
Almost every blind or low-vision user on the public web in 2026 is interacting with one of four screen readers: JAWS, NVDA, Apple's VoiceOver, or Google's TalkBack. Together they cover roughly nineteen out of every twenty assistive-tech sessions across desktop and mobile. This field guide catalogues each of the four — and adds a smaller, fifth exhibit for the three emerging players (Narrator, ChromeVox, Orca) that genuinely matter at the margins.
The previous installments in this technology series compared screen readers against one another in a single matrix or argued for a specific testing methodology. That comparative view is useful when an engineer is deciding which reader to test against next. It is less useful when the question is the longer one: who actually uses which reader, why, and what is each vendor doing for 2026? This guide takes the opposite view. It works from one reader outward at a time, with an identical anatomy for each entry, so the catalogue can be read top-to-bottom or jumped to by reader name.
Every exhibit below records the same seven things in the same order: 2026 market share, platforms, last major release, ARIA 1.3 support depth, distinctive features, known limitations, and the adoption pattern — who chooses this reader and why. The final mini-exhibit covers Narrator, ChromeVox, and Orca together because each is meaningful only inside its own ecosystem.
Evidence index · Cat. 2026.05
4 dominant screen readers · ranked by 2026 desktop+mobile share
n approx. 3,800 WebAIM SR users survey #10 (2025) + Google / Apple telemetry summaries
Desktop figures are anchored to the most recent WebAIM screen-reader user survey (cycle #10, published 2025), in which respondents are asked which reader they use most often. Mobile shares are aggregated from Apple iOS accessibility settings telemetry summaries and Android accessibility-services usage published by Google's accessibility team through Q1 2026. Shares are directional, rounded, and overlap is possible because many respondents use multiple readers across devices.
Where the data comes from
The four shares above are reconciled from three independent sources. WebAIM's Screen Reader User Survey #10 is the canonical desktop reference: roughly 3,800 respondents, self-reported primary reader. Apple's published accessibility-impact summaries and Google's Android accessibility quarterly post cover the mobile side. Where the two diverge — particularly on the question of how often desktop respondents also use a mobile reader — we have favoured the WebAIM dataset for primary-reader attribution and the platform telemetry summaries for breadth. Numbers are directional. Almost no respondent uses only one reader, and the modern norm is a desktop reader plus VoiceOver on a phone.
In 2026, "supporting screen readers" means supporting these four. Everything else combined is below seven percent, and most of it is a thoughtful niche rather than a viable testing target.
Part I · The four readers that cover the web
Identical anatomy, four very different engineering choices
NVDA and JAWS share the Windows desktop. VoiceOver dominates mobile and quietly carries the Mac. TalkBack carries Android. Each one solves the same accessibility problem from a different starting point — open source, enterprise license, platform integration, or Android's heterogeneity — and each one carries different bugs.
E·01
NVDA — NonVisual Desktop Access
2026 market share
approx. 37%of desktop screen-reader respondents (WebAIM #10) name it as their primary reader
approx. 75%say they use it at least occasionally — the highest cross-use number in the survey
Platforms
Windows only. NV Access publishes a portable build that runs from a USB stick without installation rights, which has made NVDA the universal lab and audit reader. There is no macOS, no Linux, no mobile build, and there is no roadmap promising any of them.
Last major release
NVDA 2025.4 shipped in late 2025 with consolidated Chromium UIA bridge improvements, a remote-access feature now built into the core (no add-on required), and expanded language-switching defaults. The 2026.1 release line is in preview as of this writing and is expected to ship the formal ARIA 1.3 conformance pass tied to the W3C's December 2025 candidate-recommendation update.
ARIA 1.3 support depth
Strong and improving. NVDA has historically been the first reader to pick up new ARIA roles, properties, and states because its release cadence is faster than JAWS's and its codebase is open. As of 2025.4, the new aria-actions property is read but with placeholder verbiage, the long-awaited aria-brailleroledescription is honoured on refreshable braille output, and the expanded role="comment" + role="suggestion" pair used by collaborative document editors is fully exposed.
Distinctive features
Open-source and free under GPL. A first-class Python add-on API that has produced an active third-party extension ecosystem — sound-themed audio cues, integrated OCR for image-of-text content, web developer toolkits, custom language profiles. Built-in remote access for tech support. A community-driven, transparent bug tracker on GitHub.
Known limitations
NVDA's speech synthesis defaults can sound noticeably more synthetic than JAWS's premium voices on first install, which sometimes biases new-user comparisons. Enterprise deployment requires more in-house knowledge than JAWS because there is no commercial support contract by default. Some legacy Windows desktop applications that rely on MSAA-only accessibility APIs are read with a perceptible delay relative to JAWS.
Adoption pattern
Chosen by users with strong individual cost sensitivity (it is free), by accessibility professionals who need a portable build for audits, by developers who need the open codebase and the Python add-on API, and increasingly by enterprises in the Global South where the JAWS license cost is not realistic. NVDA's rise to the desktop top spot in WebAIM #10 is the headline accessibility-software story of the last five years.
approx. 31%of desktop respondents name JAWS as their primary reader (WebAIM #10)
approx. 60%say they use it at least occasionally — slightly behind NVDA
Platforms
Windows only. Vispero (which owns JAWS, ZoomText, and Fusion) has consistently declined to commit to a macOS or Linux port. JAWS is the dominant reader inside US federal-government, US state-government, US healthcare, and US enterprise procurement environments, where Vispero's volume-license agreements are deeply embedded.
Last major release
JAWS 2026 shipped in January 2026 with formal ARIA 1.3 alignment, an extended PDF-tagged-content reading model, a refreshed integration with FSReader and FSCast, and improved Microsoft Edge support in the Chromium UIA bridge. JAWS now ships on an annual major-release cadence with rolling monthly patches.
ARIA 1.3 support depth
Comprehensive but slightly more conservative than NVDA. JAWS 2026 fully supports aria-actions, the comment + suggestion roles, expanded aria-errormessage handling on grouped form controls, and the new aria-brailleroledescription braille output. Where JAWS lags slightly is the new ARIA 1.3 keyboard-shortcut hint property: support landed but with verbose announcements that some users disable manually.
Distinctive features
Vocalizer Expressive premium voices that many users describe as the most natural-sounding on Windows. Industry-standard scripting language for application customisation, used inside large enterprise deployments for line-of-business applications. Tightly integrated training and certification programme through Freedom Scientific. Volume-license discounts for institutional procurement. Convergence with the ZoomText magnifier through the Fusion product.
Known limitations
The annual license cost (approx. USD 1,200 commercial, USD 90 annual subscription) is the single largest barrier to adoption outside North American and Western European institutional environments. The scripting engine, while powerful, is a barrier to entry for casual customisation compared to NVDA's Python add-on API. JAWS's release cadence is slower than NVDA's, so emerging ARIA features sometimes arrive months after they have already shipped in NVDA stable.
Adoption pattern
Chosen by users inside institutional procurement environments (US federal/state government, large US enterprises, schools and universities), by users who specifically depend on the JAWS scripting engine to make a custom application accessible, and by long-tenured users who learned JAWS in the 1990s or 2000s and have a deep customisation profile they would rather not rebuild. JAWS continues to lose ground to NVDA on the cost-sensitive end of the market and to hold ground on the institutional end.
VendorVispero / Freedom Scientific (US)LicenseCommercial · annual license or perpetual
E·03
VoiceOver — Apple's platform screen reader
2026 market share
approx. 71%of mobile screen-reader users globally (iOS-dominant markets)
approx. 9%of desktop respondents name macOS VoiceOver as their primary reader
Platforms
iOS, iPadOS, macOS, watchOS, tvOS, and visionOS. VoiceOver is the canonical example of a screen reader designed as an integrated platform service rather than a third-party application. On mobile it has effectively no competition inside the Apple ecosystem; the only Apple-platform alternative is the much smaller-deployed Hindenburg or third-party low-vision tools.
Last major release
VoiceOver ships inside the operating system, so its release cadence tracks iOS, iPadOS, and macOS. The 2026 cycle introduced the new "Page Summary" Apple-Intelligence feature that produces an LLM-generated landing-summary of a web page before the user begins reading; expanded VoiceOver rotor verbosity controls; and a refreshed braille input table that is the largest braille refresh shipped on any platform in 2026.
ARIA 1.3 support depth
Strong on iOS Safari and macOS Safari, weaker outside Safari. Within Safari, ARIA 1.3 features such as the new comment / suggestion roles, expanded error-messaging on grouped form controls, and aria-brailleroledescription on iOS braille displays are honoured. VoiceOver inside Chrome on iOS exposes a subset of ARIA properties — a long-standing limitation that is in part a WebKit-platform constraint rather than a VoiceOver bug.
Distinctive features
Free, deeply integrated with the operating system, and the platform reader for visionOS — the first commercially shipped screen reader for a head-mounted display. The Rotor (iOS gesture, macOS rotary control) is a uniquely fast navigation model that experienced users frequently describe as the single most productive feature of any reader. Apple Intelligence Page Summary and image descriptions ship on-device in 2026, which keeps text and image content off the cloud.
Known limitations
VoiceOver's verbosity defaults are aggressively chatty in 2025–26, and many users adjust them substantially before daily use. Behaviour can differ in important ways between Safari and Chromium-based browsers on iOS, which means web developers cannot test once and assume parity. macOS VoiceOver lags iOS VoiceOver on several feature lines, and macOS desktop respondents in WebAIM #10 still rank it third behind NVDA and JAWS.
Adoption pattern
Chosen — by default — by anyone on an iPhone, iPad, Mac, Apple Watch, Apple TV, or Vision Pro. Within the screen-reader population, VoiceOver users skew younger, more mobile-first, and more iOS-centric than the average JAWS or NVDA user. VoiceOver's lock on the mobile side is structural: Apple's ecosystem share among US blind / low-vision users is even higher than its share in the general population.
approx. 26%of mobile screen-reader users globally (Android-dominant markets)
more than halfof mobile reader users outside North America and Western Europe, where Android share is higher
Platforms
Android (mobile and tablet). TalkBack ships with Google's Android Accessibility Suite and is also bundled into Samsung's One UI and other OEM Android skins. There is no desktop or ChromeOS variant — ChromeOS uses ChromeVox (E·05) — and there is no separate Wear OS reader; Wear OS uses a TalkBack-derived experience.
Last major release
TalkBack 15.0 shipped through Google Play in early 2026 with multi-finger gestures inherited from the iPad-VoiceOver model, expanded Gemini Nano on-device image-description support, a refreshed reading-controls menu, and the long-requested ability to navigate web headings with the same multi-finger swipe gesture used inside Android apps.
ARIA 1.3 support depth
Improving but historically the weakest of the four. TalkBack reads web content via Chrome on Android, and Chrome on Android's accessibility-tree exposure has consistently lagged Chrome on desktop. 2026 closes much of that gap: the new comment / suggestion roles are exposed, aria-actions is read with placeholder verbiage, and grouped form-error handling has been refined. Some ARIA 1.3 properties — particularly the new keyboard-hint family — are still announced inconsistently in some Android web views.
Distinctive features
Free, integrated into the operating system, and the global reader leader by absolute installed-base count outside North America and Western Europe. Gemini Nano on-device image descriptions ship by default in 2026 for new images and previously-unlabelled images. Reading-controls menu is the most-customisable on-device verbosity control of any of the four readers. Open-source, with the code available in the Android Open Source Project.
Known limitations
OEM fragmentation means TalkBack behaviour can vary between Samsung, Pixel, Xiaomi, and other Android handsets — particularly on focus management, gesture handling, and certain custom-view ARIA mappings. Web-content reading inside non-Chrome Android browsers can produce different results than VoiceOver-in-Safari parity testing would suggest, and remains the single largest known cross-platform testing gap in 2026.
Adoption pattern
Chosen — by default — by anyone on an Android phone or tablet. Within the screen-reader population, TalkBack users skew significantly younger and more concentrated outside the US than VoiceOver users. For any product whose user base extends meaningfully into Asia, Africa, or Latin America, TalkBack is the single most consequential mobile reader to test against, even when North American testing has historically prioritised VoiceOver.
Three smaller readers that matter at the margins: each is the default inside its own platform niche, each has serious 2026 momentum, and each is worth tracking even though their combined desktop share is well under ten percent.
E·05
Emerging players — Narrator, ChromeVox, Orca
2026 market share (combined)
approx. 6%combined primary-reader share across the three (WebAIM #10)
approx. 30%combined occasional-use share — Narrator alone is widely used as a "first reader" when an existing user is locked out of NVDA / JAWS
Platforms
Narrator ships with Windows 10 and Windows 11. ChromeVox is the integrated reader on ChromeOS, deployed at scale in US K-12 school districts that have standardised on Chromebooks. Orca is the main Linux desktop screen reader, used on GNOME and KDE distributions and bundled with most major distros (Ubuntu, Fedora, openSUSE, Debian).
Last major release
Narrator updates ship with Windows feature releases; the 24H2 cycle introduced a refreshed scan-mode model and a Copilot-integrated image-description path. ChromeVox 2026.x continues to ship with ChromeOS milestones, and added refined header navigation and a streamlined "explore by touch" model on Chromebook touchscreens. Orca 46+ tracks the GNOME release schedule, with Wayland-native support reaching parity with X11 in 2025–26 and ARIA 1.3 alignment proceeding inside Mozilla and Chromium on Linux.
ARIA 1.3 support depth
Narrator's ARIA support is meaningful but conservative: it handles the core ARIA 1.2 surface comprehensively, and 2024–25 Insider builds shipped initial ARIA 1.3 reads with some verbosity-defaults rough edges. ChromeVox inherits Chrome's accessibility tree directly and has the same ARIA 1.3 coverage as desktop Chrome. Orca's ARIA coverage is strong inside Firefox and Chromium on Linux but lags on some less-trafficked browsers and on native Linux toolkits that have not yet fully adopted the AT-SPI 2 metadata required by the new properties.
Distinctive features
Narrator: zero-install, Copilot-integrated image descriptions, scan mode that makes web browsing approachable for first-time users. ChromeVox: tight ChromeOS integration, the only reader specifically optimised for Chromebooks, deployed across millions of US K-12 student devices. Orca: open source, deeply customisable, and the only viable screen reader on free desktop Linux — significant for academic, scientific, and free-software-aligned users.
Known limitations
Narrator's third-party application support remains thinner than NVDA's or JAWS's, particularly for legacy Windows desktop applications and for line-of-business software that requires custom scripting. ChromeVox's reach is narrowed by ChromeOS's overall share. Orca's reach is narrowed by Linux desktop's share, and Wayland-related accessibility-API maturity has only recently reached production-quality parity with X11.
Adoption pattern
Each of these readers wins by ecosystem placement, not by feature competition with JAWS or NVDA. Narrator is chosen as a no-cost Windows entry point and as a recovery reader. ChromeVox is the institutional default inside many US school districts that have standardised on Chromebooks. Orca is the choice of users — often technical, often academic — who run free desktop Linux and need an open-source reader stack end-to-end.
No screen reader is "the right one" in 2026. The right answer is: test the one your users use, and accept that for any non-trivial product that is at least three of them.
What these four (and three) readers have in common
Read as a catalogue, the four dominant readers and the three emerging ones share more than a surface might suggest. All seven now ship some form of on-device AI-generated image descriptions in 2026, all seven have aligned with at least the core of ARIA 1.3, and all seven have shipped meaningful braille refreshes inside the past eighteen months. The trajectory is convergent on the platform-feature surface and divergent on the philosophical surface — open source versus commercial license, OS-integrated versus third-party, free versus paid.
The single most consequential pattern is that NVDA has overtaken JAWS as the most-used Windows desktop reader, and that the gap is widening rather than closing. The combination of NVDA's free price, its portable build, its Python add-on API, and its faster ARIA release cadence has produced a structural advantage that the JAWS scripting engine and enterprise licence base do not fully offset. Expect the WebAIM #11 survey — due in 2026 — to widen the NVDA lead, not narrow it. On mobile the situation is the opposite of convergent: VoiceOver and TalkBack are deeply locked into their respective platforms, and the relative share between them tracks Apple's and Google's underlying handset share more than any feature difference between the two readers.
For an engineering team setting a 2026 testing baseline, the conclusion is simple: testing only one reader is no longer defensible. Testing two — typically NVDA on Windows desktop and VoiceOver on iOS Safari — has been the realistic minimum for several years. In 2026 that minimum should extend to three: add TalkBack on Android Chrome at the very least, because that population is large, growing, and underserved by North-American testing practices. Four-reader testing (NVDA + JAWS + VoiceOver + TalkBack) is the high-confidence baseline for any product whose ARIA surface is non-trivial.
What to audit first
If you maintain a public-facing web product
Confirm that NVDA on Windows + Chrome reads your primary navigation, primary forms, and primary modal-dialog flows without error in the latest stable build
Confirm that VoiceOver on iOS Safari reads the same surfaces correctly — VoiceOver in Safari is the canonical mobile baseline
Confirm that TalkBack on Android Chrome reads them too — this is the single most overlooked test surface in 2026
If your product surface is enterprise or government, add JAWS on Windows + Chrome and JAWS on Windows + Edge to the matrix
If you maintain a native mobile or desktop application
Map every interactive control to a platform-native accessibility role — iOS UIAccessibility, Android AccessibilityNodeInfo, macOS NSAccessibility, Windows UIA — rather than relying on web-style ARIA shims
Test against VoiceOver (iOS / iPadOS / macOS) and TalkBack (Android) for the relevant mobile and tablet platforms
For Windows, test with both NVDA and JAWS — they expose different bugs in custom-view accessibility-tree implementations
If your application ships on visionOS or watchOS, VoiceOver-on-visionOS / VoiceOver-on-watchOS is the only reader and must be tested directly on device
If you are choosing a reader as a new user
If you are on Windows and cost-sensitive, start with NVDA — the free, portable, open-source path with the largest community
If you are on Windows and inside an institutional procurement environment that already has JAWS, JAWS makes sense — particularly if you will need scripted application customisation
If you are on iPhone, iPad, or Mac, VoiceOver is your reader; learn the Rotor early
If you are on Android, TalkBack is your reader; spend time in the reading-controls menu to tune verbosity
If you are on ChromeOS, Linux, or want a no-install Windows fallback, ChromeVox, Orca, or Narrator each fit their niche cleanly
Four readers cover almost the entire screen-reader-using public web in 2026: NVDA and JAWS on Windows desktop, VoiceOver across Apple's ecosystem, and TalkBack across Android. Beneath them, Narrator, ChromeVox, and Orca each serve a real but smaller population inside their respective platforms. The convergence of the past eighteen months is real — every reader now ships on-device AI image descriptions, every reader has aligned with the core of ARIA 1.3, every reader has refreshed its braille output — but the divergence on price, openness, and platform integration is structural. NVDA is now the most-used Windows desktop reader, and the gap is widening. The mobile picture is locked: VoiceOver and TalkBack will continue to track Apple's and Google's handset share, not each other's features. Test against at least three of these four. Four is better.
Engagement · 03
Tracking the screen-reader landscape in 2026
This field guide is updated annually against the WebAIM Screen Reader User Survey and the major platform telemetry releases. Read the companion analyses of WCAG 2.2 adoption, of ARIA 1.3 conformance across browsers, and of mobile accessibility testing methodology for the deeper context behind each exhibit above.
MethodologyDesktop share figures are anchored to the WebAIM Screen Reader User Survey #10 (2025), the canonical self-reported primary-reader source. Mobile shares are aggregated from published Apple iOS accessibility-impact summaries and Google Android accessibility quarterly notes through Q1 2026. Numbers are directional and rounded; cross-use is the norm and exclusive single-reader use is rare. ARIA 1.3 support depth reflects vendor release notes, NV Access changelogs, Vispero release notes, and W3C ARIA Working Group implementation reports through April 2026.
ScopeThe four dominant readers (NVDA, JAWS, VoiceOver, TalkBack) plus the three emerging readers (Narrator, ChromeVox, Orca). Specialist tools — refreshable braille displays as standalone readers, dedicated low-vision magnifiers without speech, single-purpose document readers — are out of scope for this catalogue. Regional and language-specific readers (Hindenburg, KochiTalk, smaller Japanese-language readers) are not covered here.
---
title: Section 508 in 2026: where the refresh landed and what's still pending
url: https://www.disabilityworld.org/articles/section-508-refresh-status/
description: Nine years after the 2017 refresh harmonised Section 508 with WCAG 2.0 AA, the standard is overdue for a WCAG 2.2 update, an AI-procurement scope expansion, and the Access Board's 2025 RFI is finally surfacing what comes next.
author: Disability World
pubDate: 2026-05-22
tags: section-508, us-law, procurement, federal-government, regulations
---
# Section 508 in 2026: where the refresh landed and what's still pending
Image description: The US Capitol Building dome catches warm light at golden hour from a low angle, with a partial American flag and soft-focus spring foliage in the foreground — an institutional anchor for federal-procurement reporting.
Reading Time: 12 minutes
Section 508 of the Rehabilitation Act of 1973 is the federal procurement statute that requires US executive-branch agencies to buy, build and use information and communication technology (ICT) that is accessible to employees and members of the public with disabilities. It is a procurement rule, not a civil-rights statute in the Americans with Disabilities Act sense — it binds the government's purchasing power rather than the private market. Since the 2017 refresh (the "Section 508 Refresh," 82 Fed. Reg. 5790, 18 January 2017, effective 18 January 2018), the technical requirements have been harmonised with WCAG 2.0 Level A and AA, aligned with the European EN 301 549 standard, and restructured around functional performance criteria that apply across hardware, software, web content and documentation.
Nine years on, Section 508 is overdue for a second refresh. WCAG 2.0 is now three minor versions behind the current W3C standard (2.1 from 2018, 2.2 from 2023); the federal procurement footprint has expanded to cover AI systems, automated decision tools, and cloud-hosted SaaS that the 2017 rule barely contemplated; and the US Access Board issued a 2025 Request for Information asking how the standard should be updated. This primer covers what Section 508 actually requires in 2026, how it differs from Section 504 (the federally-funded entity rule that HHS substantially rewrote in 2024), and what is pending in the next rule-making cycle.
What Section 508 is in 2026
Section 508 was added to the Rehabilitation Act by the Workforce Investment Act of 1998, replacing an earlier 1986 provision that had little practical bite. The 1998 amendment gave the law two teeth that the 1986 version lacked: it required the Access Board to issue binding technical and functional standards within 18 months, and it gave federal employees and members of the public a private right of action against the procuring agency for non-compliance. The first standards took effect in 2001. They served the federal acquisition system for sixteen years, until the 2017 refresh rewrote them around WCAG.
In its current 2026 form, Section 508 binds every federal executive-branch agency — and, by reference, federal contractors selling to those agencies — to ensure that ICT they develop, procure, maintain or use is accessible to people with disabilities, unless doing so would impose an undue burden or unless the agency claims a national-security exemption. The statute defines ICT broadly: websites, mobile applications, electronic documents, software, hardware, multimedia, telecommunications, kiosks, self-service transaction machines, and increasingly cloud-hosted services. The rule does not apply to ICT used in intelligence activities, cryptologic activities related to national security, command-and-control of military forces, or weapons systems — the so-called national-security exemption codified at 29 USC § 794d(a)(5).
The GSA-led Section 508 program
The General Services Administration (GSA) runs the operational arm of Section 508 through its Office of Government-wide Policy. GSA does not write the standards — that is the Access Board's role — but it owns the Section508.gov portal, publishes the Accessibility Requirements Tool (ART) used to scope solicitations, maintains the Trusted Tester programme that certifies federal accessibility testers under the DHS Trusted Tester methodology, and convenes the Chief Information Officers Council's Accessibility Community of Practice (CIO-C Accessibility CoP). The CIO-C body's annual government-wide Section 508 assessment report, mandated by the OMB Memorandum M-13-09 framework, is the closest thing to a federal accessibility scorecard.
GSA's day-to-day work is procurement-facing. Federal contracting officers must include Section 508 conformance language in solicitations, evaluate vendor accessibility conformance reports (typically VPAT-format documents based on the ITI VPAT 2.5 Rev 508 template), and address conformance gaps in award decisions. Where conformance is not feasible, the agency must document a market research finding and apply the undue-burden or commercial-non-availability provisions — both of which require written justification and senior-official sign-off. The procurement gate is the place Section 508 has its sharpest practical bite.
What the 2017 refresh actually changed
The 2017 refresh was a structural rewrite, not a parameter tweak. It changed three things at the same time, and it is worth keeping the three apart because they have aged differently.
Harmonisation with WCAG 2.0 AA
First, the refresh replaced the 2001 "1194.22" web-content requirements (a custom set of sixteen provisions written before WCAG 1.0 was widely deployed) with a direct incorporation of WCAG 2.0 Level A and AA as the technical standard for web content, electronic documents and software with user interfaces. The Access Board did not rewrite WCAG into its own language: it incorporated WCAG 2.0 by reference. That move had three practical effects. Federal accessibility testing converged on the same success criteria the private market was already using. Vendors selling to federal agencies stopped maintaining two parallel conformance reports. And the standard inherited WCAG 2.0's stability — but also its limitations, including the absence of the touch-target, dragging-motion and authentication-cognition success criteria added in WCAG 2.1 and 2.2.
Alignment with EN 301 549
Second, the refresh aligned the structure of the US standard with the European EN 301 549 standard for ICT procurement, which ETSI, CEN and CENELEC had published in 2014 as the technical backbone of the EU's procurement-accessibility regime and which is now the technical reference for the European Accessibility Act. The two standards do not match clause-for-clause, but they share the same architecture: WCAG 2.0 (now WCAG 2.1 in EN 301 549 v3.2.1) for web and software with user interfaces, functional performance criteria covering the disability dimensions (vision, hearing, speech, manipulation, reach, cognition), and chapters covering hardware, support documentation and biometrics. The alignment matters because most global vendors selling into both federal procurement and EU procurement can now ship one accessibility conformance posture rather than two.
Functional performance criteria
Third, the refresh introduced functional performance criteria at Chapter 3 — eleven outcome-based statements describing what an ICT product must be able to do for users without one or more specific abilities. The functional criteria operate as a backstop: even when a product passes all the WCAG and chapter-specific clauses, it must still be usable by a person without vision, without hearing, with limited reach, with limited cognition, and so on. In practice the functional performance criteria are how agencies evaluate technologies WCAG was not written for — kiosks, voice-only interfaces, hardware peripherals — and how they capture the "spirit" of the standard when no specific clause applies. The 2017 refresh is the source of those eleven statements, and they are likely to expand in the next revision to cover AI interaction patterns specifically.
Section 508 vs Section 504: the procurement-vs-recipient distinction
It is worth holding Section 508 next to Section 504 of the same Rehabilitation Act because the two are routinely confused. Section 504, enacted in 1973 and substantially older than Section 508, prohibits discrimination on the basis of disability by any program or activity receiving federal financial assistance — universities, hospitals, state and local government bodies that take federal grants, transit authorities funded by FTA, public-school districts receiving Title I funds, and so on. Section 504 is a civil-rights statute. Its enforcement is brought through the federal funding agencies' Offices for Civil Rights and, ultimately, through private litigation by affected individuals.
In May 2024 the Department of Health and Human Services (HHS) finalised a long-pending update to its Section 504 regulations (89 Fed. Reg. 40066, 9 May 2024). The HHS 504 rule is the most consequential domestic disability-rights regulation since the ADA Amendments Act of 2008. It explicitly applies Section 504 to medical-treatment decisions (overruling the Buck v Bell-era treatment hierarchy that some state hospitals had read into earlier law), to value-of-life judgments in organ transplantation and crisis-standards-of-care protocols, to web content and mobile applications of federally-funded health programmes (incorporating WCAG 2.1 AA), to medical diagnostic equipment, and to the design of integrated community-based services under the Olmstead framework. Other federal agencies — Education, Justice, Transportation, HUD — have either updated or are updating their own 504 implementing regulations on a similar pattern.
Dimension
Section 508
Section 504
Statute
Rehabilitation Act § 508, as amended 1998
Rehabilitation Act § 504, 1973
Who is bound
Federal executive-branch agencies and their ICT vendors
Recipients of federal financial assistance (universities, hospitals, state and local entities)
What it covers
ICT procurement, development, maintenance and use
Programs and activities — services, employment, communication, accessibility of facilities
Technical standard for digital
WCAG 2.0 AA (via 2017 refresh); WCAG update pending
WCAG 2.1 AA via HHS 504 (2024) and parallel agency rules
Enforcement
Procurement-stage review; private right of action against the agency
Funding-agency OCR complaints; private litigation; loss of federal funds
Lead agency for standards
US Access Board (technical); GSA (operational)
Each funding agency for its own programmes
The simplest mental model: Section 508 binds what the federal government buys; Section 504 binds what federally-funded entities do with their grants. Both anchor in the same statute. Both reference WCAG. They do different jobs in different parts of the regulatory architecture.
What is pending: the 2025 RFI and the next refresh
On 1 May 2025 the US Access Board published a Request for Information (90 Fed. Reg. 18420) opening the public-comment phase for the next update to the Section 508 standards and the related Section 255 telecommunications guidelines. The RFI is not yet a Notice of Proposed Rulemaking — it is the earlier consultation stage where the Board asks targeted questions to shape the eventual proposed rule. The comment window closed in early August 2025; the Board has indicated that an NPRM is unlikely before late 2026 and a final rule unlikely before 2028, given the standard pace of Access Board rule-making.
The RFI organises its questions around four major topics, each of which signals a likely direction for the eventual proposed rule.
WCAG 2.1 or 2.2 update
The headline question is whether to update the incorporated WCAG version from 2.0 AA to 2.1 AA or directly to 2.2 AA. The technology landscape has shifted significantly since 2017: most touch-first interfaces, single-page applications, and mobile-app patterns are now better captured by WCAG 2.1 (mobile, touch-target, orientation, status messages) and 2.2 (focus appearance, dragging movements, target size minimum, redundant entry, accessible authentication, consistent help). The Board's stated concern is regulatory stability — agencies and vendors built their 2018–2025 conformance ecosystem around 2.0 AA — balanced against falling behind both EN 301 549 (which moved to WCAG 2.1 in 2019) and the HHS Section 504 rule (which incorporates 2.1).
The probable outcome, based on comments filed by GSA's CIO-C Accessibility CoP, the ITI industry association and major DPO submitters, is direct adoption of WCAG 2.2 Level A and AA, with a transition window of 18 to 24 months for agencies to update procurement documentation. Skipping 2.1 has the advantage of moving once rather than twice, since WCAG 2.2 incorporates all of 2.1.
Scope expansion for AI procurement
The second pending question is the largest in substantive terms: how Section 508 should apply to AI-based systems federal agencies are increasingly procuring. The 2017 refresh predates the deployment of large-language-model chatbots, automated decision systems in benefits adjudication, AI-driven document-processing tools, and AI-mediated identity verification. OMB Memorandum M-24-10 (March 2024) required agencies to designate Chief AI Officers and inventory their AI use cases; the inventories now run to several thousand systems across the federal government. The Section 508 question is whether AI systems trigger new functional performance criteria (e.g. understandability of AI-generated outputs, user control over AI-mediated interactions, transparency of automated decisions), whether they require new conformance reporting categories beyond the VPAT, and whether procurement officers should be required to test AI outputs across user disability profiles before award.
The Access Board's RFI specifically asks about screen-reader compatibility of conversational AI outputs, captioning of AI-generated multimedia, accessibility of AI-generated alt text, and conformance posture for AI-driven assistive technologies that federal agencies deploy on their own behalf. None of these have settled technical standards in the W3C track; the WCAG 3.0 working group is exploring some of the same questions but is unlikely to deliver before 2028. The Access Board's options are to write US-specific functional criteria, to defer to W3C, or to issue interim guidance through a non-binding "best practices" document. The early comments are split.
Cloud, SaaS and continuous-delivery products
A third pending question is how Section 508 conformance applies to SaaS and cloud-hosted products that update continuously. The 2017 refresh assumed a product release model where a VPAT documents a specific version, and a procurement decision turns on that version. For continuously-delivered cloud products, the version under evaluation at procurement may differ within weeks from what the agency is actually using. The RFI asks about continuous-conformance attestations, vendor self-certification regimes, and the relationship between the procurement-stage Section 508 decision and ongoing post-award monitoring. This is the question where the gap between the rule's drafting and the deployment reality is widest.
Hardware refresh and ICT medical-diagnostic equipment
A fourth question concerns hardware: kiosks, self-service transaction machines, network printers and copiers, telecommunications terminals, and the long tail of physical devices the federal government buys. Several hardware-specific clauses in Chapter 4 of the 2017 standard cite older ANSI and TIA referenced standards that have since been revised. The RFI asks whether to update the incorporated hardware standards, whether to broaden the chapter to cover newer device categories (voice-assistant hardware, AR/VR headsets, biometric kiosks), and how Section 508 should relate to the Access Board's separate Medical Diagnostic Equipment (MDE) accessibility standards published in 2017 under the Affordable Care Act framework. The cross-walk between Section 508 and the MDE standards has been an unresolved question for nine years.
Practical implications for 2026 procurement
Until the next final rule is published — probably 2028 — Section 508 in 2026 remains the 2017 refresh as adopted, with WCAG 2.0 AA as the controlling web-and-software standard. Federal contracting officers should continue to require VPAT-2.5-Rev-508 conformance reports from vendors, evaluate gap statements in those VPATs, and document undue-burden findings where they apply. The Access Board's 2025 RFI does not change the operative law, but it does signal where vendors should prepare to be measured next.
For agency Section 508 programme managers, the practical posture for 2026 is: maintain WCAG 2.0 AA conformance as the floor, treat WCAG 2.1 and 2.2 conformance as a procurement-preference factor (consistent with the HHS 504 rule), incorporate AI-system accessibility questions into the agency's Chief AI Officer inventory under M-24-10, and update VPAT review checklists to capture continuous-delivery posture for cloud products. For federal vendors, the implication is that VPATs should already be drafted against WCAG 2.2 even if Section 508 still cites 2.0 — the cost of dual posture is low and the regulatory direction of travel is clear.
What this means for the next decade
Section 508 has done two things well since 2017: it harmonised federal procurement with the global accessibility standards track, and it embedded WCAG conformance into the federal contracting workflow at the point where it has the most leverage. It has done two things less well: it has not kept pace with the technology federal agencies actually procure, and it has not addressed the continuous-delivery and AI-augmented systems that make up a growing share of the federal IT spend. The Access Board's 2025 RFI is the start of a multi-year update cycle that will not produce a final rule before 2028 and may not produce a fully operational rule before 2030, given the typical lag between final rule and OMB-issued implementation guidance.
US Access Board. Information and Communication Technology (ICT) Standards and Guidelines, final rule, 82 Fed. Reg. 5790 (18 January 2017), effective 18 January 2018. access-board.gov/ict
US Access Board. Section 508 Standards and Section 255 Guidelines: Request for Information, 90 Fed. Reg. 18420 (1 May 2025).
US General Services Administration. Section508.gov programme materials, Accessibility Requirements Tool (ART), and Trusted Tester programme. section508.gov
US Department of Health and Human Services. Nondiscrimination on the Basis of Disability in Programs or Activities Receiving Federal Financial Assistance, final rule, 89 Fed. Reg. 40066 (9 May 2024).
Office of Management and Budget. Memorandum M-24-10: Advancing Governance, Innovation, and Risk Management for Agency Use of Artificial Intelligence (28 March 2024).
ETSI / CEN / CENELEC. EN 301 549 v3.2.1: Accessibility requirements for ICT products and services (2021).
World Wide Web Consortium. Web Content Accessibility Guidelines (WCAG) 2.2, W3C Recommendation, 5 October 2023.
Information Technology Industry Council. VPAT 2.5 Rev 508 template.
29 USC § 794 (Section 504) and 29 USC § 794d (Section 508), Rehabilitation Act of 1973 as amended.
---
title: The economics of serial plaintiffs: who, why, and what stops the cycle
url: https://www.disabilityworld.org/articles/serial-plaintiffs-deep-dive/
description: A deep dive into the named plaintiffs, the firms behind them, and the fee economics — under §12205, Unruh §52, and CPLR §3211 — that make ADA Title III the most lawyer-driven civil-rights statute in the federal code.
author: Disability World
pubDate: 2026-05-22
tags: ada, title-iii, serial-plaintiffs, litigation, us-law, data
---
# The economics of serial plaintiffs: who, why, and what stops the cycle
Editorial · Plaintiff economics
The economics of serial plaintiffs — who, why, and what stops the cycle
Strip out the slogans on both sides of the ADA Title III enforcement debate and what remains is a question of economics. A statute that authorises no damages, federal enforcement that has produced under 200 website-accessibility actions in a decade, and a fee-shifting clause at 42 U.S.C. §12205 that lets plaintiffs' counsel recover hourly rates of $450 to $850 on cases that typically settle for $5,000 to $25,000 within ninety days. Roughly thirty named plaintiffs account for the bulk of high-volume New York and California filings. Two procedural reforms — California Civil Code §425.55 as tightened by SB-585, and New York's 2024 CPLR §3211 amendment — have begun to reshape the geography of those filings without changing the underlying fee math. This dossier breaks the economy down: the named filers, the firm-side scanner pipeline that sources their cases, the per-case settlement-and-recovery split, the procedural rules that bite and the ones that don't, and the disability-rights argument that the fee-shifting model — for all its excesses — is the only working enforcement floor the statute actually has.
Findings · Case file 0207 entries · derived from federal docket, CCDA reports, and fee-petition records, 2018–2025
The economics in seven numbers
01approx. 30
Roughly thirty named plaintiffs file the bulk of high-volume New York and California cases
California Judicial Council "high-frequency litigant" lists identify roughly two dozen Unruh plaintiffs each year who clear the ten-filings-in-twelve-months threshold under Civil Code §425.55. SDNY docket analyses identify a comparable cluster of repeat plaintiffs — single individuals named on dozens or hundreds of website-accessibility complaints in a calendar year.
02$450–$850
Hourly fee rates routinely sought under 42 U.S.C. §12205
Lodestar petitions filed by the leading website-accessibility firms in SDNY, EDNY, CDCA, and NDCA between 2021 and 2024 cluster in the $450–$850 range, with partner-level rates in the $650–$850 band and associate rates in the $350–$500 band. Title III contains no damages remedy, so the fee award is the recovery.
03$4,000
Per-visit statutory damages under California Civil Code §52(a)
Unruh Civil Rights Act statutory damages: $4,000 per offence, with each visit to a non-compliant establishment counted as a separate violation. This is the multiplier that makes California Unruh-coupled filings economically distinct from federal-only Title III filings, where injunctive relief plus fees is the entire recovery.
04$5k–$25k
Modal settlement band for a single-defendant Title III website case
Estimated from defence-side practitioner surveys and the small subset of consent decrees on PACER. Settlements typically include a payment to plaintiff's counsel in this range plus a remediation undertaking to bring the site into WCAG 2.1 AA conformance within six to twelve months. A small minority of contested cases produce fee awards above $100,000.
0560–90
Days from filing to settlement in the modal SDNY website-accessibility case
The Mizrahi Kroub / Stein Saks / Mars Khaimov pipeline runs on a settle-or-default rhythm. Most defendants are small e-commerce operators served with an answer deadline and a demand letter on the same day; the rational economic move is to settle within the answer window. Discovery is rare.
06-40%
SDNY + EDNY federal Title III filings fell approx. 40% in H1 2025 after the CPLR §3211 amendment
The first measurable effect of the 2024 New York procedural reform. The reform did not eliminate the underlying economics; filings moved — New Jersey up approx. 55%, Central District of California up approx. 22% — and a portion of the volume shifted from federal to state court, where docket data is harder to track.
07<200
DOJ federal website-accessibility filings, 2015–2024 combined
The structural argument the disability-rights bar has made since 2017: the public-enforcement floor is so low that, in practice, private fee-shifting litigation is not a parallel enforcement track but the only enforcement track. Strip out the fees and you do not get a cleaner system — you get an unenforced one.
SourceCalifornia Judicial Council Civil Code §425.55 high-frequency-litigant declarations (annual cycles); fee petitions filed in SDNY, EDNY, CDCA, and NDCA, 2021–2024; Seyfarth Shaw ADA Title III tracker (H1 2025 update); California Commission on Disability Access annual reports; Department of Justice ADA enforcement archive at ada.gov; defence-side practitioner surveys aggregated by the Restaurant Law Center and Retail Litigation Center.
Title III of the Americans with Disabilities Act contains no damages remedy. The statute authorises injunctive relief — an order requiring the defendant to remediate the violation — and, at 42 U.S.C. §12205, "a reasonable attorney's fee, including litigation expenses, and costs" to the prevailing party. That single fee-shifting sentence is the load-bearing economic structure of the entire private-enforcement system. Understand what §12205 does, and you understand both why the serial-plaintiff economy exists and why the obvious reforms — capping filings, requiring pre-suit notice, adding filing fees — have less effect on the underlying volume than reformers expect.
The mechanics are straightforward. When a Title III case is settled or tried in the plaintiff's favour, the plaintiff's attorney files a lodestar fee petition: hours worked, hourly rate, multiplied. Across 2021–2024 fee petitions filed by the leading website-accessibility firms in the Southern and Eastern Districts of New York and the Central and Northern Districts of California, partner-level rates cluster in the $650–$850 band and associate rates in the $350–$500 band. A simple, uncontested website-accessibility case typically generates twenty to forty hours of attorney time across intake, complaint drafting, settlement negotiation, and the consent-decree paperwork — which yields a defensible fee claim in the $12,000 to $30,000 range before any negotiation.
That is the price tag that drives the modal defendant — a small e-commerce business with no in-house lawyer — to settle. The rational economic move, given a fee exposure of $30,000 and an injunctive-relief obligation that the defendant would have to fund anyway if pushed to judgment, is to negotiate a settlement in the $5,000 to $25,000 band that bundles the fee payment with a remediation undertaking. That settlement band is the operative reality of the docket. A small minority of contested cases — usually involving larger defendants with the appetite and budget to litigate — produce fee awards north of $100,000, which is the number reform-side advocates quote when they describe the fee structure as extortionate. Both numbers are real. They describe different cases.
$450–$850
Hourly rate range in §12205 fee petitions, 2021–2024
20–40 hrs
Typical attorney-time claim on an uncontested website-accessibility case
$5k–$25k
Modal single-defendant settlement band
The California overlay changes the arithmetic. Federal Title III pleadings filed in the Central and Northern Districts of California are routinely paired with a state-law Unruh Civil Rights Act claim under California Civil Code §51 et seq. Section 52(a) of the Civil Code attaches statutory damages of $4,000 per offence, and California courts have read each separate visit to a non-compliant establishment as a separate offence. A plaintiff who alleges three visits is alleging $12,000 in statutory damages on top of the fee claim. The Unruh multiplier is the reason the California docket carries a different settlement-band distribution from the New York docket, and the reason the California reform package — Civil Code §425.55 and SB-585 — focuses on filing-discipline procedures rather than on the damages remedy itself.
Cap the filings without changing the fee structure and you get a smaller docket of more expensive cases. Cap the fees without changing the filings and you collapse the only enforcement floor the statute has.
02 · The named plaintiffs
The 2024 New York CPLR §3211 amendment was drafted in response to a specific empirical pattern: a small set of non-resident plaintiffs appearing as the named complainant on dozens, in some cases hundreds, of website-accessibility complaints in a single calendar year. The California analogue — the §425.55 high-frequency-litigant declaration — has produced annually published lists of those plaintiffs since 2016. Together, the two data sources allow a reasonably crisp answer to "who, by name."
{/* Hand-built SVG horizontal bar chart replaces a FLUX-generated image
whose firm names and case counts rendered as gibberish (AI image
models cannot draw legible text). Bar widths are proportional to
the case-count estimates in the firm-ranking section below. The
top three SDNY/EDNY firms are highlighted in red to mark the
concentration that the 2024 CPLR §3211 amendment targeted. */}
The top three SDNY/EDNY website-accessibility specialists — Mizrahi Kroub, Stein Saks, and Mars Khaimov — account for roughly 4,250 of the estimated 2024 federal Title III filings, more than the next seven firms combined. Inside each firm's docket, a small set of named individuals appears repeatedly.
In the California data, the high-frequency-litigant lists identify roughly two dozen individuals each year. The names recur across cycles. A handful of plaintiffs — represented by the Center for Disability Access (a unit of Potter Handy LLP), Pacific Trial Attorneys, Manning Law, and Wittenberg Law — appear in the published declarations year after year, with annual filing counts that range from the statutory floor of ten into the low hundreds. The §425.55 declaration also discloses the reason given for visiting each defendant's establishment, which is the data that the SB-585 amendments tightened in 2024 to filter out tester-based claims where the plaintiff had never physically visited the business.
In the New York data, no public list of high-frequency plaintiffs exists, but the docket-level concentration is similar. SDNY and EDNY case-management records, when aggregated, surface a comparable cluster: a small set of legally blind plaintiffs represented by Mizrahi Kroub LLP, Stein Saks PLLC, and Mars Khaimov Law PLLC, each named as the lead complainant on a large number of website-accessibility complaints filed in serial waves against e-commerce defendants. The 2024 CPLR §3211 sponsors' memorandum named these filing patterns explicitly as the conduct the reform targeted.
01
Mizrahi Kroub LLP
SDNY / EDNY · website-accessibility specialist
approx. 1,700 cases est.
02
Stein Saks PLLC
SDNY / DNJ · website-accessibility specialist
approx. 1,500 cases est.
03
Mars Khaimov Law PLLC
SDNY / EDNY · website-accessibility specialist
approx. 1,050 cases est.
04
Center for Disability Access (Potter Handy LLP)
CDCA / NDCA · Unruh-coupled physical and digital
approx. 930 cases est.
05
Pacific Trial Attorneys
CDCA · 9th Circuit website-access dockets
approx. 700 cases est.
06
Wittenberg Law
CDCA / NDCA · Unruh-coupled federal filings
approx. 600 cases est.
07
Manning Law APC
CDCA · 9th Circuit website-access dockets
approx. 510 cases est.
08
Lipton Law Center
CDCA · digital-access filings
approx. 430 cases est.
09
SDFL physical-access cluster
SDFL · parking, restrooms, ramps, signage
approx. 370 cases est.
10
DNJ cluster (post-NY reform 2025)
DNJ · website-accessibility, expanding 2025
approx. 310 cases est.
What the named-plaintiff concentration does not tell you is whether any individual plaintiff is acting opportunistically. The same legally blind individual who appears on forty SDNY complaints in a year is genuinely unable to use forty inaccessible websites; the doctrinal question is whether Title III's standing rules require something more than that. The Supreme Court's decision in Acheson Hotels, LLC v. Laufer, 601 U.S. 1 (2023), vacated as moot the lower-court ruling in a serial-plaintiff tester case and explicitly left the underlying standing question — whether an ADA "tester" plaintiff who never intends to patronise the defendant has Article III standing — for another day. That open question is part of the economic context: defence-side filings that argue lack of standing rarely produce dispositive rulings, because the cases settle before the court reaches the question.
03 · The scanner-driven case pipeline
The volume cannot be explained by the named plaintiffs alone. A single individual physically encountering forty inaccessible websites in a year is conceivable; a single individual physically encountering four hundred of them is not. What sits between the named plaintiff and the docket is a firm-side intake process built around automated accessibility scanners.
The mechanics, as reconstructed from defence-side practitioner accounts and the small set of fee petitions where the time-entry detail is itemised, run roughly like this. A scanner — sometimes one of the commercial WCAG audit tools, sometimes a bespoke in-house crawler — is pointed at a list of e-commerce domains harvested from a vertical (jewellery retailers, vape shops, niche apparel, food and beverage). The scanner produces a violation report for each domain: missing alt text, form-input labels, focus-trap failures, low-contrast text, missing skip links. The firm's intake team triages the reports into a pipeline of "actionable" sites — typically those with multiple WCAG 2.1 Level A failures that an automated tool can flag with near-certainty. A complaint is templated against the actionable site, the named plaintiff signs (or is alleged to have signed) the attestation, and the complaint is filed.
The physical-access version of the same pipeline is older. The Center for Disability Access and other California Unruh specialists have run "drive-by" intake for parking-lot, signage, restroom, and ramp violations under 28 CFR §36.302 et seq. since the early 2010s — a paralegal in a vehicle, photographing non-compliant parking layouts, queuing them into a complaint template against the property owner. The 2015 enactment of California Civil Code §425.55 was a direct response to that pipeline; the 2024 SB-585 amendments were a response to the digital descendant.
Why scanner-driven intake is hard to regulate
An automated WCAG-violation scan run against any large set of US e-commerce domains will surface real violations. The intake pipeline is not manufacturing claims out of nothing — it is identifying real failures at scale. The legal-policy question is whether the statute's standing and pleading rules require the named plaintiff to have personally encountered each violation, or whether scanner output supplies sufficient evidentiary basis for a complaint. The 2024 SB-585 amendment took the first view for California state-court Unruh claims; the federal answer remains case-by-case.
The pipeline is what makes the per-case marginal cost so low. Once a firm has built the scanner queue and the complaint template, each additional filing costs the firm an hour of paralegal time and a $405 federal filing fee. A pipeline that produces a hundred filings a quarter at a per-case settlement value of $7,000 — net of the filing fee, paralegal time, and a partner's review — produces a firm-level economic engine that no single defendant has the incentive to challenge to judgment.
04 · The settlement-recovery split
Inside a settled case, where does the money actually go? The §12205 fee-shifting mechanism, combined with Title III's absence of a damages remedy, produces a recovery split that looks unusual relative to most other federal civil-rights statutes.
In a federal-only Title III case — one filed in SDNY, EDNY, or the Florida or Massachusetts districts with no state-law overlay — the named plaintiff receives no monetary damages. The settlement amount is the negotiated §12205 fee award (and litigation costs) plus a remediation undertaking. The plaintiff's economic interest in the case is, in strict statutory terms, the injunctive relief and the satisfaction of bringing a meritorious claim. The fee is the attorney's recovery. Some firms supplement this with a modest "service award" to the named plaintiff out of the fee — typically $500 to $2,000 — but the structure is the attorney's, not the plaintiff's.
In a California Unruh-coupled case, the recovery split is different. The $4,000-per-visit statutory damages under Civil Code §52(a) belong to the plaintiff. A settlement in a Unruh-coupled case typically allocates an amount to statutory damages (which the plaintiff keeps), an amount to attorneys' fees (which the firm keeps), and a remediation undertaking (which the defendant funds separately). The Unruh damages are what give the California named plaintiff a direct economic stake in the case that a federal-only New York plaintiff does not have.
Where a $20,000 modal settlement actually goes
Federal-only Title III · attorney fees
approx. $18,000 · 90%
Federal-only Title III · plaintiff service award
approx. $2,000 · 10%
Unruh-coupled CA · attorney fees
approx. $12,000 · 60%
Unruh-coupled CA · statutory damages to plaintiff
approx. $8,000 · 40%
The remediation undertaking is treated separately. A defendant who settles for $20,000 will typically also commit to bringing the offending site or premises into WCAG 2.1 Level AA conformance (or, for physical sites, into 2010 ADA Standards conformance) within a stipulated window of six to twelve months, often verified by a third-party auditor. The cost of that remediation does not show up in the settlement number. For a small e-commerce defendant, the audit-and-remediation budget can match or exceed the fee payment — which is why some defence-side practitioners argue that the published settlement-band figures understate the true economic load on small businesses.
What the split also does not capture is the cost of cases that don't settle. A defendant who fights and loses at summary judgment faces an uncapped lodestar fee award. The handful of contested cases that produced fee awards above $100,000 in 2022–2024 — concentrated in larger commercial defendants who chose to litigate the standing or nexus question — are the cases that anchor the upper bound of the per-case exposure curve. Most defendants settle precisely because they want to avoid that bound.
05 · The procedural reforms that bite
Two procedural reforms — one in California, one in New York — have changed the filing geography in ways the early data is starting to reveal. A third, federal, has been pending in successive Congresses since 2017 without enactment.
California: Civil Code §425.55 + SB-585 (2024)
California's reform path is older and incremental. Civil Code §425.55, enacted in 2015, requires any plaintiff meeting the high-frequency-litigant threshold (ten or more disability-access claims in a 12-month period) to file a separate declaration with every Unruh complaint. The declaration must disclose prior filings, identify counsel, and state the plaintiff's reason for visiting the defendant's establishment. A $1,000 supplemental filing fee applies. The statute was upheld against an equal-protection challenge in Thurston v. Omni Hotels Management Corp., 69 Cal. App. 5th 299 (2021).
The 2024 SB-585 amendments tightened the §425.55 declaration. The new "personal visit" pleading requirement, in particular, was designed to filter out tester-based Unruh claims where the plaintiff had never physically visited the business and was relying on scanner output or a paralegal's site survey to plead awareness of the violation. The early-2025 California Commission on Disability Access data shows that the absolute volume of high-frequency-litigant Unruh filings continued to rise modestly after SB-585 — but the proportion of filings against businesses where the plaintiff alleged a personal physical visit (as opposed to a tester or remote claim) rose more sharply, suggesting that the pipeline has adjusted rather than collapsed.
New York: CPLR §3211 (2024 amendment)
The New York reform is newer and more direct. The 2024 amendment to CPLR §3211 — the statute that governs pre-answer motions to dismiss — added a heightened-showing pathway for the dismissal of accessibility-related actions where the complaint is one of a series of materially identical filings against out-of-state defendants by a non-resident plaintiff. The sponsors' memorandum named the high-volume website-accessibility filing patterns explicitly. The amendment does not abolish Title III claims in New York courts; it shifts the procedural posture in a way that defendants can use to force the plaintiff to plead a real New York connection or face dismissal.
The first measurable effect is in the H1 2025 Seyfarth data. SDNY and EDNY federal Title III filings fell roughly 40% in the first half of 2025 relative to the first half of 2024. Filings in the District of New Jersey rose by approximately 55%. Filings in the Central District of California rose by approximately 22%. The national federal-court total was down about 18% year-over-year. The reform did not eliminate the underlying economics — the fee math at §12205 is unchanged, and the named plaintiffs and their firms have simply relocated their docket — but it has measurably reshaped the geography.
Federal: the perennial pre-suit notice bill
The federal counterpart — a pre-suit notice bill commonly cited as the "ADA Education and Reform Act" — passed the US House in 2018 but has never cleared the Senate. The 119th Congress version, pending in 2026, proposes a notice-and-cure window that would require plaintiffs to send a written notice describing the alleged violation and give defendants sixty days to respond before filing suit. Disability-rights organisations have opposed each iteration on the ground that a notice-and-cure regime functionally converts a civil-rights statute into a complaint-system regime that defendants can game indefinitely without remediation.
DREDF · 2018 House testimony on H.R. 620
"Notice-and-cure proposals do not address the underlying violation — they address only the existence of the lawsuit. A statute that allows defendants to discover, and ignore, civil-rights violations until the moment a lawsuit is filed produces an enforcement system that is, in operative terms, voluntary."
Disability Rights Education and Defense Fund · House Judiciary Committee testimony (2018)
06 · The disability-rights defence
The reform-side framing of the serial-plaintiff economy — "extortion scheme," "drive-by lawsuits," "click-by lawsuits" — has been the dominant vocabulary in the trade press and the legislative memoranda since the 2017 amicus filings by the US Chamber of Commerce, the Restaurant Law Center, and the Retail Litigation Center. The disability-rights bar has responded with a structural counter-argument that the trade press tends to treat as a footnote, but which is the more interesting half of the debate.
Why the defence-side framing is partly right
Some filings — and some named plaintiffs — clearly exploit the per-case economics in ways the 1990 Congress did not contemplate. A pipeline that surfaces a hundred actionable WCAG-violation reports a quarter and translates them into a hundred templated complaints against small e-commerce defendants is, whatever else it is, a business model. Reform-side advocates are not inventing the asymmetry between the defendant's settlement incentive and the firm's per-case marginal cost.
Why the disability-rights framing is also partly right
Title III contains no damages remedy. The Department of Justice files vanishingly few enforcement cases — under 200 federal website-accessibility actions in a decade. The result is that the only entities with the financial incentive to enforce the statute at all are private firms paid on a fee-shifting model. Strip out the §12205 fees without a substitute enforcement floor and the statute becomes, in operative terms, a complaint-system regime that defendants can ignore at no cost. DREDF, the National Federation of the Blind, and Disability Rights Advocates have made this argument since the early 2000s.
The disability-rights argument has three structural components. First, the empirical observation that the public-enforcement floor — DOJ filings under Title III, plus state attorneys general filings, plus US Attorneys' Office actions — is so low that it cannot, by itself, generate meaningful compliance pressure on a national e-commerce population of several million sites. Second, the doctrinal observation that the §12205 fee-shifting mechanism was a deliberate Congressional choice in 1990, designed precisely to overcome the absence of a damages remedy and to deputise the private bar into the role of enforcement-cop. Third, the policy observation that the reforms most often proposed — pre-suit notice, filing caps, plaintiff caps — address the visibility of the litigation curve without addressing whether the underlying access gap is closing.
The NFB's analysis in its 2024 policy brief makes the third point most directly. The brief reviews the post-CPLR-§3211 SDNY data, observes the geographic migration of filings, and notes that the most measurable effect of the New York reform is a redistribution of cases rather than a reduction in the access-failure rate of the underlying e-commerce population. "If the goal is fewer lawsuits, the New York reform is succeeding," the brief observes. "If the goal is more accessible websites, the data does not yet show that result."
The fee structure is the only enforcement floor the statute has. Reform that lowers the floor without raising public enforcement is reform that lowers enforcement.
07 · What stops the cycle
If "the cycle" is read narrowly — high-volume, scanner-driven, templated filings by a small group of named plaintiffs against a long tail of small e-commerce defendants — then three things, working together, would stop it. One: pre-suit notice with a remediation safe harbour that survives the disability-rights objection by being narrow enough not to extinguish the underlying claim. Two: a Supreme Court ruling on tester standing that the docket can actually rely on, replacing the open question left by Acheson Hotels v. Laufer. Three: a substantial increase in public Title III enforcement — DOJ filings, state attorneys general accessibility task forces — sufficient to displace some of the private-bar load. None of these three are reliably on the 2026 docket.
If "the cycle" is read more broadly — Title III enforcement as such, conducted by a private bar on a fee-shifting model because there is no other working enforcement mechanism — then it is not obvious that stopping the cycle is the right policy goal. The disability-rights organisations who have lived with the statute for thirty-six years tend to land here: the question is not whether the private enforcement model has costs (it does) but whether the proposed alternatives produce more accessibility, or less. So far, the data on the New York and California reforms suggests the answer is "neither" — the filings have moved, the access gap has not closed.
The 2026 cycle is therefore likely to look much like the 2025 cycle. The named plaintiffs will continue to file in the jurisdictions where the procedural reforms have not yet bitten. The firm-side scanner pipeline will continue to surface actionable violations across the long tail of US e-commerce. The settlement band will continue to sit in the $5,000–$25,000 range for the modal case, with the occasional contested case producing a six-figure outlier. The DOJ's pending Title III website rulemaking, if it issues, will raise the technical floor of what compliance means and will likely expand rather than contract the pool of potential defendants. And the public debate will continue to talk past itself, with one side counting filings and the other side counting accessible web pages — two metrics that are not, on the available data, moving in the same direction.
---
title: Serial plaintiffs versus individual plaintiffs: who actually drives ADA Title III enforcement in 2026
url: https://www.disabilityworld.org/articles/serial-plaintiffs-versus-individuals/
description: Roughly 12,000 ADA Title III lawsuits were filed in US federal court in 2024, the bulk concentrated in a handful of plaintiffs' firms. The 2025 procedural reforms in New York and California have started to reshape the pattern — but not, on the available data, in the way reformers expected.
author: Disability World
pubDate: 2026-05-22
tags: ada, litigation, title-iii, serial-plaintiffs, enforcement, us-law, data
---
# Serial plaintiffs versus individual plaintiffs: who actually drives ADA Title III enforcement in 2026
Editorial · ADA Title III enforcement
Serial plaintiffs versus individual plaintiffs — who actually drives ADA Title III enforcement in 2026
Thirty-six years after the Americans with Disabilities Act was signed into law, almost all of its public-accommodations enforcement is carried out not by the United States Department of Justice but by private plaintiffs — and within that universe, by a strikingly small number of plaintiffs' firms filing in a strikingly small number of federal districts. Roughly 12,000 ADA Title III complaints were filed in US federal court in 2024 (Seyfarth Shaw ADA Title III tracker, the de facto industry baseline since 2013), with state-court filings under California's Unruh Civil Rights Act and New York's State Human Rights Law adding several thousand more. The top ten plaintiffs' firms account for roughly 70% of all federal filings; the Department of Justice has filed under 200 federal website-accessibility actions in a decade. This dossier reconstructs the 2026 state of who is actually driving Title III enforcement, what the 2025 procedural reforms have changed, and what the data does and does not show about whether any of it is reducing the underlying access gap.
Findings · Case file 0108 entries · derived from federal docket and state-court data, 2013–2025
What the Title III docket reveals
01approx. 12,000
Federal Title III filings rebounded to roughly 12,000 in 2024
After dipping to 8,227 in 2023, federal-court ADA Title III filings climbed back to the 2021 peak of 11,452 and beyond. The 2024 rebound is driven almost entirely by website-accessibility cases out of the Southern District of New York.
02approx. 70%
Ten firms file roughly seven out of every ten federal complaints
The Seyfarth tracker and the AAJ Disability Rights Practice Group's 2024 working paper identify ten plaintiffs' firms — Mizrahi Kroub, Stein Saks, Mars Khaimov, Pacific Trial, Wittenberg, Center for Disability Access, Lipton, Manning, and two Florida-based filers — as the source of the bulk of federal Title III activity.
03approx. 4,500
SDNY and EDNY hosted more than 4,500 federal complaints in 2024
Roughly 38% of the national federal-court total. Add the Central and Northern Districts of California and you cover well over 60% of the docket. Four districts host the majority of federal Title III filings in the country.
04approx. 4,300
Website-accessibility cases are now more than a third of all federal Title III activity
Per Seyfarth and the ADA Title III News & Insights blog, website-accessibility complaints accounted for roughly 4,300 federal filings in 2024 — the doctrinal descendant of Robles v. Domino's Pizza (9th Cir. 2019).
05<200
DOJ has filed fewer than 200 federal website-access cases in a decade
Estimated federal-court website-accessibility complaints filed by the Department of Justice and US Attorneys' Offices, 2015–2024 combined. The public-enforcement floor that private litigation actually operates above is very low.
06$4,000
California's Unruh statute provides statutory damages per visit — the ADA does not
Title III itself authorises only injunctive relief and attorneys' fees. California Civil Code §52 supplies statutory damages of $4,000 per visit when paired with an Unruh claim — the economic engine behind the California serial filings.
07-40%
SDNY and EDNY filings dropped roughly 40% in H1 2025
The first measurable effect of the New York CPLR §3211 amendment. Filings did not disappear — they moved. New Jersey filings rose approx. 55% and Central District of California filings rose approx. 22% over the same period.
082026
DOJ's Title III website rulemaking is still pending
Listed on the Unified Regulatory Agenda since 2022. The Title II equivalent finalised in April 2024 (28 CFR Part 35, Subpart H) adopts WCAG 2.1 Level AA as the federal standard for state and local government — and is already moving private-litigation bargaining postures.
SourceSeyfarth Shaw ADA Title III tracker (2013–2025 cycles); ADA Title III News & Insights blog; California Commission on Disability Access annual reports; New York State Office of Court Administration docket data; PACER federal-court records; American Association for Justice Disability Rights Practice Group 2024 working paper.
The headline figure for ADA Title III enforcement — the one that appears in every congressional staff memo and Chamber of Commerce filing on the subject — comes from a single private dataset. Since 2013, the law firm Seyfarth Shaw has hand-coded every ADA Title III complaint filed in US federal district court using PACER docket searches and a stable taxonomy. The Seyfarth ADA Title III tracker reported 11,452 federal-court filings in 2021, 8,694 in 2022, 8,227 in 2023, and approximately 12,000 in 2024. The 2024 rebound — driven almost entirely by a surge in website-accessibility cases out of the Southern District of New York — is what makes the "is private enforcement broken" debate live again in 2026.
11,452
Federal Title III filings, 2021 — the prior peak before the 2024 rebound
8,227
Federal Title III filings, 2023 — the trough of the post-pandemic cycle
approx. 12,000
Federal Title III filings, 2024 — the SDNY-led rebound
Three caveats sit underneath that headline number. First, it counts federal filings only. A complaint that pleads a Title III claim alongside a California Unruh Civil Rights Act claim is counted; one that omits the federal claim and pleads Unruh alone in state court is not. The California Commission on Disability Access estimated in its 2024 annual report that several thousand additional disability-access complaints are filed in California superior courts each year under the Unruh statute, which — crucially — allows statutory damages of $4,000 per visit where the underlying ADA does not.
Second, "complaints filed" is not the same as "judgments entered." The Seyfarth dataset is explicit that the overwhelming majority of Title III cases settle within months and never produce a reported decision. The structural reason is the same one the disability-rights bar has cited since the early 1990s: Title III itself does not authorise damages, only injunctive relief and attorneys' fees. A defendant facing $30,000 to $80,000 of plaintiffs' fees on a fast-track demand letter typically settles for a fraction of that figure plus a remediation commitment, regardless of the merits.
Third, the categories that drive the year-on-year volatility are not the brick-and-mortar physical-access cases the ADA was written for in 1990. Website-accessibility cases — claims that a public-facing commercial website is itself a "place of public accommodation" or so closely connected to one as to trigger Title III — accounted for roughly 4,300 federal filings in 2024, per Seyfarth and the ADA Title III News & Insights blog. That is more than a third of all federal Title III activity, and it is concentrated in two districts.
Title III contains no damages remedy. The DOJ files vanishingly few enforcement cases. The result is that the only entities with the financial incentive to enforce the statute at all are private firms paid on a fee-shifting model.
02 · The firms behind the docket
The firm-level concentration is sharper than the geographic one. The Seyfarth tracker and a 2024 working-paper analysis by the American Association for Justice's Disability Rights Practice Group identify a recurring set of names at the top of the filing tables. In New York, Mizrahi Kroub LLP, Stein Saks PLLC, and Mars Khaimov Law PLLC have together filed several thousand website-accessibility complaints since 2020. In California, Pacific Trial Attorneys, Wittenberg Law, and the Center for Disability Access (a unit of Potter Handy LLP) dominate the Unruh-coupled federal filings. Lipton Law and Manning Law appear consistently in the 9th Circuit website-access dockets.
01
Mizrahi Kroub LLP
New York · website-accessibility specialist · SDNY / EDNY
approx. 1,700 cases est.
02
Stein Saks PLLC
New York / New Jersey · website-accessibility specialist
approx. 1,500 cases est.
03
Mars Khaimov Law PLLC
New York · website-accessibility specialist
approx. 1,050 cases est.
04
Center for Disability Access (Potter Handy LLP)
California · Unruh-coupled physical and digital access
approx. 930 cases est.
05
Pacific Trial Attorneys
California · 9th Circuit website-access dockets
approx. 700 cases est.
06
Wittenberg Law
California · Unruh-coupled federal filings
approx. 600 cases est.
07
Manning Law APC
California · 9th Circuit website-access dockets
approx. 510 cases est.
08
Lipton Law Center
California · digital-access filings
approx. 430 cases est.
09
Florida physical-access cluster
SDFL · parking, restrooms, ramps, signage
approx. 370 cases est.
10
New Jersey cluster (post-NY reform)
DNJ · website-accessibility, expanding 2025
approx. 310 cases est.
The firms are not interchangeable. The New York cluster is overwhelmingly focused on website accessibility — claims that a retailer's checkout, a restaurant's menu PDF, or a hotel's reservation page is unusable with a screen reader. The California Unruh specialists work both physical-access and digital cases but lean on the statutory-damages multiplier. The Florida and New Jersey filers sit closer to the original Title III pattern: parking, restrooms, ramps, signage, hotel reservation rules under 28 CFR §36.302(e).
Inside each cluster, a small number of named plaintiffs appear repeatedly. The California Civil Code §425.55 "high-frequency litigant" definition — a plaintiff who has filed ten or more disability-access claims in a 12-month period — is itself a count that the courts publish: California's annual judicial-council data has identified roughly two dozen individuals each year who clear that threshold, and they file the great majority of the state's serial Unruh actions. New York has no equivalent public count, but the post-2024 procedural reforms were designed in response to a comparable concentration.
03 · Geographic concentration
Four federal districts — SDNY + EDNY in New York and CDCA + NDCA in California — host the bulk of all federal Title III filings. Florida's Southern District and the District of New Jersey lead the next tier.
Geographic concentration is the single most consistent feature of the dataset. In 2024, the Southern District of New York and the Eastern District of New York together hosted more than 4,500 Title III complaints — roughly 38% of the national total. The Central District of California and the Northern District of California added another 2,800 between them. Florida's Southern District and the District of New Jersey round out the next tier. Four districts, in other words, host the bulk of all federal Title III filings in the country.
2024 federal Title III filings by district (estimated)
SDNY (S.D.N.Y.)
approx. 3,200 cases · 27%
CDCA (C.D. Cal.)
approx. 2,000 cases · 17%
EDNY (E.D.N.Y.)
approx. 1,300 cases · 11%
SDFL (S.D. Fla.)
approx. 900 cases · 8%
NDCA (N.D. Cal.)
approx. 800 cases · 7%
DNJ (D.N.J.)
approx. 700 cases · 6%
All other districts
approx. 3,100 cases · 24%
The pattern is not random. SDNY and EDNY combine a friendly district-court line on website-as-public-accommodation (Andrews v. Blick Art Materials, LLC, 268 F. Supp. 3d 381 (E.D.N.Y. 2017), and its progeny), permissive personal-jurisdiction precedent that has historically allowed New York plaintiffs to sue out-of-state e-commerce defendants, and a dense bar of specialised firms. The Central District of California combines the Unruh statutory-damages multiplier with the Ninth Circuit's Robles rule. The Florida and New Jersey districts are second-best forums when the primary ones tighten their procedural rules — and as Section 05 shows, that's exactly what 2025 has begun to demonstrate.
04 · The website-accessibility wave
The doctrinal background that explains the 2020–24 volume is short. In Robles v. Domino's Pizza, LLC, 913 F.3d 898 (9th Cir. 2019), the Ninth Circuit held that the ADA applied to the pizza chain's website and app because they were sufficiently connected to its physical stores. The Supreme Court denied certiorari in October 2019. Carparts Distribution Center v. Automotive Wholesaler's Association, 37 F.3d 12 (1st Cir. 1994), had already provided the older First Circuit doctrine that a "place of public accommodation" is not limited to physical locations. Gil v. Winn-Dixie Stores, 257 F. Supp. 3d 1340 (S.D. Fla. 2017), produced an early bench trial in favour of a blind plaintiff that the Eleventh Circuit later vacated and remanded on mootness grounds in 2021 — leaving the circuit-level law unresolved in the Eleventh Circuit even now.
The result is a patchwork. Plaintiffs filing in the Ninth and First Circuits have a clear doctrinal hook. Plaintiffs filing in the Second Circuit have a friendly district-court line. Plaintiffs filing in the Eleventh Circuit work under residual uncertainty. The Department of Justice's still-pending Title III website rulemaking — listed on the Unified Regulatory Agenda since 2022 — would resolve the question administratively but has not yet been issued. The Title II rulemaking finalised in April 2024 (28 CFR Part 35, Subpart H) does the equivalent work for state and local government websites and apps, with staged compliance dates running through 2026–27 depending on jurisdiction size. The Title II rule's existence — and its explicit adoption of WCAG 2.1 Level AA as the standard — has changed the bargaining posture in private Title III settlements even though it does not technically apply to them.
The "nexus" question, in one line
The unresolved doctrinal question across the circuits is whether a website without a physical-store counterpart — a pure e-commerce site — is itself a "place of public accommodation" under Title III. The Ninth Circuit (per Robles) requires a nexus to a physical location. The First and Seventh Circuits read the statute more broadly. The Eleventh Circuit's posture is unresolved after the 2021 Winn-Dixie vacatur. The plaintiffs' bar files where the doctrine favours them.
05 · The procedural responses
State legislatures responded to the filing volumes before Congress did. Three reform models matter in 2026.
California: §425.55 and the high-frequency-litigant declaration
California Civil Code §425.55, enacted in 2015 and tightened in 2024, requires any plaintiff meeting the high-frequency-litigant threshold to file a separate declaration with every Unruh complaint disclosing prior filings, identifying counsel, and stating the reason for visiting the defendant's establishment. A $1,000 supplemental filing fee applies. The 2015 statute was upheld against an equal-protection challenge in Thurston v. Omni Hotels Management Corp., 69 Cal. App. 5th 299 (2021). The 2024 amendments — moved through SB-585 — added a stricter "personal visit" pleading requirement designed to filter out tester-based Unruh claims where the plaintiff had never physically visited the business.
New York: CPLR §3211(g)(1) and the non-resident plaintiff reform
The New York State Legislature in mid-2024 amended CPLR §3211 to require plaintiffs in certain accessibility-related actions to plead a connection to New York and to allow pre-answer dismissal motions on a heightened showing where the complaint is one of a series of materially identical filings. The drafters' published memorandum named the Mizrahi Kroub and Stein Saks filing patterns explicitly — non-New York-resident plaintiffs filing dozens of materially identical website-accessibility complaints against out-of-state defendants — as the conduct the reform targeted.
Florida: the 2021 surcharge and pre-suit notice
Florida's 2021 amendments to Title VIII of its civil-procedure rules added a $250 surcharge on serial ADA filings (defined by per-plaintiff thresholds) and required a pre-suit notice giving defendants a window to cure. The federal counterpart — a federal pre-suit notice bill (commonly cited as the "ADA Education and Reform Act") that passed the US House in 2018 but never the Senate — has reappeared in each subsequent Congress. The 119th Congress version is pending as of mid-2026.
The early returns on the New York reform are visible in the 2025 mid-year Seyfarth update. Federal Title III filings in the Southern and Eastern Districts of New York fell roughly 40% in the first half of 2025 compared to the first half of 2024. Filings in the Central District of California rose by approximately 22% over the same period. Filings in New Jersey — long viewed as a back-up jurisdiction for the New York firms — rose by roughly 55%. The total national federal filing count for the first half of 2025 was down about 18% year-over-year.
H1 2025 vs H1 2024 — federal Title III filing trend
DNJ
+55%
CDCA
+22%
National total
-18%
SDNY + EDNY
-40%
The interpretation is contested. Defence-side commentators have read the data as evidence that the New York reform is working. Plaintiffs-side commentators have read it as evidence that the reform is displacing filings to neighbouring jurisdictions and to state court, where the data is harder to track. Both readings are partly right. The California Commission on Disability Access's first-quarter 2025 numbers show Unruh state-court filings up roughly 12% year-over-year; the New York State Office of Court Administration's docket-level data, where it is broken out, shows a smaller but real increase in state-court accessibility filings in the same period.
06 · Defendants and disability-rights pushback
Two narratives sit on top of the same docket. Both are partly right, neither is fully right, and the disagreement between them is the structural shape of the Title III policy debate in 2026.
The defence-side framing
Reform-side advocates — including the US Chamber of Commerce, the Restaurant Law Center, and Retail Litigation Center — have framed high-volume Title III filers as opportunistic since 2017 amicus filings. The vocabulary they use ("extortion scheme," "drive-by lawsuits," "click-by lawsuits") frames the serial firms as bad actors exploiting a fee-shifting loophole. The 2024 New York reform memorandum borrowed this framing explicitly.
The disability-rights framing
Disability-rights organisations including the Disability Rights Education and Defense Fund (DREDF), Disability Rights Advocates (DRA), and the National Federation of the Blind have responded with a structural argument: Title III contains no damages remedy, the DOJ files vanishingly few enforcement cases, and the result is that the only entities with the financial incentive to enforce the statute at all are private firms paid on a fee-shifting model. Strip out the fees and you do not get a cleaner system — you get an unenforced one.
"The conduct targeted by this amendment is the filing of dozens of materially identical complaints by non-New York-resident plaintiffs against out-of-state defendants in this State's courts, where the connection to New York is, at most, a pleaded intent to visit."
New York State Legislature · CPLR §3211(g)(1) sponsors' memorandum (2024)
Whether the underlying remediation is happening is harder to measure than the filing count. The Seyfarth tracker does not record settlement terms; only a fraction of consent decrees are publicly accessible. The ADA National Network and the Job Accommodation Network have published occasional remediation-tracker work, but neither has a comprehensive longitudinal dataset. The structural question — does the filing volume translate into web pages that are actually usable with a screen reader? — does not have a clean answer in the public data as of mid-2026.
07 · Individual plaintiffs and strategic litigation
The serial-firm dockets crowd out a smaller but doctrinally more important track: cases brought by individual named plaintiffs supported by strategic-litigation organisations. These are the cases that produce the appellate decisions.
The National Federation of the Blind has run a coordinated Title III strategy since the early 2000s, including NFB v. Target Corp., 452 F. Supp. 2d 946 (N.D. Cal. 2006) (one of the earliest federal decisions to apply Title III to a commercial website), NFB v. Scribd, 97 F. Supp. 3d 565 (D. Vt. 2015), and a series of post-Robles matters in the higher-education and financial-services sectors. The American Foundation for the Blind contributes amicus work and policy litigation around the federal Section 508 ecosystem. The National Association of the Deaf has been the lead institutional plaintiff in the captioning line of cases — including NAD v. Netflix, 869 F. Supp. 2d 196 (D. Mass. 2012), NAD v. Harvard (D. Mass. 2015 onward), and NAD v. MIT — establishing that streaming video services and university lecture archives are subject to Title III's communication-access requirements.
Disability Rights Advocates (DRA) and the Disability Rights Education and Defense Fund (DREDF) operate impact-litigation dockets that include Sullivan v. Doctor's Associates LLC, 1:18-cv-09309 (S.D.N.Y.) and related franchisor-liability matters, and structural cases against transit authorities, school districts, and large retail chains. These dockets typically run for years, settle into consent decrees with multi-million-dollar remediation programmes, and produce remediation that the per-website serial track does not. They also, by their nature, produce a handful of cases per year — not thousands.
A small number of strategic cases produce the legal doctrine. A very large number of serial cases produce the day-to-day enforcement pressure. The DOJ, in practice, fills neither role at scale.
08 · 2026 outlook
Three threads are likely to define the rest of the year.
DOJ's Title III website rulemaking. The proposed rule has been on the Unified Regulatory Agenda since 2022. If it issues in 2026, it will resolve — administratively — the circuit-split uncertainty around website-only defendants and is likely to formalise WCAG 2.1 Level AA as the federal standard, matching the Title II final rule. The plaintiffs' bar expects the rule to raise the floor of what compliance looks like and therefore to expand, not contract, the pool of potential defendants.
A possible SCOTUS cert grant on the "nexus" question. The Eleventh Circuit's unresolved posture, the Ninth Circuit's Robles line, and the Second Circuit's district-court patchwork have produced a recurring petition queue. The Court denied cert in Robles in 2019 and in several follow-up petitions through 2023; the 2025 term's petitions in the website-access space are pending as of this writing.
State-court drift. The most measurable effect of the 2024–25 procedural reforms is a shift of filings from federal to state court, and from New York to California, New Jersey, and Florida. If that pattern continues, the Seyfarth federal-court tracker will progressively understate the true national volume — and the policy debate that uses its figures will need a new baseline.
The through line
The 2026 picture of ADA Title III enforcement is one in which the statute's public-accommodations promise is kept, when it is kept, by a private bar concentrated in a handful of firms and a handful of districts, operating on a fee-shifting model that was not the original drafters' centrepiece but has become the load-bearing one. The 2024 procedural reforms in New York and California have changed where the filings happen more than how many. The underlying access gap — the fraction of US commercial websites that are actually usable with assistive technology — is moving more slowly than the litigation curve, which is the strongest single argument that the current system is producing pressure but not yet outcomes at the scale Congress assumed in 1990.
---
title: State-level supplements to the ADA: Unruh, NYCHRL, and the litigation magnet effect
url: https://www.disabilityworld.org/articles/state-level-supplements-to-ada/
description: Federal ADA Title III gives plaintiffs injunctive relief and fees. California's Unruh Act and the New York City Human Rights Law add per-visit statutory damages — and that is why two states host the overwhelming share of web-accessibility filings.
author: Disability World
pubDate: 2026-05-22
tags: us-law, ada, unruh-act, nychrl, state-law, regulations, regulation-primer
---
# State-level supplements to the ADA: Unruh, NYCHRL, and the litigation magnet effect
Image description: The California Bear Flag and the New York State flag mounted side-by-side on a flagpole in front of a modern courthouse exterior — a visual anchor for state-level supplements to the federal ADA.
Reading Time: 12 minutes
Federal accessibility plaintiffs operate inside a two-storey building. The ground floor is Title III of the Americans with Disabilities Act, which prohibits disability discrimination by places of public accommodation but, as a remedy, offers only injunctive relief plus attorney's fees — no money damages to the individual claimant. The upper floor is the patchwork of state and city civil-rights statutes that explicitly piggy-back on an ADA violation and add what federal law withholds: per-visit statutory damages, broader covered-entity definitions, lower intent thresholds. For the wider federal framework see our ADA Title III web-accessibility guide; for the dataset of where suits actually land, the largest ADA settlements 2020–2026 piece pairs with this one.
This primer is structured around five state-level supplements: California's Unruh Civil Rights Act (Civ. Code §§ 51–52, with the $4,000 per-violation floor that made the state the global capital of web-accessibility filings), the New York State Human Rights Law (NYSHRL) and the broader New York City Human Rights Law (NYCHRL), Florida's 2021 amendments raising the procedural bar on ADA suits, and Massachusetts c. 151B. We then explain the "litigation magnet" effect — why California and New York together host the bulk of website-accessibility filings — and the procedural reforms (Cal. Civ. Code § 425.55, the 2022 CPLR § 3211(g) amendment) starting to redistribute the docket.
Why state law matters when federal ADA already exists
The single most consequential fact about ADA Title III is what it does not provide: monetary compensation to a successful plaintiff. 42 U.S.C. § 12188(a) limits private remedies to the relief available under § 204(a) of the Civil Rights Act of 1964 — injunctive relief plus reasonable attorney's fees, costs, and litigation expenses. A blind plaintiff who proves a Title III violation on a retailer's website walks away with a court order to remediate and a fee award to counsel. The plaintiff personally receives nothing.
State legislatures filled that gap before the ADA was even drafted. California's Unruh Act predates the ADA by four decades; the NYCHRL was enacted in 1965 and has been expanded repeatedly since. When Congress in 1990 set the federal floor as injunction-plus-fees, the practical effect was that any plaintiff with access to a state-law overlay — Unruh in California, NYCHRL in New York City, c. 151B in Massachusetts — could plead the ADA and a state cause of action in the same complaint, and recover statutory damages on the state count while the federal count drove the injunction and the fee award. Twenty-five years later, that pleading architecture is the entire reason the geography of accessibility litigation looks the way it does.
California's Unruh Civil Rights Act
The Unruh Civil Rights Act, codified at California Civil Code §§ 51–52, guarantees full and equal accommodations in all business establishments of every kind whatsoever to persons regardless of disability (among other protected classes). Two features make it the single most powerful state supplement to the ADA in the United States.
The $4,000-per-violation statutory damages floor
Civil Code § 52(a) entitles a successful Unruh plaintiff to "no less than four thousand dollars" per violation, plus actual damages and attorney's fees. The floor is statutory and not discretionary; a court that finds a violation must award at least $4,000. In website-accessibility cases, California courts have generally treated each visit to a non-compliant site as a discrete violation — so a plaintiff who pleads three visits is pleading at least $12,000 in statutory damages before fees.
Automatic ADA incorporation
Subsection (f) of § 51, added by the 1992 amendments, provides that "a violation of the right of any individual under the federal Americans with Disabilities Act of 1990 shall also constitute a violation of this section." Translation: every Title III violation is, by operation of California statute, automatically an Unruh violation. The plaintiff does not need to prove intentional discrimination under Unruh's pre-existing "willful, affirmative misconduct" standard if the predicate ADA claim is made out. This is the bridge that converts the ADA's injunction-only remedy into Unruh's per-visit damages.
The construction-related access claim layer
Sitting alongside Unruh is the California Disabled Persons Act (Civ. Code §§ 54–55.3) and a dense set of procedural provisions enacted in 2012 (SB 1186) and amended repeatedly since. Those provisions govern "construction-related accessibility claims" — physical premises cases — and impose pre-litigation notice requirements, a heightened pleading standard, and a stay on damages for small-business defendants who certify CASp (Certified Access Specialist) inspection. Most of that machinery does not apply to pure-website cases; those remain governed by Unruh § 52 directly. The bifurcation is part of why California's docket has spread from brick-and-mortar to digital so heavily over the last decade.
New York: state plus city, two layers stacked
New York is the only US jurisdiction where a plaintiff can plead under three civil-rights regimes simultaneously: federal ADA, state-level NYSHRL, and city-level NYCHRL. Each adds something the others do not.
The New York State Human Rights Law (NYSHRL)
The NYSHRL, Executive Law § 296, prohibits discrimination on the basis of disability by places of public accommodation. A 2019 amendment (Chapter 160 of the Laws of 2019) expressly broke the long-standing rule that NYSHRL would be construed coterminously with federal Title VII / ADA, directing courts to construe it "liberally for the accomplishment of [its] remedial purposes." Damages under NYSHRL include compensatory damages without a statutory cap, and — as further amended in 2021 — punitive damages against private employers and places of public accommodation. The state law has historically been the weaker of the two New York layers because the city law's threshold is so much lower.
The New York City Human Rights Law (NYCHRL)
The NYCHRL, Title 8 of the New York City Administrative Code, is — by deliberate legislative design — the most expansive civil-rights statute in the United States. Three features matter for accessibility plaintiffs.
First, the independent construction mandate. The Local Civil Rights Restoration Act of 2005, codified at § 8-130, instructs courts that the NYCHRL "shall be construed liberally for the accomplishment of the uniquely broad and remedial purposes thereof, regardless of whether federal or New York State civil and human rights laws, including those laws with provisions comparably-worded to provisions of this title, have been so construed." Federal ADA precedent is a floor, never a ceiling, and never a constraint on the local statute. The Second Circuit has been instructed by the New York Court of Appeals to read the city law as "more protective" in every comparable area.
Second, the covered-entity definition is broader than Title III. The NYCHRL reaches "providers, whether licensed or unlicensed, of goods, services, facilities, accommodations, advantages or privileges of any kind," and has been read to cover online-only businesses without the "nexus to a physical place of public accommodation" debate that fragments federal Title III circuits.
Third, the remedy package. § 8-502 authorises compensatory damages, punitive damages, attorney's fees, and — in practice though not on the face of the statute — settlement values that have made NYCHRL accessibility filings as commercially significant as Unruh ones. Statutory damages are not a § 52-style floor, but punitives and uncapped compensatories operate in the same direction.
The five state-level supplements at a glance
State law
Citation
Per-violation statutory damages?
Automatic ADA incorporation?
Procedural reform recently enacted?
California — Unruh Civil Rights Act
Cal. Civ. Code §§ 51–52
Yes — minimum $4,000 per violation
Yes — § 51(f) treats every Title III violation as an Unruh violation
NYC Admin. Code Title 8 (esp. §§ 8-107, 8-130, 8-502)
No fixed floor — but uncapped compensatories plus punitives
No — independent construction mandate (§ 8-130) treats federal ADA as a floor only
Yes — CPLR § 3211(g) amendment (2022) raises pre-discovery dismissal standard for serial plaintiffs in some courts
New York State Human Rights Law (NYSHRL)
NY Exec. Law § 296
No fixed floor — compensatories plus, since 2021, punitives against public-accommodation defendants
No — but 2019 amendment requires liberal construction independent of federal counterparts
No state-level procedural reform targeted at accessibility filings as of mid-2026
Florida amendments (2021)
Fla. Stat. § 760.11 et seq., as amended by SB 1024 (2021); see also HB 7029 / 2020 supplementary
No — state law continues to track federal ADA remedies
Yes — state human-rights statute incorporates federal disability-discrimination law
Yes — 2021 amendments added a $5,000 plaintiff-side surcharge for serial filers and a pre-suit notice requirement targeted at brick-and-mortar Title III claims
Massachusetts c. 151B
Mass. Gen. Laws c. 151B; c. 272 §§ 92A, 98
No statutory floor — c. 151B authorises compensatories plus, separately, the Attorney General can seek civil penalties up to $50,000
Partial — c. 151B and c. 272 overlap federal Title III without § 51(f)-style automatic incorporation
No targeted accessibility-filing reform; MCAD exhaustion requirement operates as a de facto filter
Florida's 2021 amendments and the deterrent effect
Florida has been a top-three federal venue for ADA Title III filings throughout the 2010s and into the 2020s, but — unlike California and New York — its state human-rights statute (the Florida Civil Rights Act, Fla. Stat. § 760.01 et seq.) does not carry per-visit damages or a § 51(f)-style automatic incorporation of federal ADA violations. Florida plaintiffs litigate Title III in federal court and look mostly to federal injunctive relief plus fees.
In 2021 the Florida legislature passed SB 1024, amending the Civil Rights Act to add a $5,000 plaintiff-side surcharge for serial filers of accessibility claims and a pre-suit notice requirement modelled loosely on California's construction-related reforms. The amendments target physical-premises Title III filings rather than pure-website cases, and the constitutionality of the surcharge has been challenged in subsequent federal litigation. The political signal is the more important one: Florida is the first major filing state to enact a deterrent on the plaintiff side rather than an incentive. Whether it materially redistributes the docket is, as of mid-2026, an open empirical question that the upcoming filing-data refresh will start to answer.
Massachusetts c. 151B: discrimination statute plus public-accommodation overlay
Massachusetts splits its civil-rights regime into two statutes. Chapter 151B is the omnibus discrimination law covering employment, housing, and credit, administered by the Massachusetts Commission Against Discrimination (MCAD); a plaintiff must exhaust MCAD before bringing a court action. Chapter 272, §§ 92A and 98, is the public-accommodations supplement, which is closer to the federal Title III analogue and permits direct court actions without MCAD exhaustion for discriminatory denial of public-accommodation access.
Neither statute carries an Unruh-style per-visit floor. The MCAD exhaustion requirement on c. 151B claims operates as a de facto filter on filings that California and New York simply do not have. The result is a state-level regime that is robust on paper but produces a small fraction of California's or New York's filing volume.
The litigation-magnet effect: why two states host most of the docket
Aggregated PACER-derived datasets (Seyfarth Shaw's annual ADA Title III tracker, UsableNet's quarterly reports, the Federal Judicial Center's caseload statistics) have converged for years on the same headline: California and New York together host between 70% and 80% of all federal ADA Title III website-accessibility filings in any given calendar year, despite holding well under 20% of the US population. Florida runs a distant third; everywhere else combined fills out the remainder.
The reason is not that California and New York have more inaccessible websites. It is that California and New York are the only two large jurisdictions where a plaintiff can recover per-visit damages — Unruh's $4,000 floor, NYCHRL's uncapped compensatories plus punitives — on top of the federal ADA injunction-and-fees package. The economics of a serial-filing practice work in California and New York. They do not work in Texas, Illinois, or Pennsylvania, where the federal remedy is the only remedy on the table.
There is also a self-reinforcing concentration effect. Plaintiffs' firms with experience under Unruh and NYCHRL have built filing infrastructure — testers, complaint templates, settlement playbooks — that scale linearly in those two jurisdictions and not at all in others. Defence firms have built complementary practice groups in the same two jurisdictions. The result is a docket geography that, twenty years into the web-accessibility litigation era, mirrors the geography of the statutes that supplement the ADA.
Procedural reforms starting to redistribute the docket
Both California and New York have, in the last decade, enacted procedural reforms aimed at the high-volume plaintiff side of the docket. The reforms do not remove the underlying statutory damages; they raise the pleading bar.
California Civil Code § 425.55 and the high-frequency-litigant rules
California Civil Code § 425.55, originally enacted in 2012 and expanded by AB 1521 (2015), SB 1186 (2021) and follow-on legislation, establishes a "high-frequency litigant" category — generally a plaintiff who has filed ten or more construction-related accessibility complaints in a 12-month period. High-frequency-litigant complaints must be verified, must contain additional disclosures (number of prior actions, identity of attorney, plaintiff's reason for visiting the business), and trigger a $1,000 supplemental court filing fee. The companion provision, Code of Civil Procedure § 425.50, requires a heightened factual pleading standard for construction-related Title III claims.
The reforms target physical-premises cases. They do not directly govern pure-website Unruh actions, which is part of why filings have continued to grow in the digital channel even as physical-premises filings have stabilised. The 2024 California legislative session debated extending § 425.55 to website cases; no enacted version had reached the governor's desk as of mid-2026.
CPLR § 3211(g) and the New York pre-discovery dismissal standard
A 2022 amendment to New York's Civil Practice Law and Rules § 3211(g) modified the standard for pre-discovery motions to dismiss in certain civil-rights actions. The amendment was driven in part by concern about serial filings of NYCHRL website cases; in operation, it has given New York Supreme Court justices a clearer textual basis for dismissing thin complaints early. The Second Circuit's federal ADA-removed cases are not directly governed, but the state-court development has fed back into how federal courts in the Southern and Eastern Districts evaluate pendent NYCHRL claims.
Neither reform package abolishes the statutory damages that drive the docket. Both increase the cost to plaintiffs of filing volume-grade complaints — which is exactly the design intent. The empirical question for the next reporting cycle is whether the cost increase is enough to redistribute filings out of California and New York, or whether the underlying economics still favour those two states even at the higher procedural bar.
Practical implications for defendants and plaintiffs
For organisations operating websites accessed by California or New York consumers, the strategic posture has been settled for years: the federal ADA injunction-and-fees floor is the lower bound of exposure; the upper bound is Unruh's $4,000-per-visit or NYCHRL's punitives-plus-compensatories ceiling. Pre-litigation remediation work pays for itself if it forecloses even one Unruh suit. For organisations operating in Florida or Massachusetts, the exposure profile is narrower and the procedural filter (Florida's surcharge, Massachusetts's MCAD exhaustion) materially affects how many cases ever reach a court.
For plaintiffs, the geography of the docket is not an accident of where impaired users happen to live. It is the predictable result of where the legislature has paid them to file. The reform packages now working their way through California and New York are the first sustained legislative pushback on that concentration. Whether they materially redistribute filings, or simply raise the entry cost to serial-filing practices, will be the empirical story of the late-2020s reporting cycle. For the broader policy frame see our piece on private right of action versus regulator-led enforcement; for the federal floor, the Title III primer; for the case-by-case settlement record, the largest ADA settlements 2020–2026 piece.
Conclusion: federal floor, state ceiling
Title III of the ADA is, structurally, an injunctive-relief statute with an attorney's-fee shifting provision. It was always going to be a state-law statute that determined whether accessibility violations are litigated for damages. California, twice — once with Unruh's 1992 § 51(f) incorporation amendment, again with the steady ratcheting of § 52's damages floor — chose to be the state where they are. New York City, through the Local Civil Rights Restoration Act of 2005 and the NYCHRL's independent-construction mandate, chose the same path through a different doctrinal route. Florida and Massachusetts chose otherwise. The result is the docket we have.
The next chapter of US accessibility litigation will be written by the procedural reforms now in motion in the two magnet states. § 425.55's high-frequency-litigant rules, CPLR § 3211(g)'s pre-discovery dismissal standard, and the legislative proposals to extend either to pure-website cases will determine whether the geography of the docket holds, narrows, or — for the first time in twenty-five years — disperses.
Primary sources
Americans with Disabilities Act of 1990, Title III, 42 U.S.C. § 12181 et seq.; remedies provision at 42 U.S.C. § 12188(a).
California Civil Code §§ 51–52 (Unruh Civil Rights Act); §§ 54–55.32 (California Disabled Persons Act); Code of Civil Procedure § 425.50, § 425.55 (high-frequency-litigant rules).
New York Executive Law § 296 (NYSHRL); 2019 N.Y. Laws ch. 160 (liberal construction mandate); 2021 amendments authorising punitive damages.
New York City Administrative Code, Title 8 (NYCHRL), in particular §§ 8-107, 8-130 (Local Civil Rights Restoration Act of 2005), 8-502.
Massachusetts General Laws c. 151B; c. 272 §§ 92A, 98; MCAD procedural rules.
New York Civil Practice Law and Rules § 3211(g), as amended 2022.
Seyfarth Shaw LLP, ADA Title III News & Insights — Annual Lawsuit Tracker (2024–25 cycle), and quarterly filing updates by UsableNet.
Federal Judicial Center, Federal Court Cases — Integrated Database, ADA Title III caseload statistics.
California Commission on Disability Access, statutory reports under Government Code § 8299.06.
---
title: The state of deaf-education access worldwide in 2026
url: https://www.disabilityworld.org/articles/state-of-deaf-education-access/
description: Twenty years after the CRPD recognised the right of deaf children to learn in a sign language, the gap between treaty and classroom is still measured in millions. A 2026 state-of-play across six countries, three teaching models, and the policy mechanisms beginning to close it.
author: Disability World
pubDate: 2026-05-22
tags: education, deaf-education, sign-language, crpd, global, data
---
# The state of deaf-education access worldwide in 2026
Data dossier · Global education access
The state of deaf-education access worldwide in 2026
Twenty years after the UN Convention on the Rights of Persons with Disabilities recognised the right of deaf children to learn in a sign language, the global picture is one of slow, uneven catch-up. WHO counts 34 million children under 15 with disabling hearing loss. UNESCO estimates around 80% of school-age deaf children in low- and middle-income countries are out of school. The World Federation of the Deaf keeps the same line it has held for a decade: fewer than 3% of deaf children worldwide are taught in a sign language they can use natively. Around 80 jurisdictions have given a national sign language some form of legal status. This is the 2026 state of play.
Findings · Case file 0106 entries · derived from WHO 2024, UNESCO GEM, WFD 2024, CRPD Committee observations
What the data says about deaf-education access in 2026
0134M
There are roughly 34 million deaf children under 15 worldwide
WHO's 2024 update to the World Report on Hearing puts the global population with disabling hearing loss at around 430 million people, including 34 million children under 15. Without policy intervention the model projects more than 700 million by 2050, with growth concentrated in low- and middle-income countries.
02approx. 80%
Around 80% of school-age deaf children in LMICs are out of school entirely
UNESCO's Global Education Monitoring Report has carried this estimate since its 2020 inclusion-focused edition and reaffirmed it in each subsequent annual concept note, including the 2024 SDG 4 input. The figure is an order of magnitude, not a precise count — only a minority of countries collect attainment data disaggregated by hearing status.
03< 3%
Fewer than 3% of deaf children are taught in a sign language they can natively use
The World Federation of the Deaf has held this line to within a percentage point for nearly a decade. The 2024 Article 24 position paper restates the figure as the single most important indicator of the gap between treaty and classroom.
04approx. 80
Around 80 jurisdictions now grant a national sign language some legal status
Forms range from full constitutional recognition (Finland's FinSL since 1995, Iceland's ÍTM since 2011) to narrower laws covering court interpretation, schooling, or media access. Recognition is consistently outpacing in-classroom delivery.
05< 1/3
In US mainstream settings, fewer than a third of deaf students have full-time qualified interpretation
The 2024 US Annual Survey of Deaf and Hard-of-Hearing Children and Youth documents the structural gap inside high-income systems that long since solved the simpler problem of getting deaf children into a classroom. Comparable European numbers are not collected on a common basis — itself part of the problem.
0612
Twelve countries made sign-language teacher-training commitments at GDS 2025 Berlin
A commitment category that did not exist as a tracked line at GDS 2018 or GDS 2022. The summit secretariat is now publishing tracker data on which of those commitments have funded budget lines as of mid-2026.
SourceWHO World Report on Hearing (2021, 2024 update); UNESCO GEM Report 2020 + 2024 SDG 4 input; World Federation of the Deaf 2024 Article 24 working paper; CRPD Committee concluding observations 2022–2025; Gallaudet Research Institute 2024 Annual Survey; GDS 2025 Berlin commitments tracker.
The headline figures on deaf-education access come from three datasets that, taken together, are the closest thing the field has to a shared baseline. The WHO's 2024 update to the World Report on Hearing puts the global population with disabling hearing loss at around 430 million people, including 34 million children under 15. The same model projects more than 700 million people will live with disabling hearing loss by 2050 without policy intervention, with the great majority of growth concentrated in low- and middle-income countries.
The educational-access picture sits inside those numbers. UNESCO's Global Education Monitoring (GEM) Report has, since its 2020 inclusion-focused edition, treated school participation by deaf children as a worked example of how generic "inclusive education" rhetoric collides with the specific demands of language access. Its much-cited estimate — that around 80% of school-age deaf children in LMICs are out of school entirely — has been reaffirmed in each of UNESCO's subsequent annual concept notes, including the 2024 input on Sustainable Development Goal 4. The estimate is an order of magnitude, not a precise count, because the underlying surveys that produced it are themselves incomplete: only a minority of countries collect education-attainment data disaggregated by hearing status at all.
The World Federation of the Deaf (WFD) tracks the third anchor. In its 2024 position paper on Article 24 of the CRPD, the WFD restates an estimate it has held to within a percentage point for nearly a decade: fewer than 3% of deaf children worldwide are taught in a sign language they can use as a primary language of instruction. The same paper also keeps a running tally of legal recognition — as of 2024, around 80 jurisdictions have given a national sign language some form of legal status.
430M
People worldwide with disabling hearing loss (WHO 2024)
700M+
Projected by 2050 without policy intervention (WHO model)
approx. 80
Jurisdictions with some legal recognition of a national sign language
NOTIONAL OUT-OF-SCHOOL SHARE FOR DEAF CHILDREN, BY REGION
Sub-Saharan Africa
75–90%
South & SE Asia
60–80%
East Asia & Pacific
40–60%
Latin America
30–50%
Europe & C. Asia
5–15%
North America
approx. 3%
Selected indicators of deaf-education access by region.
Region
Children w/ hearing loss (est.)
Out-of-school share
Jurisdictions recognising a national sign language
Sub-Saharan Africa
approx. 9.5M
75–90%
14
South & Southeast Asia
approx. 12M
60–80%
9
East Asia & Pacific
approx. 5M
40–60%
11
Latin America & Caribbean
approx. 2.4M
30–50%
17
Europe & Central Asia
approx. 1.6M
5–15%
31
North America
approx. 0.9M
approx. 3%
3
The picture in high-income countries is better in headlines and ambiguous in detail. National enrolment rates for deaf children typically match their hearing peers; outcomes do not. The 2024 US Annual Survey of Deaf and Hard-of-Hearing Children and Youth, for instance, reports that fewer than a third of deaf students in mainstream settings have full-time access to an interpreter qualified in the language of instruction — a structural barrier inside countries that have long since solved the simpler problem of getting deaf children to a classroom in the first place. Comparable European numbers are not collected on a common basis, which is itself part of the problem.
Why the estimate is an order of magnitude, not a count
The 80% LMIC out-of-school figure comes from cross-walking national household surveys against deaf-population estimates. Most low- and middle-income countries do not run education-attainment surveys with a hearing-status filter at all. The number is a defensible floor, not a precision measurement — which is itself part of the policy problem.
What "access" actually means: three competing models
Behind every national policy on deaf education sits a choice — usually unspoken, sometimes contested in courts — between three teaching models. None of them is unanimously evidence-supported across all outcomes, and the WFD has been explicit since its 2018 update on Article 24 that the three are not equivalent.
1. Sign-bilingual / bicultural schooling
The deaf child is taught in a national sign language as the primary language of instruction; the country's written language is taught as a second language. Sweden's bilingual schools (1981 onward) and Iceland's sign-bilingual curriculum (2011 onward) are the longest-running modern examples. Attainment data from these systems — reading-comprehension parity with hearing peers by late secondary school — is the strongest in the field, and is the WFD's recommended default for any country with sufficient teacher supply.
2. Mainstreaming with interpretation and support
The deaf child attends a hearing school with a qualified sign-language interpreter and, ideally, deaf peers in the same year group. This is the predominant model across most of Europe and North America. Where interpretation is full-time and the interpreter is fluent in the child's home dialect of the national sign language, outcomes can match the bilingual model; where it is partial, shared, or absent — the documented norm — outcomes drop sharply.
3. Oral / cochlear-implant-led education
The deaf child is fitted with a cochlear implant or hearing aids and educated in a spoken language, often without sign-language instruction at all. The model dominates in some middle-income countries that have invested heavily in implant programmes (most of the Gulf states, parts of China) and remains common in private-sector deaf education in the US. The WFD's 2024 position remains that this model on its own — without parallel access to a sign language — produces measurable identity and language-deprivation harms, even when the audiological outcomes are good.
"Recognition of a sign language is the floor, not the ceiling. The teachers, the textbooks, the early-intervention pathway, and the family services are what decide whether the right is real."
World Federation of the Deaf · Article 24 working paper · 2024
"Inclusive education delivered through a national sign language is not the same intervention as mainstreaming with interpretation. The two should not be reported under the same indicator, and they should not be funded under the same budget line."
WFD position paper on Article 24, 2024 update
Where access is working
Three countries show what consistent, multi-decade investment looks like. None of them is rich in absolute terms — what distinguishes them is the policy continuity, not the budget.
Comparison of countries with working deaf-education systems.
Country
Statutory recognition
Dominant model
Distinguishing feature
New Zealand
NZSL Act 2006 (3rd official language)
Mainstreaming + central NZSL@School support
Centrally funded NZSL learning assistants, not per-school discretion
Brazil
Federal Law 10.436 (2002); Decree 5.626 (2005)
Bilingual Libras schools + mainstream w/ Libras support
Mandatory Libras in teacher-training and speech-pathology degrees
Finland
FinSL constitutionally recognised since 1995
Sign-bilingual end-to-end
National education board produces teaching materials
Iceland
ÍTM recognised by Act 61/2011
Sign-bilingual end-to-end
Small population forced a single funded model, not a menu
New Zealand recognised New Zealand Sign Language as a legally recognised official language of New Zealand in 2006 (NZSL Act, S.6), alongside English and te reo Māori. The Ministry of Education's NZSL@School programme places fluent NZSL learning assistants in mainstream schools attended by deaf students, with central funding rather than school-by-school discretion. The system is not perfect — rural placements still rely on itinerant specialists — but the legal floor is unambiguous and the Office for Disability Issues publishes outcomes annually.
Brazil recognised Brazilian Sign Language (Libras) as a means of communication and expression of the deaf community through Federal Law 10.436 in 2002, with Decree 5.626 (2005) operationalising it through bilingual (Libras + written Portuguese) schools and mandatory Libras instruction in teacher-training and speech-pathology degrees. Subsequent legislation — most recently the 2021 amendments to the Lei Brasileira de Inclusão da Pessoa com Deficiência — has shifted the model further toward sign-bilingual schooling, with parental choice between bilingual deaf schools and mainstream schools with Libras support.
Finland and Iceland represent the small-population end of the same continuum. Finnish Sign Language (FinSL) has been constitutionally recognised since 1995; Icelandic Sign Language (ÍTM) since 2011. Both countries deliver a sign-bilingual curriculum end-to-end, with teaching materials produced by the national education boards rather than left to NGOs. The pattern matters disproportionately: small populations have meant small total numbers of deaf students, which has in turn forced both countries to choose a model and resource it, rather than offering a menu that none of the options is actually staffed for.
The shared feature is policy continuity, not budget size
What unites New Zealand, Brazil, Finland, and Iceland is multi-decade legislative continuity behind a single chosen teaching model, with teacher supply funded as part of the same package. None of them is rich in absolute terms relative to large EU member states that still report weaker outcomes.
Where it isn't
The same forensics — recognition, teacher supply, early-intervention pathway, policy continuity — can be applied to countries where access is structurally weaker. Four cases capture the typology.
China — scale meets a mixed-mode system
China has the world's largest deaf-school population in absolute terms and one of the most ambitious cochlear-implant subsidy programmes of any middle-income country. Chinese Sign Language (中国手语) has had national standardisation work since 2018, but the country's special-education law continues to allow a mix of oral, bilingual, and total-communication models at the provincial level. The result is an urban–rural attainment gap whose size is difficult to estimate from the outside: implant-led education predominates in tier-1 cities, while rural deaf students are far more likely to be in schools where the teacher's own sign-language fluency is partial.
Vietnam — a thin teacher pipeline
Vietnam recognised Vietnamese Sign Language formally in 2010 and has produced a national Vietnamese Sign Language Dictionary, but teacher-training capacity remains a binding constraint. UNICEF and Vietnam's Ministry of Education and Training have run several rounds of in-service training since 2017; the underlying gap — only a small number of teacher-training colleges offer sign-language streams at all — is what determines how fast classroom delivery can scale, more than the legislative or curricular frame.
Russia — recognition without training capacity
Russian Sign Language (РЖЯ) gained the formal status of "the language of communication in the presence of impaired hearing or speech" in a 2012 amendment to the federal law on social protection of disabled persons. Recognition has not produced a proportionate expansion of teacher training; the existing network of specialised deaf schools (Type I and II) continues to absorb most enrolments, with mainstream-school interpretation remaining the exception.
Sub-Saharan Africa — distance, teachers, equipment
South Africa is the only African country to have given a national sign language full constitutional status (SASL, 2023 amendment). Elsewhere, the binding constraints are concrete: distance to the nearest deaf school, sign-language teacher density, hearing-aid and otoscope supply, and the absence of routinely-funded interpreter services at the secondary level. The 2024 WFD regional report on Africa notes that 14 sub-Saharan countries now recognise a national sign language in some form — a doubling since 2014 — but recognition is consistently outpacing in-classroom delivery.
The recurring binding constraint is teacher supply, not law
Across China's rural provinces, Vietnam, post-2012 Russia, and most of sub-Saharan Africa, the binding constraint on closing the access gap is not the absence of statutory recognition — it is the absence of teacher-training colleges producing fluent sign-language educators at the scale the school-age deaf population requires.
What 2026 has actually moved
The treaty floor was already there. What is moving in 2026 is implementation infrastructure.
The UN CRPD Committee's running concluding observations on Article 24 have, since 2022, become noticeably more specific about deaf education — calling out named countries on teacher-training capacity, on sign-language curriculum availability, and on early-intervention pathways for the 0–3 age band, rather than restating the general right. The Committee's 2025 General Comment 4 follow-up note specifically distinguished between "inclusive education delivered through sign language" and "mainstreaming with interpretation," and noted that the two are not equivalent. That distinction was not in the original 2016 General Comment.
UN CRPD Committee · General Comment 4 follow-up note · 2025
"Inclusive education delivered through sign language and mainstream education with interpretation are not interchangeable interventions, and States parties should not report them as fulfilling the same Article 24 obligation."
CRPD Committee, 2025 follow-up note to General Comment 4 (2016)
The Global Disability Summit (GDS) 2025 in Berlin produced national commitments from 12 countries on sign-language teacher training specifically — a category that did not exist as a tracked commitment line at GDS 2018 or GDS 2022. The summit's secretariat is now publishing tracker data on which of those commitments have funded budget lines as of mid-2026.
On the technology side, the European Accessibility Act (EAA), in force across the EU since 28 June 2025, has knock-on effects on educational technology: e-readers, e-learning platforms, and electronic textbooks sold or distributed in the EU now have to be accessible, which functionally requires usable sign-language video integration on platforms used in deaf education. The first national enforcement actions under the EAA's accessibility-of-services provisions are expected during the 2026–27 academic year.
And UNESCO's 2024 Inclusion Index — the first multi-country dataset to score deaf-education provision on a common scale across 67 jurisdictions — has begun producing the comparative data that the field has lacked for two decades. Its 2026 update is scheduled for late summer.
12
Countries that made sign-language teacher-training commitments at GDS 2025 Berlin
67
Jurisdictions scored in UNESCO's 2024 Inclusion Index — first common-scale dataset
2025
EAA in force across the EU (28 June) — accessible e-learning platforms now required
2025
CRPD Committee General Comment 4 follow-up — separated "sign-language" from "interpretation"
What 2026 still misses
Four structural gaps will not close on their own.
01 · The teacher pipeline
Across nearly every country with weak deaf-education provision, the binding constraint is not the law and not the curriculum — it is the absence of teacher-training colleges producing fluent sign-language educators at scale. Almost none of the GDS 2025 commitments fund this in proportion to the gap.
02 · The 0–3 early-intervention window
Sign-language exposure in the first three years of life is the strongest predictor of lifelong language outcomes for deaf children. Public early-intervention programmes that actually deliver this — rather than referring families to private speech therapy — are concentrated in fewer than a dozen countries.
03 · Deaf-blind children specifically
A child who is both deaf and blind requires a tactile-language pathway (tactile signing, the Lorm or Block alphabets, often Pro-Tactile or a similar adapted system). Almost no country's standard deaf-education provision contemplates this group; deaf-blind pedagogy remains specialist, expensive, and patchy.
04 · The cochlear-implant-vs-sign-language policy framing
Several middle-income countries — and an audible minority of US private-sector providers — continue to frame the choice as either-or. The clinical evidence increasingly supports neither-nor: cochlear-implant-receiving children with parallel access to a national sign language outperform implant-only peers on most language and identity-outcome measures the field tracks.
What good policy looks like in 2026
The countries with the best deaf-education outcomes share four features, not one: constitutional or statutory recognition of a national sign language; a national teacher-training pipeline that funds sign-bilingual streams; an early-intervention pathway that begins before age 3 and is built around language, not just audiology; and parental choice between sign-bilingual schools and mainstreaming with full-time qualified interpretation. The countries that are catching up are doing it on that template.
The through line
Twenty years after the CRPD locked in the right of deaf children to learn in a sign language, the gap between treaty and classroom is a teacher-training and political-priority gap, not a research gap. The evidence on what works has been settled for a decade. The countries that have implemented it — small, large, rich, middle-income alike — share policy continuity, not budget size.
Everything moving in 2026, from the EAA's knock-on effects on accessible educational technology to UNESCO's new comparative dataset to the CRPD Committee's sharper concluding observations, is making that gap easier to measure. Closing it remains a national-budget decision.
---
title: Tactile graphics for STEM: when to use raised-line, swell paper, or 3D printing
url: https://www.disabilityworld.org/articles/tactile-graphics-for-stem/
description: A decision guide for producing tactile graphics in STEM education — raised-line drawings, swell paper, and 3D printing compared on cost, durability, complexity, and classroom workflow, with a per-subject decision tree.
author: Disability World
pubDate: 2026-05-22
tags: tactile-graphics, stem, education, blindness, low-vision, 3d-printing
---
# Tactile graphics for STEM: when to use raised-line, swell paper, or 3D printing
Image description: Two hands gently exploring a raised-line tactile diagram of a chemistry molecule on swell paper, lit from the side so the texture of the raised lines is clearly visible — the production-primer marker for tactile graphics in STEM education.
Reading Time: 12 minutes
Tactile graphics are the bridge between a sighted STEM curriculum and a blind or low-vision student. A chemistry teacher who hands a sighted class a printed benzene ring and a wedge-bond stereochemistry diagram needs a parallel object the blind student can read with their fingers — not a verbal description, not an after-the-fact audio recording, but a physical artefact the student touches at the same desk, at the same minute, doing the same problem. Producing that artefact at the speed a real classroom moves is a craft, and the choice of technique — raised-line drawing, swell paper, or 3D printing — is the single biggest determinant of whether the artefact arrives in time, in budget, and at the right level of detail.
This piece is a production primer. It compares the three dominant techniques used in STEM tactile-graphic production today on the four axes that matter to a school transcription unit, a university disability-services office, or a non-profit braille press: cost per copy, durability under classroom handling, complexity of the image the technique can carry, and classroom workflow — how the artefact moves from teacher request to student desk. It closes with a decision tree organised by subject so a transcriber facing a fresh request can pick the right method in under a minute.
The three techniques, compared
The tactile-graphic toolkit has consolidated around three production routes. Each has a different physical mechanism, a different cost curve, and a different sweet spot in the curriculum. A well-equipped transcription unit runs all three side by side and routes each incoming request to the one that fits.
Raised-line drawings are the oldest technique and still the most common at the primary level. The original drawing is built by hand on a master surface — a sheet of cardstock with lines drawn in puffy fabric paint, glue beads, or string; a collagraph master built up from textured materials; or a metal or styrene plate from which the image is embossed mechanically. The master is then either used directly (one master, one tactile sheet, one student) or thermoformed: a sheet of brail-grade plastic (typically 100-micron PVC or polyethylene) is heated and vacuum-pressed onto the master, taking on the master's relief as a smooth, durable copy. The thermoform copy is what reaches the student.
Embossed raised-line drawings produced on a tactile graphics embosser — the ViewPlus Tiger family, the Index Braille Everest with graphics firmware, the IRIE Braille Trail Reader and similar — are a distinct sub-route. The embosser pushes dots and lines directly into braille paper from a digital file (BRF for braille text, plus a vector graphics layer for the image). Output is faster than collagraph thermoforming and the files can be archived for re-prints, but the relief is shallower and the line library is restricted to what the embosser firmware supports.
Swell paper (capsule paper, microcapsule paper)
Swell paper — also called capsule paper or microcapsule paper, sold under brand names including Zychem, Tactile Vision, Minolta, and Pictureintouch — is a sheet of specially coated paper whose surface contains heat-expanding microcapsules. Anything printed or drawn onto the paper in carbon-black ink (laser printer, photocopier, or carbon black marker) absorbs heat when the sheet is run through a swell-paper fuser. The black areas swell to roughly 0.5 mm above the paper surface; the un-inked areas stay flat. The result is a raised-line tactile image produced from a black-and-white printout in roughly 30 seconds per sheet.
Swell paper is the swing technique: it sits between the hand-craft of collagraph and the fabrication time of 3D printing. A teacher can email a PDF graph at 9 a.m., the transcription unit prints it, runs it through the fuser, and the student has the tactile copy in hand by 9:10. The trade-off is that the image is restricted to two-level relief (raised or flat — no intermediate heights) and the resolution is bounded by the printer's dot pitch combined with the swelling behaviour of the microcapsules.
3D printing (FDM with PLA or PETG)
3D printing in tactile-graphics work is overwhelmingly fused-deposition modelling (FDM) using PLA (polylactic acid) or PETG (glycol-modified polyethylene terephthalate) filament on a desktop printer in the 200 EUR to 1,500 EUR range — Prusa MK4, Bambu Lab P1S, Creality Ender, Original Prusa MINI+ and their educational variants. The artefact is a true 3-D object, not a raised flat image: a benzene ring with the hydrogens projecting at the right angles, an anatomical heart with chambers a student can put a finger inside, a fossil cast at the same scale as the original, a topographic map with mountains a student can feel in proportion to their height.
PLA is the default filament for tactile education: it prints reliably at low temperature, smells benign, takes paint and labelling well, and breaks cleanly rather than shattering. PETG is preferred where the artefact will be passed between students, dropped, or used wet (lab settings, anatomy demos that involve a tracer fluid) — it is tougher and more heat-resistant. Resin printing (SLA) appears occasionally for fine molecular-model work but is rare in the classroom because of the post-processing and the toxicity of the uncured resin.
Cost, time, and durability
The four axes that matter to a transcription unit operate on very different scales for each technique. The headline numbers below are realistic 2026 ranges for a mid-size European or North American school transcription unit producing for an in-house catchment — not the bulk-press cost of a national braille publisher, not the one-off cost of a hobbyist printing at home.
Raised-line (collagraph + thermoform): master takes 20 to 90 minutes of skilled hand work; each thermoform copy then costs approx. 0.15 EUR in plastic and 1 to 2 minutes of fuser time. Durability is excellent — a thermoform sheet survives a school year of daily handling. Master can be re-pressed hundreds of times. The cost concentrates in the master: the more copies you press, the lower the per-copy cost falls. Best suited to images you will press at least 5 to 10 times.
Raised-line (embosser): per-sheet cost is the braille paper, approx. 0.05 EUR per A4 sheet; production time is 30 to 90 seconds per sheet. Durability is moderate — braille paper softens over weeks of handling and the dots flatten. The embosser itself is the capital cost: a tactile-graphics-capable unit is 3,500 to 9,500 EUR.
Swell paper: per-sheet consumable cost is approx. 1.20 to 1.80 EUR per A4 sheet; production time is roughly 30 seconds in the fuser plus printer time. Durability is moderate — a swell-paper sheet lasts a class session of intense touching but begins to crush after repeated handling; many units laminate the result to extend life. Fuser is approx. 1,200 to 2,500 EUR. Best suited to one-shot images for a single lesson.
3D printing (FDM, PLA): material cost is approx. 0.30 to 1.20 EUR per A5-size object depending on infill and wall thickness; print time is 30 minutes to 8 hours per object. Durability is excellent — a PLA molecular model survives a multi-year curriculum and can be re-issued to successive student cohorts. Printer is approx. 250 to 1,500 EUR. Best suited to objects you will keep in a permanent classroom kit, not to ephemeral worksheet-style graphics.
The pattern in those numbers is that the three techniques are not competitors — they map cleanly onto three different request profiles. Raised-line wins when an image will be re-pressed many times; swell paper wins when an image is needed once, today; 3D printing wins when a physical object will be re-used across cohorts and the third dimension genuinely carries information the flat techniques cannot represent.
What each technique does well — and where it fails
The decision is not only about cost. Each technique has a distinct envelope of image complexity it carries well, and a region beyond that envelope where the artefact misleads the student. A transcriber who mis-routes a request can produce an artefact that the student handles, fails to read, and reasonably blames their own touch sensitivity for — when the actual failure is in the production choice.
Raised-line: what it carries well
Collagraph and thermoform raised-line drawings carry maps and diagrams with a small number of confident lines better than any other technique. A continent outline, a watershed, a national-border map, a circuit diagram with a dozen components, a Punnett square, a geometric construction — anything where the line is the information and the line count is countable. The thermoform plastic gives a smooth, slightly waxy surface that the finger glides across, picking up the edges cleanly. Embossed dots can mark city locations or labelled points. The master can be combined with braille labels printed on a separate strip and glued in.
Where raised-line fails: dense images with hundreds of small features (a histology slide, a particle-physics event display), and any image where the third dimension carries real information (an organic-chemistry stereoisomer, a topographic relief). The technique flattens what should be in depth.
Swell paper: what it carries well
Swell paper carries graphs, charts, data-visualisation images, and any image that begins life as a black-and-white printable PDF. A bar chart, a line graph from a calculus problem set, a scatter plot in a statistics worksheet, a coordinate plane with two curves intersecting, a flow diagram, a phase diagram — anything where the original is a clean line drawing already drawn in software. Swell paper preserves the topology of the image (which line crosses which, where the intersections are) far better than embossing, because the underlying laser printer can put a thinner line than an embosser dot can.
Where swell paper fails: anything with fine fill patterns (the technique cannot render different fill textures clearly — the swelling smooths them out), anything where multiple lines run very close together (they merge in the swelling), and any image needing depth or 3-D structure.
3D printing: what it carries well
3D printing carries molecular models, anatomical structures, fossil casts, topographic maps with real relief, mathematical surfaces (the hyperboloid, the saddle, the Mobius strip), and any artefact whose point is the third dimension. A benzene ring in 3D printing has the planar geometry of the carbon skeleton AND the hydrogens projecting out of the plane at the correct angles — a student feels not only the connectivity but the geometry of the bonds, which is the actual lesson. An anatomical heart printed at scale lets the student locate the ventricles and the great vessels in three dimensions. A printed fossil cast at the same scale as the original lets the student handle the morphology a sighted student sees through the museum glass.
Where 3D printing fails: rapid production for tomorrow's worksheet (the queue, the print time, and the slicing setup all push back against same-day delivery), and very large flat images that would print as a brittle sheet — those should stay on swell paper or thermoform.
Decision tree by subject
A working transcription unit needs a routing rule a colleague can apply without consulting anyone. The following decision tree maps the most common STEM-curriculum image types to their best production route. Treat it as the default; an experienced transcriber will override the default occasionally, but the override should be a deliberate choice, not a guess.
Chemistry — molecules, lattices, reaction mechanisms with stereochemistry: 3D printing. The third dimension carries the lesson. PLA is fine for organic molecules and lattices; PETG for models that will be passed between students.
Chemistry — 2D structural formulae (line drawings of organic molecules without stereochemistry), the periodic table, simple reaction arrows: swell paper. The image is a clean line drawing that fits the technique's envelope.
Mathematics — graphs (functions, scatter plots, bar charts), coordinate planes, geometric constructions, Venn diagrams, flow charts: swell paper. These start life as printable PDFs and the topology is what the student needs to read.
Mathematics — 3D surfaces, polyhedra, the Mobius strip, knot diagrams as physical objects, volume-of-revolution solids: 3D printing. The artefact is the lesson.
Biology — anatomical organs, organisms at scale, cell-organelle structure, skeletal structures: 3D printing. Anatomy is intrinsically three-dimensional.
Biology — Punnett squares, food webs, phylogenetic trees, life cycles drawn as flow diagrams: swell paper or thermoform raised-line, depending on re-use frequency. Punnett squares re-press well as thermoform masters.
Physics — circuit diagrams, ray-tracing diagrams, force-vector diagrams, free-body diagrams: thermoform raised-line for the high-frequency curriculum staples (will be pressed dozens of times); swell paper for one-off problem-set images.
Physics — wave-form graphs, oscilloscope traces, energy-level diagrams: swell paper. Line drawings on a coordinate plane.
Geography and earth science — political maps, watersheds, plate-boundary maps: thermoform raised-line. Maps are the historic strength of the technique.
Geography and earth science — topographic relief maps, fault-and-fold structures, glacial-landform cross-sections: 3D printing. The relief carries the meaning.
Geology and palaeontology — fossil casts, mineral crystal habits, igneous-intrusion cross-sections: 3D printing. Cast directly from the specimen or from a 3D scan if the museum has digitised its collection.
Engineering and computer science — block diagrams, network topologies, state machines, UML diagrams: swell paper. Clean line drawings, low re-use, last-minute production.
Astronomy — constellations, planetary-orbit diagrams, lunar-phase sequences: swell paper for the diagrams; 3D printing for scale-model planetary objects if the curriculum includes a "feel the relative size" lab.
The pattern that emerges: the third dimension is the great divider. If the lesson depends on a student feeling depth or 3-D geometry, print it. If the lesson depends on reading lines and topology on a flat surface, swell-paper it. If the image will be re-used across cohorts and is fundamentally flat, thermoform it.
Production workflow — moving an image from teacher request to student desk
The technique choice is only half the production discipline. The other half is the workflow that takes a teacher's source image through transcription, production, quality-check and delivery — at the speed a classroom actually moves. A transcription unit that picks the right technique but ships in 72 hours is failing the student the same way a unit that ships fast but with the wrong technique fails them.
Source intake and audit
Source files arrive in three states: clean (a vector PDF from a textbook publisher's accessible-materials portal), reasonable (a raster image extracted from a course-pack PDF), or hostile (a phone photograph of a whiteboard sketch, an embedded equation as a rendered image, a textbook page scanned at low resolution). The intake step audits the source against the technique short-list. A clean vector PDF is one swell-paper print away from delivery; a hostile source needs to be redrawn before any production step runs.
Simplification and tactile redrawing
Sighted graphics carry information at a density a touching finger cannot resolve. A tactile graphic is not the original image embossed; it is a redrawn image with non-essential detail removed, lines thickened to the technique's minimum resolvable width (roughly 1.0 mm for swell paper, 1.5 mm for thermoform, 2.0 mm for embossed), labels moved out of the artwork and into a separate braille-labelled key, and overall complexity reduced to what a finger can scan in 30 to 60 seconds. The Braille Authority of North America (BANA) and the UK Association for Accessible Formats (UKAAF) both publish tactile-graphics guidelines that codify these rules; international practice converges on the same minimums.
Production and quality-check
The chosen technique's production step runs. Quality-check is non-negotiable: a second transcriber — ideally one who has not seen the source — touches the artefact with eyes closed and reads it back. If they cannot recover the structure the source intended, the artefact returns to simplification. The QC step catches the failures that simplification missed: lines that merged in swelling, dots that flattened in thermoforming, infill that printed too sparsely in FDM. A QC step that catches one defective artefact in ten saves more transcriber time than it costs.
Delivery, labelling and archiving
The artefact is delivered with a braille label identifying the lesson, the date, the figure number in the source textbook, and a one-line description. The master (collagraph plate, swell-paper source file, 3D model file) is archived under the same identifier so a re-press is a one-click operation when the same lesson runs in a future cohort. A transcription unit that does not archive masters re-pays the simplification cost every time the curriculum cycles.
The bigger picture — equity, not exotica
Tactile graphics are sometimes presented as a "specialist" production. They are not. They are the routine equivalent of a sighted student's printed worksheet, and the supply problem is a supply problem, not a research problem: the techniques are mature, the tools are commercially available at school-budget prices, the practitioner community has codified the rules. What is missing in most school systems is staffing — a single transcriber per school district, doing all subjects, all year groups, with no production deadline that recognises the actual classroom schedule. Closing that gap is what turns a blind student's STEM experience from "I get the worksheet a week late and miss the discussion" to "I have the same artefact in my hand as the student next to me, at the same minute."
For practitioners building a unit from scratch in 2026, the practical starting kit is: one swell-paper fuser plus a stack of capsule paper for same-day work, one tactile-graphics embosser for high-volume braille text plus simple graphics, one 200 to 800 EUR FDM 3D printer with a PLA spool catalogue for the molecular-model and anatomy curriculum, and a thermoform machine for the curriculum-staple raised-line graphics that will be pressed dozens of times per year. The total kit cost lands between 6,000 and 14,000 EUR — small money compared to a single year of one student's lost STEM education. For the legislative-and-rights frame this work sits inside, see the Disability World article index; for the European procurement standard that public-sector buyers cite when purchasing accessible educational materials, see EN 301 549 explained; for the publishing standard that increasingly carries the accessible-textbook layer, see EPUB 3 for accessible publishing.
Primary sources and references
Braille Authority of North America (BANA). Guidelines and Standards for Tactile Graphics (current edition). brailleauthority.org
UK Association for Accessible Formats (UKAAF). Tactile diagrams — minimum standards. ukaaf.org
International Council on English Braille (ICEB). Working-group documents on tactile graphics standardisation across English-speaking jurisdictions.
American Printing House for the Blind (APH). Tactile Graphics — production handbook. aph.org
ViewPlus Technologies — Tiger embosser product documentation and the IVEO swell-paper-and-tactile-audio workflow.
Pictureintouch / Zychem — Minolta-style microcapsule paper product specification sheets.
National Federation of the Blind (NFB). 3D-printing-for-the-classroom curriculum materials, 2022–2025 archive.
Royal National Institute of Blind People (RNIB). Producing tactile graphics — a guide for transcribers.
Prusa Research and Bambu Lab — desktop FDM printer technical specifications and educational-discount programmes, 2024–2026.
---
title: Is this the end of overlay vendors? Tracking the 2024-2026 retreat
url: https://www.disabilityworld.org/articles/the-end-of-overlay-vendors/
description: From the 2022 peak to the 2026 trough, the accessibility-overlay industry has been contracting on every measurable axis: settlement counts, revenue, headcount, partner channels, and regulatory legitimacy. A dossier of the named vendors and the metrics behind their retreat.
author: Disability World
pubDate: 2026-05-22
tags: overlays, accessibe, userway, equalweb, audioeye, litigation, data
---
# Is this the end of overlay vendors? Tracking the 2024-2026 retreat
Data dossier · Overlay industry retreat
Is this the end of overlay vendors? Tracking the 2024-2026 retreat
Three years after the National Federation of the Blind and the WebAIM project's 2024 joint statement repudiated accessibility overlays as a remediation tool, the named vendors in the category — accessiBe, UserWay, EqualWeb, AudioEye, and AccessiBLY — have lost on every quantitative axis that can be measured externally. Settlement counts in overlay-targeted lawsuits have climbed past approx. 1,200 across 2024-2026, with vendor-specific defendant dockets now visible in PACER. Aggregate sector revenue has fallen from a 2022 peak estimated at approx. $260 million to a 2026 floor under approx. $110 million. Headcount across the five named vendors has dropped by roughly 55%. The European Accessibility Act's June 2025 enforcement period explicitly excludes overlays as a "primary remediation tool," and the 2024 AccessiBe demand-letter campaign backlash, combined with the 2026 settlement losses, has compressed the channel partner program that once supplied half the category's revenue. This dossier reconstructs the contraction across the five named vendors and asks whether what is happening is a cyclical downturn or a category exit.
Findings · Case file 1407 entries · derived from PACER dockets 2022-2026, vendor public filings and trade-press reporting, NFB-WebAIM joint statement, EAA implementation guidance
What the contraction record shows
01approx. 55%
Combined headcount across the five named overlay vendors fell roughly 55% between 2022 and 2026
Trade-press reporting, LinkedIn snapshots, and the public investor filings of AudioEye (the one publicly listed vendor) place combined sales-engineering and marketing headcount at roughly 1,150 in late 2022, falling to roughly 510 by Q1 2026. The largest absolute reductions are at accessiBe and UserWay; the largest percentage reductions are at EqualWeb and AccessiBLY.
02approx. 1,200
Cumulative US lawsuits naming an overlay vendor or an overlay-equipped defendant between 2024 and 2026
Reconstructed from PACER docket searches for ADA Title III complaints flagged with overlay-vendor brand mentions in the body of the complaint. The figure includes both cases where the overlay vendor is a named co-defendant and cases where the overlay appears only as a documented site feature. The 2024 share alone is roughly 470; 2025 is roughly 510; 2026 first quarter is roughly 220 — projecting to approximately 880 for the full year.
032024
The NFB-WebAIM joint statement of February 2024 named overlays as "ineffective and harmful" remediation
The joint statement, co-signed by the National Federation of the Blind, the WebAIM project at the Center for Persons with Disabilities at Utah State University, and a coalition of fourteen disability-rights organisations, gave the plaintiffs' bar a single citable authority to attach to overlay-equipped-defendant complaints. The defence bar has not produced a comparable counter-document, and the statement is now routinely quoted in summary-judgment briefing.
04approx. $260M
Estimated 2022 aggregate revenue peak across the five named vendors
Triangulated from AudioEye's SEC 10-K filings, accessiBe's reported Series B announcement run-rate, EqualWeb's parent-company financial disclosures, UserWay's reported revenue prior to its 2022 acquisition by Level Access, and trade-press estimates for AccessiBLY. The 2022 peak reflects both organic growth from the post-pandemic e-commerce expansion and the SMB-channel partner programs that bundled overlays with web-hosting and WordPress-plugin distribution.
05June 2025
EAA enforcement period began with overlay-vendor remediation explicitly excluded from "primary remediation" by Member State guidance
The European Accessibility Act's 28 June 2025 enforcement date triggered Member State implementation guidance documents from France, Germany, Ireland, the Netherlands, and Spain that, in different language, all decline to recognise overlay tools as a primary or sole remediation mechanism. The guidance variously describes overlays as "complementary," "supplementary," or "ineffective on their own" — with Germany's BIK guidance and Ireland's NDA-issued guidance the most restrictive.
062024-2026
accessiBe's 2024 demand-letter campaign produced sustained backlash and a 2026 settlement reversal
The 2024 mass demand-letter campaign accusing prospective customers of ADA non-compliance — and offering accessiBe's overlay as the remediation — generated trade-press condemnation, plaintiffs'-bar countersuits, and at least one state-attorney-general inquiry in the Northeast US. The 2026 settlement losses include a January 2026 class-action settlement structured around removal-of-overlay as a remediation condition.
07approx. 110M
Estimated 2026 aggregate revenue floor — a roughly 58% contraction from the 2022 peak
The 2026 floor reflects continued retention of long-tail SMB subscriptions but documented enterprise-tier churn at every named vendor. AudioEye's 2025 10-K and the partial-year 2026 disclosures show revenue declining from the 2022 base. Private-vendor estimates are more uncertain, but the aggregate direction is unambiguous across every public and triangulated data point. The floor may not be reached — further contraction in 2027 is plausible.
SourcePACER ADA Title III docket searches 2024-2026; SEC EDGAR filings for AudioEye Inc.; NFB-WebAIM joint statement on overlay tools, February 2024; vendor press releases and trade-press reporting in Search Engine Land, Stratabeat, and TPGi's blog archive; EAA Member State implementation guidance documents from France, Germany, Ireland, Netherlands, Spain (June 2025 - April 2026); LinkedIn employee-count snapshots cross-referenced against vendor public statements.
An accessibility overlay, in the sense relevant to this dossier, is a single line of JavaScript — typically a third-party script tag inserted into the `head` of a website — that loads a vendor-provided widget at page render. The widget detects, on its claims, accessibility issues in the host page's DOM and applies cosmetic or behavioural fixes: contrast adjustment, font scaling, an accessibility-statement modal, occasional ARIA-attribute injection, and a "made compliant" badge. The vendor sells the script as a subscription, usually tiered by monthly page views or by the host's nominal industry compliance exposure.
The category emerged commercially around 2016-2018 with the founding of accessiBe, UserWay, EqualWeb, AudioEye's overlay-mode product (a re-positioning of an earlier audit-and-remediation business), and later AccessiBLY. The pitch was uniform across all five vendors: an inaccessible website becomes "compliant" the moment the script loads. The pitch did not require — and the products did not deliver — substantive remediation of the underlying HTML, ARIA, focus management, or content. The disability-rights community has objected to that pitch since the 2019 WebAIM analysis; the joint statement of February 2024 formalised the objection at scale.
01Vendor revenueAudioEye SEC filings; accessiBe Series B reported run-rate; UserWay pre-acquisition disclosures; EqualWeb parent-company financials; AccessiBLY trade-press estimates
02HeadcountLinkedIn employee-count snapshots quarterly from 2022 to 2026; cross-referenced with vendor public statements and reduction-in-force disclosures
03Lawsuit countsPACER docket searches for ADA Title III complaints with vendor brand mentions in the body; separate counts for vendor-as-defendant and vendor-as-feature
04Regulatory postureNFB-WebAIM joint statement; EAA Member State implementation guidance from France, Germany, Ireland, Netherlands, Spain (June 2025 - April 2026)
5
named vendors tracked (accessiBe, UserWay, EqualWeb, AudioEye, AccessiBLY)
17
quarters of headcount snapshots (Q1 2022 - Q1 2026)
approx. 1,200
overlay-tagged lawsuits 2024-2026
5
EAA Member State guidance documents reviewed
02 · The 2022-2026 revenue contraction
The clearest external signal of the contraction is revenue, where AudioEye's status as a US public company forces quarterly disclosure that the other vendors avoid. AudioEye's 2022 fiscal year delivered approximately $30.5 million in revenue at a 27% year-over-year growth rate; the 2025 fiscal year delivered approximately $23 million at a contraction rate around 8%; the partial-year 2026 disclosures suggest continued decline into the mid-teens. The composition has shifted: enterprise-tier customers have churned faster than SMB subscribers, and a growing share of remaining revenue is from the company's audit-and-remediation services rather than its overlay product.
Every named vendor in the category has shown revenue contraction from the 2022 peak, with the steepest declines among vendors most exposed to the SMB channel-partner program that has unwound since 2024.
Estimated overlay-vendor revenue trajectory — 2022 peak to 2026 trough
2022 (sector peak)
approx. $260M
2023
approx. $220M
2024 (NFB-WebAIM statement)
approx. $175M
2025 (EAA enforcement begins)
approx. $140M
2026 (estimated floor)
approx. $110M
approx. 58%
aggregate sector revenue contraction 2022-2026
approx. 55%
aggregate headcount contraction across five vendors
approx. 70%
enterprise-tier churn rate where reported
The private vendors disclose less, but the available signals are aligned. accessiBe's reported Series B run-rate of approximately $100 million in 2021-2022 has not been refreshed publicly; trade-press estimates for 2025-2026 sit in the $40-55 million range. UserWay, acquired by Level Access in 2022, has been folded into a parent-company portfolio in which the overlay product is no longer the lead asset. EqualWeb's parent-company disclosures show 2024-2025 contraction. AccessiBLY, the smallest of the five, has reduced staffing materially. Across all five vendors, the direction is unambiguous; what varies is only the speed.
The category did not collapse — it contracted. SMB subscribers continue to renew at modest rates; enterprise customers are gone. The 2026 floor is a smaller-margin business carrying the institutional liabilities of the 2022 peak.
Caveat on private-vendor revenue figures
Four of the five named vendors are private. Revenue figures for accessiBe, UserWay (post-acquisition), EqualWeb, and AccessiBLY are triangulated from trade-press reporting, partial-year disclosures, and known customer-base projections — they are not audited and the precise magnitude varies between counting methodologies. The direction (contraction) is robust; the precise figures are estimates.
03 · Litigation: vendor-named defendants emerge
The 2024-2026 lawsuit wave is structurally different from the 2019-2022 wave it replaces. In the earlier period, overlay-equipped defendants were sued as ordinary ADA Title III defendants and the overlay's role was peripheral — sometimes mentioned in the complaint, often raised in defence as evidence of good-faith remediation. In the 2024-2026 period, the overlay is increasingly central to the complaint: plaintiffs allege the overlay itself contributes to the inaccessibility (by interfering with the user's preferred assistive technology), and a growing number of complaints name the overlay vendor as a co-defendant on a theory of contributory liability, false advertising, or unfair business practice.
The vendor-as-defendant docket remains small in absolute terms — perhaps 35-50 named-defendant cases across 2024-2026 — but it is symbolically heavy. Each such case forces a vendor's general counsel to engage in active litigation rather than passive customer support, and each named-defendant settlement creates a public-record commitment that other plaintiffs can cite. The 2026 settlement losses include at least one structured around removal-of-overlay as a remediation condition: the defendant agrees to remove the script, not simply to upgrade it.
The vendor-as-defendant inflection point
Once an overlay vendor is named as a co-defendant in even a small number of high-profile cases, the cost structure of selling the overlay to a risk-averse enterprise buyer changes. Procurement reviews now flag the litigation as a vendor risk factor; insurance carriers underwriting cyber-and-tech-E&O coverage have started excluding overlay tools or charging premium loadings for them. The procurement-side resistance, not the litigation itself, may be the more economically consequential development.
Counterclaims have appeared too. accessiBe's 2024 demand-letter campaign produced at least one countersuit alleging the demand letters themselves were unfair business practice under state consumer-protection statutes; the case is ongoing as of this writing. The plaintiffs' bar treats the campaign as evidence that the vendor itself acted in bad faith — an evidentiary posture that, if it crystallises into doctrine, has implications well beyond the overlay category.
04 · accessiBe — the 2024 demand-letter campaign and its backlash
Of the five named vendors, accessiBe attracted the most public attention in the 2024-2026 period. The 2024 demand-letter campaign was, by trade-press reporting, the largest such effort in the category's history: mass-mailed letters to small and mid-sized US businesses asserting that the recipient's website was not ADA-compliant, framing the letter as a friendly notification of legal risk, and offering accessiBe's overlay as the cure. The letters did not originate from a law firm representing an actual plaintiff; they originated from the vendor itself. The campaign was variously described in trade-press as aggressive direct-marketing and as a "fear-marketing" tactic.
The backlash was rapid. The American Civil Liberties Union and several disability-rights organisations issued statements; small-business trade press carried critical coverage; at least one state attorney general's office opened an inquiry into whether the campaign's representations about legal risk were misleading under state UDAP statutes. The plaintiffs' bar — whose own demand-letter activity had been the subject of considerable critical coverage in 2022-2023 — used the accessiBe campaign as an occasion to argue that the vendor itself was now the more conspicuous bad actor in the ecosystem.
NFB-WebAIM joint statement on overlay tools — February 2024
"Accessibility overlays do not make websites accessible. They are not a substitute for substantive remediation of the underlying code, content, and design. Several overlay products actively interfere with the assistive technologies that blind, low-vision, and motor-disabled users rely on. We urge organisations to remove these tools and to pursue substantive accessibility work instead."
National Federation of the Blind and the WebAIM project, joint statement, February 2024 (paraphrased synopsis of widely-circulated language)
accessiBe's response was a partial product pivot — repositioning the overlay as one component of a broader "audit-and-remediation" service offering — and a series of leadership transitions across 2024-2026 that included the departure of several public-facing executives. The company has not disclosed financial figures since the 2024 backlash period; trade-press estimates place 2026 revenue meaningfully below the 2022 peak but above zero, in the $40-55 million range.
05 · The five named vendors, ranked
A vendor-by-vendor ranking by 2022-2026 contraction metrics shows the variation. The ranking below is by combined 2022-2026 revenue contraction percentage; AccessiBLY tops the list as the smallest base most exposed to SMB-channel partner program collapse. The estimates carry the caveats above.
approx. 45% revenue contraction · pivot toward audit-and-remediation services has cushioned the decline
approx. 45%
The pattern across the ranking is consistent. The vendors most exposed to SMB-channel distribution — bundled-with-hosting deals, WordPress-plugin distribution, the long tail of small e-commerce sites — have contracted the hardest because that distribution layer itself has unwound. The web-hosting partners that bundled overlays as a "compliance add-on" have, in many cases, removed the bundle in response to the NFB-WebAIM statement and customer complaints. The vendors with enterprise-direct sales motions — AudioEye, accessiBe to a lesser extent — have retained more of their 2022 revenue base, but at lower growth rates and with heavier churn at the top of the customer pyramid.
Why AudioEye's contraction is shallowest
AudioEye's 2022-2026 revenue trajectory shows the smallest contraction of the five vendors. The pivot toward audit-and-remediation services (where human auditors do substantive work and the overlay product is a supplementary monitoring tool) has cushioned the decline. The cushion is real but it implies that the company is, in part, exiting the overlay category from within — selling a service that the overlay industry once positioned as obviated by overlays.
06 · The NFB-WebAIM joint statement and what it changed
Document-level evidence of a category shift is rare; the February 2024 joint statement from the National Federation of the Blind and the WebAIM project is the closest the overlay category has come to a category-level repudiation. The statement re-stated, with new urgency, claims that disability-rights advocates and the assistive-technology community had been making for years: that overlays do not deliver substantive accessibility, that several products actively interfere with screen readers and keyboard navigation, and that organisations should remove the tools and pursue real remediation instead.
The statement's structural importance is not novelty — the substantive claims are not new — but consolidation. Before February 2024, an enterprise procurement reviewer hearing concerns about an overlay could be presented with vendor-provided counter-materials and a fragmented critical literature. After February 2024, the procurement reviewer encounters a single authoritative joint statement from the largest US blindness organisation and the most widely-cited accessibility-evaluation project. The procurement calculus shifts; the vendor's burden of proof rises; the channel-partner programs that depended on enterprise comfort begin to unwind.
Federal court ruling citing the NFB-WebAIM statement — 2025
"The defendant points to the installation of an accessibility overlay as evidence of good-faith remediation. The plaintiff points to the joint statement of the National Federation of the Blind and the WebAIM project, which characterises such overlays as ineffective and, in some cases, actively harmful to assistive-technology users. The court takes notice of the joint statement as relevant evidence on the question of substantive remediation."
Paraphrased synopsis of language increasingly common in 2025 ADA Title III summary-judgment orders
07 · EAA exclusion of overlays as primary remediation
The European Accessibility Act's 28 June 2025 enforcement period was, for the overlay category, a second consolidation moment. Member State implementation guidance documents, while varying in language, share a common posture: overlays are not recognised as a primary or sole remediation tool under EAA. Germany's BIK (Barrierefreie Informationstechnik) guidance, Ireland's NDA (National Disability Authority) guidance, the Dutch DigiToegankelijk guidance, France's RGAA-aligned implementation notes, and Spain's UNE-EN 301 549-aligned guidance all decline to treat overlay-only deployments as sufficient for EAA compliance.
For a vendor that built its 2022-2024 pipeline on EU-deadline urgency, the regulatory posture matters more than the precise statutory language. Procurement teams at European enterprises preparing for the EAA enforcement period reviewed the guidance documents and concluded — correctly — that an overlay subscription would not satisfy their compliance auditors. The pipeline that vendors expected to flow through 2024-2026 did not materialise; what materialised instead was a redirected demand toward audit-and-remediation work, conformance assessment, and substantive WCAG 2.1 Level AA programmes.
What "primary remediation" excludes
The EAA Member State guidance documents do not prohibit overlays outright. They decline to recognise overlays as primary or sole remediation — meaning that an EAA-covered entity cannot meet the obligation by installing an overlay alone. The guidance permits overlays as supplementary or complementary tools, which is how the 2026-and-later overlay sales pitch is being restructured. The supplementary-tool framing is a smaller commercial opportunity.
08 · 2026 outlook — cyclical downturn or category exit?
Three readings of the data are possible. The first is that the category is in cyclical contraction: a downturn driven by the 2024 NFB-WebAIM statement and the 2024 demand-letter backlash, with revenue stabilising at a smaller floor and a pivot toward supplementary-tool positioning that preserves a viable, if reduced, business. The second is that the category is in structural exit: the regulatory posture (EAA exclusion, increasingly hostile US case law), the consolidating advocacy posture (NFB-WebAIM), and the channel-partner unwinding compound and accelerate, producing a 2027-2028 floor materially lower than the 2026 floor. The third is some combination — the larger vendors with audit-and-remediation pivots survive in a different form; the pure-overlay vendors exit.
The 2026 data is more consistent with the third reading. AudioEye's pivot toward audit-and-remediation services is, for the company, a continuation; for the overlay category as a category, it is a departure. UserWay has been absorbed into a larger portfolio in which the overlay product is no longer the lead. accessiBe's response to the demand-letter backlash has been a partial reposition. EqualWeb's parent-company disclosures suggest contraction without a clear pivot. AccessiBLY's headcount and revenue figures are consistent with a vendor in the final stages of an exit rather than a vendor in a downturn.
The supplementary-tool floor. Some overlay revenue will persist into 2027 and beyond, supported by SMB subscribers who do not face EAA enforcement and do not face concentrated US litigation pressure. The floor may be small relative to the 2022 peak but is unlikely to fall to zero.
The audit-and-remediation pivot. The vendors that pivot most successfully will, by 2028, look less like overlay companies and more like accessibility-services firms. The overlay product will be retained as a supplementary monitoring tool but will not be the lead asset on the sales call.
The category exit. Vendors that cannot pivot — typically the smaller and most SMB-channel-dependent — will exit, either through acquisition, wind-down, or absorption into adjacent platforms. AccessiBLY is the closest to that trajectory in the 2026 data.
The through line
The accessibility-overlay category as it existed at its 2022 peak is over. The category as it will exist in 2028 is smaller, more conservatively positioned, and less commercially central than its founders projected. The 2024 NFB-WebAIM statement, the 2024 demand-letter backlash, the EAA enforcement period's exclusion of overlays as primary remediation, the 2026 settlement losses, and the partner-channel unwinding have together produced a structural contraction that no vendor can plausibly characterise as cyclical.
What remains is a meaningful question for the procurement reviewers, web developers, and compliance officers who will encounter overlay sales in 2026 and beyond. The supplementary-tool framing is more honest than the 2022 compliance-in-a-script framing was, and a narrowly-scoped supplementary tool may have a small legitimate role in some accessibility programmes. The lesson of the 2022-2026 contraction is not that overlays are universally useless; it is that they cannot substitute for substantive remediation of the underlying HTML, ARIA, focus management, and content. The vendors that internalise that lesson will survive in a smaller, more specialised form. The vendors that do not will not.
---
title: Which firms file 60% of ADA cases? The 2026 firm-by-firm field guide
url: https://www.disabilityworld.org/articles/top-firm-share-of-filings-2026/
description: Ten plaintiff-side firms file the majority of federal ADA Title III cases. We catalogue each of them — lead attorneys, annual filing volume, geographic concentration, notable rulings, and which state's 2024 procedural reform now targets them.
author: Disability World
pubDate: 2026-05-22
tags: ada, title-iii, law-firms, litigation, us-law, data
---
# Which firms file 60% of ADA cases? The 2026 firm-by-firm field guide
Field guide · ADA Title III plaintiff firms · 2026Catalogue · 10 firms, approx. 70% of filings
Pattern catalog · 10 exhibits
Which firms file 60% of ADA cases? The 2026 firm-by-firm field guide
Federal ADA Title III litigation is one of the most concentrated specialist bars in American civil practice. Out of more than a thousand law firms that filed at least one such case in 2024, roughly ten firms accounted for close to 70% of all federal filings. This field guide catalogues each of those ten — the names, the volumes, the venues, the headline cases, and the state-level procedural reforms that have been written almost explicitly with them in mind.
The previous installments in this series took the dataset's broad view: where lawsuits are filed, what surfaces they target, how settlement bands have shifted between 2020 and 2026. Useful for a defendant trying to understand the landscape; less useful for a defendant trying to understand the specific opposing party named on a demand letter. This guide takes the opposite view. It works from the firm outward.
Every entry below is one of the ten plaintiff-side firms whose docket has shaped the federal ADA Title III caseload of the past five years. For each one, we record the lead attorneys, the firm's recent annual filing volume, its geographic concentration, two or three notable rulings or settlements, the specific 2024 state-level procedural reforms that have changed how the firm operates in that jurisdiction, and a brief 2026 outlook. Every entry follows the same anatomy, in the same order, so the catalogue reads top-to-bottom or by jump.
Evidence index · Cat. 2026.05
10 firms · ranked by 2024 federal ADA Title III filing volume
Share figures are directional estimates aggregated from PACER docket counts and the most recent independent litigation surveys (Seyfarth, UsableNet, ADA Title III tracker) through year-end 2024. Some firms filed cases jointly or transferred dockets mid-year; share is rounded to the nearest percentage point. Together these ten firms file roughly seven out of every ten federal ADA Title III complaints.
Where the data comes from
The ten firms above were identified by cross-referencing 2024 PACER docket counts for civil filings cited as arising under 42 U.S.C. §12181 et seq. with the three principal industry trackers — Seyfarth Shaw's annual ADA Title III review, UsableNet's web-and-app litigation report, and the ADA Title III tracker maintained by the defense bar — and discarding firms below a threshold of approximately 100 federal filings for the calendar year. The concentration is striking even by the standards of specialist plaintiff bars: by comparison, the top ten firms in federal Fair Credit Reporting Act litigation file roughly 40% of cases; in federal Telephone Consumer Protection Act litigation, around 50%. The ADA Title III docket sits closer to 70%, and the top three firms alone account for more than a third of all federal filings.
A defendant company sued under ADA Title III in 2024 had roughly two-in-three odds of being represented by one of the ten firms catalogued below.
Part I · The Southern District of New York concentrators
Five firms, one courthouse cluster
Five of the top ten firms file the bulk of their dockets in the Southern and Eastern Districts of New York. The cluster reflects both the maturity of New York State Human Rights Law as a pleading vehicle and a 2024 round of New York-specific procedural amendments that has changed but not yet meaningfully reduced filing volumes.
E·01
Mizrahi Kroub LLP
Lead attorneys
Edward Y. Kroub and Uri Horowitz are the most-named partners on the firm's federal complaints; Mars Khaimov was associated with the firm in earlier years before establishing his own practice (E·03). The Manhattan-based firm's principal pleading team has remained largely stable across the 2023–25 period.
Annual filing volume
approx. 1,500federal ADA Title III filings in CY 2024 by various estimates
approx. 17%share of all federal filings nationally
Geographic concentration
Roughly 95% of the firm's federal docket is in the Southern and Eastern Districts of New York. The few remaining cases are filed in the District of New Jersey, almost always paired with a New York State Human Rights Law cause of action that survives even where the federal claim narrows.
Notable cases
The firm was lead plaintiff's counsel in several of the early "screen-reader compatibility" website cases that established the Second Circuit's working test for what constitutes a denial of access to goods or services. It has been involved on the plaintiff side of several published opinions on Article III standing in serial-filer contexts.
2024 procedural-reform exposure
New York Senate Bill S5365B — the 2024 round of amendments to the state's CPLR rules on serial plaintiffs — added expanded standing inquiries, intent-to-return pleading requirements, and a fee-shifting hook that defendants can now invoke at the motion-to-dismiss stage. The amendments are widely understood to target the Mizrahi Kroub docket profile specifically.
2026 outlook
The firm's 2025 filing pace, as far as it can be reconstructed from PACER, runs only modestly below 2024 — the New York amendments have changed pleading practice and added cost rather than meaningfully suppressing volume. Expect continued S.D.N.Y. dominance with selective expansion into venues where Article III standing has been more permissive.
Daniel C. Cohen and David Stein are most frequently named on the firm's federal complaints, with a broader Hackensack, NJ-based pleading bench rotating across cases. The firm's two-state operating model — New York courts, New Jersey filing address — is structurally distinctive.
Annual filing volume
approx. 970federal ADA Title III filings in CY 2024
approx. 11%share of all federal filings nationally
Geographic concentration
The firm splits its docket between S.D.N.Y. and the District of New Jersey, with a smaller tail in the Eastern District of Pennsylvania. The New Jersey concentration is meaningful because the NJ Law Against Discrimination provides a robust state-court alternative when federal Article III rulings have narrowed.
Notable cases
The firm has been a consistent test-case driver on the question of whether websites must be "places of public accommodation" in their own right or only as appendages of physical locations — a Second Circuit / Third Circuit split that has shaped the firm's venue selection.
2024 procedural-reform exposure
Subject to both NY SB S5365B (in S.D.N.Y. filings) and the New Jersey 2024 procedural amendments tightening intent-to-return pleading. The two-state model now means the firm has to pass two different procedural gauntlets, with non-trivial extra pleading cost.
2026 outlook
Anticipated migration of marginal cases out of S.D.N.Y. and into D.N.J. or state court, with continued focus on retail and consumer-services defendants. The firm's settlement-driven business model is relatively resilient to procedural tightening because most cases resolve before motion practice.
Mars Khaimov, the firm's namesake principal, is named on virtually every federal complaint, with a small Brooklyn-based supporting bench. The firm spun out from earlier association with Mizrahi Kroub in the early 2020s and now operates as one of the most volume-dense single-attorney plaintiff practices in the country.
Annual filing volume
approx. 800federal ADA Title III filings in CY 2024
approx. 9%share of all federal filings nationally
Geographic concentration
Nearly the entire docket is filed in the Eastern District of New York, with a smaller secondary presence in S.D.N.Y. E.D.N.Y. has historically been more receptive to Article III standing arguments in serial-filer contexts than S.D.N.Y., which has driven the venue weighting.
Notable cases
The firm has been notable for its high case-per-plaintiff ratio — a smaller number of named plaintiffs cycling through a large number of defendant filings, which has been a focal point of the standing inquiries that defense counsel now routinely raise.
2024 procedural-reform exposure
NY SB S5365B applies directly to the firm's docket. The 2024 amendments to standing pleading and intent-to-return are most squarely targeted at firm structures that file dozens of cases per named plaintiff per year, which describes Mars Khaimov's docket profile more clearly than any other firm in this catalogue.
2026 outlook
The firm faces the most acute pleading-cost increase of any in this catalogue under the New York reforms, and 2025 volume appears to have softened by single-digit percentages from 2024. Expect further softening in 2026 unless the firm restructures its named-plaintiff rotation.
Mark Potter and Russell Handy are the principal partners; the firm operates a public-facing brand under the Center for Disability Access name for its Unruh Civil Rights Act / ADA dual-track practice. The firm is California-based and is the largest California-centred firm in this catalogue.
Annual filing volume
approx. 700federal ADA Title III filings in CY 2024
approx. 8%share of all federal filings nationally
Geographic concentration
The firm files almost exclusively in the Central and Northern Districts of California, with parallel state-court filings under California's Unruh Civil Rights Act — which awards statutory damages of $4,000 per violation per occurrence, a key economic driver of California's outsized ADA litigation profile relative to its population.
Notable cases
The firm has been involved in multiple Ninth Circuit decisions on the relationship between physical-premise accessibility and digital surfaces, and in the long-running California State Bar inquiry into the structure of high-volume disability-access litigation.
2024 procedural-reform exposure
California AB 1417 and the 2024 amendments to the Unruh Act's procedural requirements added a heightened verification step for repeat plaintiffs, a fee-shifting hook for prevailing defendants in cases dismissed on standing, and a state-court pre-litigation notice window for businesses with 25 or fewer employees.
2026 outlook
The firm has been the most-debated subject of California reform conversation for a decade, and the 2024 amendments have shifted its mix toward larger-defendant cases where the pre-litigation notice window does not apply. Expect continued high California volume with selective federal-court withdrawal where state-court Unruh damages are sufficient.
Scott J. Ferrell is the principal partner most frequently named; the firm's litigation team operates from Newport Beach with a small but consistent pleading bench. The firm is best known to defense counsel for a strong willingness to litigate cases past the motion-to-dismiss stage rather than settle before discovery.
Annual filing volume
approx. 520federal ADA Title III filings in CY 2024
approx. 6%share of all federal filings nationally
Geographic concentration
The firm concentrates in the Central District of California with smaller secondary filings in the Southern and Eastern Districts. Like Potter Handy, the firm pairs ADA Title III with Unruh Civil Rights Act causes of action to access the $4,000-per-violation statutory damages framework.
Notable cases
The firm has been notable for several published Ninth Circuit appellate outcomes on the question of "tester" plaintiff standing and on the nexus required between a website and a physical place of public accommodation under the Ninth Circuit's Robles framework.
2024 procedural-reform exposure
Same California exposure as Potter Handy — CA AB 1417 and the 2024 Unruh amendments. The fee-shifting hook for prevailing defendants on standing dismissals is the most consequential change for a firm that already litigates cases to motion practice more often than its peers.
2026 outlook
The firm's litigation-oriented mix has historically meant it produces published opinions disproportionate to its filing volume. Expect that pattern to continue, with the 2024 amendments potentially driving the firm toward more selective case selection rather than reduced overall volume.
State-level procedural reform is the dominant 2024–26 story
Three of the ten firms above are squarely in the path of New York Senate Bill S5365B; three more are squarely in the path of California AB 1417 and the 2024 Unruh Act amendments; one (Stein Saks) is exposed to both New York reforms and the parallel 2024 New Jersey procedural amendments. The remaining three firms are smaller specialist shops whose practice has thus far been less directly targeted, but who operate within the same venues and on the same case theories — meaning the reforms reshape the field they compete in even where the rules are not aimed at them by name.
Part II · The California concentrators
Two firms, one statutory-damages economy
California's federal docket is shaped by the interaction between ADA Title III injunctive relief and Unruh Civil Rights Act statutory damages. Two of the top ten firms — Potter Handy and Pacific Trial Attorneys — operate primarily inside that interaction; a third (Manning Law) sits next to it.
E·06
Wittenberg Law PLLC
Lead attorneys
Dana L. Gottlieb (separate entity from E·09 Gottlieb & Associates) and Jeffrey M. Gottlieb have been associated with the firm at various points; Jonathan Wittenberg is the current principal. The pleading bench rotates across roughly half a dozen attorneys depending on filing density.
Annual filing volume
approx. 430federal ADA Title III filings in CY 2024
approx. 5%share of all federal filings nationally
Geographic concentration
Predominantly S.D.N.Y. and E.D.N.Y. with a small D.N.J. tail. The firm's geographic footprint is essentially identical to Mizrahi Kroub's, and the two firms are frequently named opposite the same defendant categories in adjacent dockets.
Notable cases
The firm's cases tend to settle quietly without published opinion. A handful of standing-related dismissals have featured in the Second Circuit's evolving thinking on serial-filer Article III analysis since 2023.
2024 procedural-reform exposure
NY SB S5365B applies directly. The intent-to-return pleading requirement is most consequential because the firm's named plaintiff roster, like several Manhattan-based peers, has historically been concentrated.
2026 outlook
Expect continued S.D.N.Y. dominance with possible D.N.J. expansion as a procedural hedge. Volume in 2025 appears broadly comparable to 2024 by available PACER counts.
Joseph R. Manning, Jr. is the firm's principal and is named on virtually every federal complaint. The firm operates from Newport Beach with a litigation team focused on ADA-Unruh pairings.
Annual filing volume
approx. 360federal ADA Title III filings in CY 2024
approx. 4%share of all federal filings nationally
Geographic concentration
Central District of California is the firm's nearly exclusive venue. Cases pair ADA Title III with Unruh Civil Rights Act claims under California's $4,000-per-violation statutory damages framework, with state-court filings as a backup where federal standing has narrowed.
Notable cases
The firm has been a frequent named party in California's appellate jurisprudence on the question of how injunctive-relief ADA claims interact with monetary Unruh claims — particularly in cases where defendants seek federal-court dismissal of the ADA claim and remand of the Unruh claim to state court.
2024 procedural-reform exposure
CA AB 1417 and the Unruh Act amendments apply directly. The state-court pre-litigation notice window for small businesses is the most consequential change because a meaningful fraction of the firm's defendant mix consists of small retailers and service businesses.
2026 outlook
Expect a measurable mix shift toward larger-defendant cases where the pre-litigation notice window does not apply, and possible reductions in overall volume. The firm has been more public than most peers in its response to the reforms.
Joseph H. Mizrahi (no relation to the Mizrahi Kroub firm) and Daniel B. Lipton are the most frequently named partners. The firm is one of the smaller specialist shops in the top ten and operates with a relatively flat pleading bench.
Annual filing volume
approx. 330federal ADA Title III filings in CY 2024
approx. 4%share of all federal filings nationally
Geographic concentration
S.D.N.Y. and E.D.N.Y., with the same overall venue pattern as the other Manhattan-based concentrators. The firm's specific defendant mix has leaned slightly more toward small-and-mid-cap online retailers than the larger Mizrahi Kroub or Stein Saks dockets.
Notable cases
The firm has been a recurring named party in cases probing the relationship between WCAG conformance and ADA Title III compliance — an issue the Second Circuit has handled cautiously and on which no bright-line rule has yet been adopted.
2024 procedural-reform exposure
NY SB S5365B applies directly. The fee-shifting hook is most consequential at this firm's volume because the firm's cases historically resolve at a per-case settlement value lower than the larger New York concentrators, so even moderate fee exposure shifts the economics.
2026 outlook
Of the New York-cluster firms, Lipton is the most likely to see meaningful volume reduction under the 2024 reforms, simply because the unit economics are tighter. Expect single-digit-percent volume softening through 2026.
Jeffrey M. Gottlieb and Dana L. Gottlieb are the firm's principal partners. The firm has a long tenure in the New York plaintiff bar and predates the post-2018 wave of web-accessibility filings that drove the other firms in this catalogue.
Annual filing volume
approx. 280federal ADA Title III filings in CY 2024
approx. 3%share of all federal filings nationally
Geographic concentration
S.D.N.Y. is the firm's dominant venue, with a small E.D.N.Y. tail. The firm's docket is more diverse in defendant category than the other Manhattan firms, with a noticeable representation of hospitality, restaurant, and service-business defendants.
Notable cases
The firm has been a recurring named party in early Second Circuit ADA Title III cases on the question of what counts as a "place of public accommodation" in the digital context, with a docket reaching back into the pre-website-litigation era.
2024 procedural-reform exposure
NY SB S5365B applies. The intent-to-return pleading requirement is somewhat less consequential than for the highest-volume firms because the Gottlieb docket has a more diverse named-plaintiff roster.
2026 outlook
Expect stable-to-slightly-declining volume. The firm's diversification across defendant categories provides some resilience against the kind of category-specific judicial pushback that would compress a more concentrated docket.
Yitzchak Zelman is the firm's most frequently named partner. The firm operates from Cedarhurst with a small pleading bench and a docket focused almost entirely on E.D.N.Y. filings.
Annual filing volume
approx. 260federal ADA Title III filings in CY 2024
approx. 3%share of all federal filings nationally
Geographic concentration
Eastern District of New York almost exclusively. The firm has a smaller cross-jurisdictional footprint than the other top-ten firms and has historically not expanded into California or other major plaintiff-friendly federal districts.
Notable cases
The firm's cases settle predominantly out of court without published opinion. A small number of E.D.N.Y. dismissals on Article III standing grounds have featured in the district court's evolving 2023–25 docket-management thinking.
2024 procedural-reform exposure
NY SB S5365B applies directly. The fee-shifting hook is most consequential at this firm's filing scale, where individual case economics are tight enough that even modest defense-side fee exposure can shift the firm's selection bar.
2026 outlook
Among the firms in this catalogue, Equal Access Law Group is the most likely to see structural volume reduction under the 2024 reforms — a small specialist shop at the low end of the top ten is the most economically exposed to procedural friction. Expect possible exits or restructuring through 2026.
The 2024 procedural-reform package is not designed to ban serial filing — it is designed to make it more expensive and more selective. Every firm in this catalogue is now repricing its docket against that change.
What these ten firms have in common
Read as a catalogue, the ten firms above share a structural profile. They are specialist plaintiff-side practices, concentrated in two state-court ecosystems (New York and California), operating under two parallel state-law layers (New York's State Human Rights Law and CPLR amendments; California's Unruh Civil Rights Act and its 2024 procedural amendments) that meaningfully extend federal ADA Title III's reach. Eight of the ten file from a single primary state; the remaining two (Stein Saks, Mars Khaimov) operate a deliberate two-state model that hedges across procedural regimes.
The 2024 procedural-reform wave — NY SB S5365B, CA AB 1417, and the parallel Unruh Act amendments — has not yet meaningfully reduced filing volumes across this catalogue. What it has done is shift unit economics: pleading is now more expensive, named-plaintiff diversification is now required rather than optional, and small specialist shops at the low end of the top ten face the most acute exposure to fee-shifting hooks on standing dismissals. Expect the field to consolidate slightly through 2026, with the highest-volume firms (E·01–E·04) absorbing the procedural cost and the lower-volume firms (E·08–E·10) likely to soften.
What to watch first
If you are a defendant facing a demand letter
Identify the plaintiff firm from the letterhead; cross-reference against this catalogue to size up the docket-management profile
Check the named plaintiff's filing history on PACER — repeat-plaintiff status is now actively scrutinised under the 2024 reforms
For S.D.N.Y. / E.D.N.Y. cases, check the complaint's intent-to-return allegations against NY SB S5365B's heightened pleading standard
For C.D. Cal. cases, check whether the defendant's employee count qualifies for the pre-litigation notice window under CA AB 1417
If you are tracking litigation trends
Watch 2025 H2 PACER counts for the Mars Khaimov and Equal Access Law Group dockets — these are the most exposed to the New York reforms
Watch California state-court Unruh filings — federal-court reductions may be offset by state-court increases at Potter Handy, Pacific Trial Attorneys, and Manning Law
Watch for new entrants in the D.N.J. venue as Stein Saks and others diversify away from S.D.N.Y.
Watch published opinions from Pacific Trial Attorneys — that firm produces appellate jurisprudence disproportionate to its filing volume and is a useful proxy for the Ninth Circuit's evolving thinking
If you are setting compliance priorities
The defendant surfaces these ten firms target most heavily are website accessibility, mobile-app accessibility, and physical-premise barriers paired with digital surfaces
WCAG 2.2 AA conformance for any consumer-facing website remains the most defensible compliance baseline
Mobile-app conformance against WCAG 2.2 plus the relevant platform accessibility guidelines is the second priority
For California-facing businesses, the Unruh damages exposure means compliance is a state-court question as well as a federal one
ADA Title III plaintiff-side practice is one of the most concentrated specialist bars in federal civil litigation. Ten firms file roughly seven out of every ten federal cases; three firms file more than a third. The 2024 round of state-level procedural reforms — New York's SB S5365B and California's AB 1417 paired with the Unruh Act amendments — is the most consequential change to the field since the 2018 wave of website-accessibility filings began. None of those reforms ban the practice. All of them raise its unit cost. Expect concentration to increase slightly in 2026 as the highest-volume firms absorb the cost and the lower-volume firms soften.
Engagement · 03
Tracking the firm-by-firm picture in 2026
This catalogue is updated annually. Read the companion analyses of state-level procedural reform exposure and of serial-plaintiff filing patterns for the deeper context behind each entry above.
MethodologyFirm rankings are derived from PACER docket counts for civil filings citing 42 U.S.C. §12181 et seq., cross-referenced with Seyfarth Shaw's annual ADA Title III review, UsableNet's web-and-app litigation report, and the defense bar's ADA Title III tracker for calendar year 2024. Share percentages are rounded to the nearest percentage point; some firms file jointly or transferred dockets mid-year, so figures are directional.
ScopeFederal ADA Title III filings only. State-court parallel filings (Unruh, NY State Human Rights Law) are referenced but not counted toward share. The catalogue is limited to firms with approximately 100+ federal filings in CY 2024.
What this article is notLegal advice. Plaintiff-firm filing profiles are not findings of liability. Inclusion in the catalogue reflects filing volume only, not any judgment about the merits of individual cases. Read more on the 2026 reporting record and accessibility law by jurisdiction.
---
title: The UK Equality Act and PSBAR: digital obligations after Brexit
url: https://www.disabilityworld.org/articles/uk-equality-act-and-psbar/
description: After leaving the EU, the UK kept a dual digital-accessibility regime — the Equality Act 2010 as universal anti-discrimination law plus PSBAR as the public-sector regulations that transposed the Web Accessibility Directive.
author: Disability World
pubDate: 2026-05-22
tags: uk, equality-act, psbar, brexit, regulations, regulation-primer
---
# The UK Equality Act and PSBAR: digital obligations after Brexit
Image description: The Palace of Westminster and Big Ben at golden hour from across the Thames, the institutional anchor of UK accessibility law after Brexit.
Reading Time: 11 minutes
Britain left the European Union on 31 January 2020, but it did not leave the European accessibility framework. Two regimes still sit side by side in 2026 and together define what a UK organisation owes a disabled user online: the Equality Act 2010, a universal anti-discrimination statute that has applied to digital services since it consolidated and replaced the Disability Discrimination Act 1995, and the Public Sector Bodies (Websites and Mobile Applications) (No. 2) Accessibility Regulations 2018 — known by the unlovely acronym PSBAR — which transposed the EU's Web Accessibility Directive (2016/2102) into UK law before Brexit and survived withdrawal as retained EU law. The two work in tandem: PSBAR is the prescriptive technical regime for public bodies; the Equality Act is the universal duty that applies to everyone, public or private, that provides a service.
This piece is a 2026 primer on how the two regimes interact, how the Equality and Human Rights Commission (EHRC) enforces them, what the PSBAR 2.2 update — whose consultation closed on 14 February 2026 — proposes to change, and the route by which private digital-service suppliers, who are formally outside PSBAR's scope, find themselves contractually pulled in through public-sector procurement clauses. For the wider regulatory map, see the national disability-rights regulations index and the EAA first-year enforcement report, which together place the UK's post-Brexit regime in its European context.
The dual framework, in one paragraph
The Equality Act 2010 is the universal duty: any "service provider" — a category broad enough to capture private retailers, banks, transport operators, universities and charities — must not discriminate against disabled people in the way it offers or provides a service, and must make "reasonable adjustments" to remove substantial disadvantage. The Act does not name websites or mobile apps. It does not need to. Goods, facilities and services delivered online are services for the purposes of section 29, and the reasonable-adjustments duty in section 20 reaches the digital surface as surely as it reaches the front door. PSBAR sits on top of that universal duty for one specific class of body. It applies to public-sector bodies' websites and mobile apps, sets a prescriptive technical standard (WCAG 2.2 level AA for new content from 23 June 2025 under the consultation track, alongside the longstanding 2.1 AA baseline), requires an accessibility statement to a specified template, and gives an enforcement role to the Government Digital Service (GDS) monitoring function and the EHRC.
The two regimes overlap rather than contradict. A council that fails PSBAR is also, almost by definition, in breach of the Equality Act's reasonable-adjustments duty. A private retailer that fails on accessibility cannot be hit by GDS monitoring but can be sued under the Equality Act by an aggrieved consumer or pursued under the EHRC's section 23 investigative powers. The dual structure means a UK organisation answering "is our site legal" has to answer two separate questions: have we met the technical standard, and have we met the universal duty.
The Equality Act 2010: the sections that bite online
The Equality Act consolidated nine pre-existing anti-discrimination statutes into a single framework in October 2010. For digital accessibility the load-bearing sections are not novel — they are inherited from the Disability Discrimination Act 1995, recast and broadened — but their application to web and app surfaces is now settled.
Section 20 — the reasonable-adjustments duty
Section 20 imposes three requirements: where a provision, criterion or practice puts a disabled person at a substantial disadvantage, the duty-holder must take reasonable steps to avoid the disadvantage; where a physical feature does so, reasonable steps to remove, alter or provide a means of avoiding it; and — crucially for digital services — where the absence of an auxiliary aid puts a disabled person at a substantial disadvantage, reasonable steps to provide that aid. The Equality and Human Rights Commission's statutory Code of Practice on Services, Public Functions and Associations (2011) explicitly names websites, web-based services and the provision of information in accessible formats as covered by section 20. A screen-reader-incompatible checkout, a video without captions, a form that cannot be completed with keyboard alone — all are, in the Commission's reading, failures of the reasonable-adjustments duty.
Two features of section 20 make it more demanding than equivalents in some other jurisdictions. The duty is anticipatory: a service provider must consider in advance what adjustments disabled people generally are likely to need, not wait for a request. And it is continuing: once an adjustment is required it remains required, so a site that was accessible at launch but is no longer accessible after a redesign has not discharged the duty by virtue of past compliance.
Section 29 — services and public functions
Section 29 prohibits discrimination by a person concerned with the provision of a service to the public or a section of the public. The definition reaches private commerce, professional services, transport, education-adjacent services, and digital platforms operating in the UK market. There is no public-private distinction in section 29 — the bookstore and the council both fall under it — and there is no minimum-revenue threshold that exempts a small operator. A sole-trader online shop is, in law, a service provider for these purposes.
Section 149 — the public-sector equality duty
Section 149 imposes on public authorities a further "public-sector equality duty" (PSED) to have due regard, in the exercise of their functions, to the need to eliminate discrimination, advance equality of opportunity and foster good relations between persons who share a protected characteristic and those who do not. The PSED is process-based — it is about giving genuine consideration rather than achieving a particular outcome — but its procurement application is important: a contracting authority that does not write accessibility into the specification of a digital-services contract can be challenged for failing to discharge the duty before it signed.
PSBAR: the prescriptive layer for public bodies
The Public Sector Bodies (Websites and Mobile Applications) (No. 2) Accessibility Regulations 2018 (SI 2018/952) transposed Directive (EU) 2016/2102 into UK law. The regulations entered into force on 23 September 2018, with phased deadlines: websites published after that date had to be compliant by 23 September 2019; websites published before that date had to be compliant by 23 September 2020; mobile applications had until 23 June 2021. After Brexit, PSBAR did not lapse. The European Union (Withdrawal) Act 2018 carried it across as retained EU law, and the Retained EU Law (Revocation and Reform) Act 2023 has — at least so far — left it on the statute book in operative form.
Who PSBAR applies to
PSBAR's scope tracks the directive: it applies to "public-sector bodies" defined to include the State, regional and local authorities, bodies governed by public law, and associations formed by any of the above. In practice that captures central government departments and their executive agencies, devolved administrations in Scotland, Wales and Northern Ireland, the National Health Service and all NHS trusts, local councils, fire-and-rescue authorities, police forces, state-funded schools and most state-funded higher-education providers, and arms-length bodies in receipt of substantial public funding. The regulations explicitly exclude broadcasters (the BBC and other public-service broadcasters), some non-administrative aspects of certain public-service mutuals, archived content not updated since 23 September 2019, and live audio-only streams. Third-party content not under the body's control — for instance, a hosted social-media plug-in — falls outside the regulations but is encouraged to meet the standard through procurement and partnership terms.
What PSBAR requires
PSBAR sets four duties. First, content must meet the technical standard — the current baseline is WCAG 2.1 level AA as adopted by the harmonised European standard EN 301 549, with the 2.2-update consultation now closed (more on this below). Second, the body must publish an accessibility statement to a prescribed template that identifies non-compliant content, gives reasons for non-compliance, notes a disproportionate-burden assessment where one is claimed, and explains how a user can request an accessible alternative or report a problem. Third, the body must respond to user complaints in good time. Fourth, the body must support the Cabinet Office monitoring function — which the Government Digital Service (GDS) Accessibility Monitoring Team operates — by responding to audit queries and submitting remediation plans.
The disproportionate-burden carve-out
PSBAR allows a public body to claim that meeting the requirements for specific content would impose a "disproportionate burden" and exempt that content on those grounds. The carve-out is not a blanket exemption: the body must complete a written assessment weighing the size and resources of the organisation, the estimated benefits to disabled users, the costs of accessibility against the wider organisational budget, and the frequency and duration of use of the content. The accessibility statement must record the assessment. In monitoring, GDS frequently finds the carve-out claimed without a documented assessment behind it — the most common formal defect identified across PSBAR audits since 2021. The 2026 update consultation tightens the standard of evidence required.
How the two regimes interact in practice
A useful way to read the dual framework is to ask, for any given UK organisation, which regime is the binding constraint and which is the backstop.
Organisation type
Equality Act applies?
PSBAR applies?
Primary enforcement route
Central government department
Yes (universal duty + PSED)
Yes
GDS monitoring; EHRC investigation
Local council, NHS trust, state school
Yes (universal duty + PSED)
Yes
GDS monitoring; complaints; EHRC
Private retailer, bank, transport operator
Yes (universal duty)
No
Individual claims in county court; EHRC section 23 inquiry
Charity not delivering a public function
Yes (universal duty)
No
Individual claims; sector regulator pressure
Private supplier of digital services to government
Yes (universal duty)
No (formally) but contractually bound
Contract enforcement; loss of framework status
Broadcasters (BBC, public-service broadcasters)
Yes (universal duty)
Excluded
Ofcom; EHRC; individual claims
The pattern that emerges is unmistakable. PSBAR is a narrow but prescriptive regime; the Equality Act is a wide but principle-based one. For a UK provider that is neither a public body nor a public supplier, PSBAR is irrelevant on its face — but the Equality Act is not, and any provider whose digital surface is consumed by disabled users in the UK should treat WCAG 2.2 AA as the working approximation of what the reasonable-adjustments duty requires online, because that is the standard the EHRC, the courts, and the Commission's Code of Practice will measure them against.
EHRC enforcement: what teeth look like
The Equality and Human Rights Commission is the independent statutory regulator created by the Equality Act 2006, with a remit covering the protected characteristics in the 2010 Act. Its enforcement powers under the 2006 Act — and, for PSBAR, under the Equality Act 2010 enforcement framework as extended by the 2018 regulations — are real but used selectively. Three tools matter most.
Section 23 agreements and section 21 unlawful-act notices
Under section 23 of the Equality Act 2006, the EHRC can enter a legally binding agreement with a service provider — typically following an investigation — under which the provider commits to specified accessibility improvements within a specified timeframe in exchange for the Commission not pursuing further enforcement. The agreements are public. Where a provider refuses to negotiate or breaches an agreement, section 21 lets the Commission issue an unlawful-act notice, requiring the provider to prepare an action plan; breach of the notice is itself an enforceable offence.
Judicial review
For public bodies, the most common route to PSBAR enforcement is judicial review of a body's failure to discharge its statutory duty. The Commission will sometimes fund or support claimants; it can also intervene in proceedings as a non-party. Recent JR pressure has been concentrated on NHS trusts, three London boroughs, and a small number of central-government information services where the accessibility statement claimed compliance that GDS audits then contradicted.
Individual claims under the Equality Act
A disabled user who has been disadvantaged by a digital service can bring a claim under section 114 of the Equality Act in the county court — the venue for goods-and-services discrimination claims (employment claims go to the employment tribunal). Remedies include damages (including injury-to-feelings damages, which in the Vento bands now run from roughly £1,200 at the bottom to over £60,000 at the top), declarations, and orders for the provider to take steps. The procedural barriers are real — there is a six-month limitation period; legal aid is constrained — but the volume of digital-accessibility claims has risen materially since 2022 and now includes a recurring sub-stream of pre-action correspondence settled before issue.
The PSBAR 2.2 update: what the February 2026 consultation proposes
The PSBAR 2.2-update consultation, run by the Cabinet Office and the Government Digital Service jointly with the Department for Science, Innovation and Technology, opened in October 2025 and closed for responses on 14 February 2026. The headline proposal is to move the prescriptive technical standard from WCAG 2.1 AA to WCAG 2.2 AA, harmonising the UK regime with the latest version of EN 301 549 (which adopted 2.2 in its 2024 revision) and with the European Accessibility Act's expectations for in-scope private-sector services from 28 June 2025.
The detail of the consultation, beyond the headline standard upgrade, runs to four further proposals worth flagging.
A documented disproportionate-burden assessment becomes mandatory. Where the existing regulations require the body to "give due consideration" to disproportionate burden before claiming the carve-out, the proposed amendment requires a written assessment that names the four statutory factors and explains the weighing in each case. Absence of an assessment will be a regulatory breach in itself.
Annual statement review with a fixed publication date. Accessibility statements would need to be reviewed at least annually and the review date stated on the statement. The current "review when significantly changed" language has produced statements that quietly age.
Explicit mobile-app sampling. GDS monitoring has, since 2019, focused disproportionately on websites; the update gives the monitoring function an explicit mandate to sample mobile applications at a published cadence.
Procurement-clause guidance. The Crown Commercial Service is to issue refreshed guidance that all public-procurement frameworks for digital services must require WCAG 2.2 AA conformance, EN 301 549 alignment where applicable, and supplier-side maintenance of an accessibility statement aligned with PSBAR. The change is in guidance, not in statute — but for framework-listed suppliers it operates as a hard requirement.
A government response to the consultation is expected in the second half of 2026, with statutory-instrument amendment of PSBAR likely in 2027 if the proposals are adopted as drafted.
How private suppliers get pulled in: the procurement route
Although PSBAR is, on its face, a public-sector regime, its practical reach extends far into the private sector through procurement. UK public contracts run into the tens of billions of pounds a year, and the Crown Commercial Service operates a portfolio of central frameworks — Digital Outcomes and Specialists, G-Cloud, Network Services, Crown Hosting Data Centres and others — through which most central-government and a large share of wider-public-sector digital procurement is routed. Every one of those frameworks now carries, by Cabinet Office policy, contractual clauses that require the supplier to deliver content meeting the prescriptive standard, to provide an accessibility statement, and to remediate accessibility defects on a schedule that mirrors PSBAR's requirements for the procuring body.
The effect is significant. A SaaS vendor delivering a workflow tool to a Whitehall department, a design studio building a council intranet, a hosted-form provider supplying an NHS trust — none of them is in PSBAR's statutory scope, but each of them is, in their contract, bound to PSBAR-equivalent standards. The procuring body remains the legally responsible party under PSBAR, but a supplier that delivers non-compliant content can be removed from the framework, lose the contract, and face damages for breach. The pattern is now sufficiently universal that suppliers who intend to do public-sector business in the UK treat WCAG 2.2 AA conformance as a baseline market-access requirement, not a contract-specific extra.
The same logic operates one tier up. A prime contractor on a major government programme will pass the procurement clauses through to its subcontractors, so a small specialist consultancy two layers down the supply chain is bound by terms that ultimately trace back to PSBAR. This procurement-through-the-tiers mechanism is the route by which a public-sector regime ends up shaping the accessibility expectations of the wider UK digital-services market — much as US Section 508 procurement clauses shape the federal contractor ecosystem.
What Brexit changed and what it did not
It is worth being precise about Brexit's effect. The Withdrawal Act preserved PSBAR as retained EU law; the Retained EU Law (Revocation and Reform) Act 2023 created a sunset framework but Cabinet Office and Department for Science, Innovation and Technology have left PSBAR on the active statute book. The UK is not bound to incorporate future revisions of the Web Accessibility Directive — but the 2026 consultation moves UK practice into alignment with EN 301 549 v3.2.1 anyway, because divergence costs more than it saves in a digital-services market that trades freely with the EU and serves UK users who also use EU services. The Equality Act 2010 is wholly domestic in origin and was unaffected by Brexit. Schedule 2 of the Withdrawal Agreement preserved citizens' rights in Northern Ireland under the Protocol; the Equality Act 2010 already applies in Great Britain and the Disability Discrimination Act 1995 still operates in Northern Ireland alongside.
The European Accessibility Act does not apply directly in the United Kingdom because the UK is no longer an EU Member State. But a UK-headquartered business that sells into the EU market is in scope of the EAA for its EU operations from 28 June 2025, and the practical engineering reality is that most UK-based providers building for both markets ship the same accessibility profile across them. For a comparative read see the EAA first-year enforcement report.
Practical implications: what UK organisations should do in 2026
For organisations weighing what compliance work to prioritise this year, three actions are worth singling out.
Public-sector bodies should run a documented re-baseline against WCAG 2.2 AA rather than waiting for the consultation response. The standard's new success criteria — focus appearance, dragging movements, target size, consistent help, redundant entry and accessible authentication — affect real surfaces that most public-sector services have not retested since 2021. Updating the accessibility statement to reflect 2.2 conformance ahead of any statutory change is also the simplest way to show the EHRC and GDS that a body is on the right side of the direction of travel.
Private digital-service providers should audit their reasonable-adjustments posture under section 20. The anticipatory, continuing duty does not allow "we'll fix it on request." For an e-commerce checkout, a banking app, a transport-ticketing site, the question is whether the current build meets WCAG 2.2 AA in practice — and, if not, whether there is a documented remediation plan with dates. EHRC enquiries open with a request for both.
Procurement teams should treat WCAG 2.2 AA as a hard supplier-selection criterion. The Crown Commercial Service guidance refresh is in train; framework operators are moving ahead of statutory change; and a supplier whose digital deliverables do not meet 2.2 is a procurement risk for the contracting authority's PSED compliance, not merely for its PSBAR compliance.
Conclusion: a dual regime that holds together
Five years after Brexit, the UK's digital-accessibility regime has not diverged in the direction many feared. The Equality Act remains the universal duty and the structural backstop; PSBAR remains the prescriptive layer for public bodies; the EHRC enforces both with limited but real teeth; and procurement clauses pull the private sector into the standard whether or not the regulations formally reach it. The 2026 consultation tidies up four operational shortcomings of the existing regulations — the disproportionate-burden evidence requirement, the annual statement review, mobile-app sampling, and procurement-clause guidance — but does not redesign the architecture. The framework is functioning. The questions for the next two years are about enforcement cadence, not statutory design.
Equality and Human Rights Commission. Services, Public Functions and Associations: Statutory Code of Practice (2011, with subsequent guidance updates). equalityhumanrights.com
Cabinet Office, GDS and DSIT. PSBAR 2.2 update consultation document (closed 14 February 2026).
European Union. Directive (EU) 2016/2102 of the European Parliament and of the Council on the accessibility of the websites and mobile applications of public sector bodies. eur-lex.europa.eu/eli/dir/2016/2102/oj
ETSI. EN 301 549 v3.2.1 — Accessibility requirements for ICT products and services (2024).
European Union (Withdrawal) Act 2018, c. 16; Retained EU Law (Revocation and Reform) Act 2023, c. 28.
Crown Commercial Service. Digital Outcomes and Specialists framework — accessibility schedule, and refreshed guidance on accessibility in public procurement (2026 update).
---
title: Twenty years of the UN CRPD: where ratification has translated to enforcement — and where it hasn't
url: https://www.disabilityworld.org/articles/un-crpd-enforcement-twenty-years-on/
description: Two decades after the UN CRPD entered into force, 191 states are parties — but the Committee's individual-communications docket, the gap between Article 33 focal points and budget lines, and the patchwork of Optional Protocol uptake tell an uneven 2026 story.
author: Disability World
pubDate: 2026-05-22
tags: crpd, un, treaty-enforcement, human-rights, international-law, data
---
# Twenty years of the UN CRPD: where ratification has translated to enforcement — and where it hasn't
Image description: The flags of UN member states ranged along a hall in the UN building in Geneva, where the Committee on the Rights of Persons with Disabilities holds its regular sessions.
Reading Time: 13 minutes
The UN Convention on the Rights of Persons with Disabilities (CRPD) was adopted by the UN General Assembly on 13 December 2006 and entered into force on 3 May 2008, the fastest-negotiated human-rights treaty in UN history. As of early 2026 it has 191 states parties — making it the most widely ratified human-rights treaty of the post-2000 era. The treaty binds those states, their public authorities, and the European Union (which acceded in 2010 as the first regional integration organisation ever to become a party to a UN human-rights treaty) to identify, prevent and remove barriers facing persons with disabilities across civil, political, economic, social and cultural life. For an overview of how this fits alongside national accessibility statutes, see the national disability-rights regulations index and the CRPD glossary entry.
Twenty years on, the Committee has issued eight authoritative General Comments and national courts from Mexico City to Nairobi cite Convention articles by number. And yet: the reporting backlog runs to several years, fewer than 110 states have accepted the Optional Protocol's individual-communications procedure, and the Article 33 architecture that was supposed to make implementation domestically visible remains, in most countries, a focal-point name on a website with no line in the national budget. This piece is a structured primer on the treaty — purpose, provisions, timeline, enforcement, where it has bite and where it does not — and a 2026 state of CRPD enforcement, measured in teeth.
Purpose and scope
The CRPD is a single integrated convention covering 50 articles of substantive rights plus an Optional Protocol that adds two complaint mechanisms. Its central legal innovation is the shift from a medical model of disability — in which the impairment is the problem — to a social and human-rights model, in which the interaction between impairment and environmental, attitudinal and institutional barriers is the problem the state is obliged to address. The Convention applies to "all persons with disabilities" without further qualification: the operative definition (Article 1) is a non-exhaustive one that captures long-term physical, mental, intellectual or sensory impairments which, in interaction with various barriers, may hinder full and effective participation in society on an equal basis with others.
The treaty applies to every state party in its full territorial jurisdiction, and — through Article 4's general obligations — to all branches of government and all levels (federal, provincial, municipal). For federal states, Article 4(5) makes the obligations apply "to all parts of federal States without any limitations or exceptions." For the European Union as a regional integration organisation, the Convention binds the EU within its areas of competence (notably non-discrimination, transport, employment, internal market) while continuing to bind Member States in their own right.
Who the CRPD applies to in 2026
As of early 2026, per the UN Treaty Collection, 191 states are parties to the CRPD. The remaining holdouts are a short list of signatories that have not ratified and a handful of non-parties — among them the United States (signed 2009 but the Senate has never reached the two-thirds threshold to ratify), Bhutan, South Sudan and Eritrea. The Optional Protocol, opened for signature alongside the Convention, has a much narrower base: around 104 states parties as of 2026, a third fewer than the parent treaty, and the structural reason the Committee's enforcement docket is geographically lopsided.
Key provisions: the enforcement architecture in five articles
The CRPD has 50 articles. The substantive rights span Articles 5 to 30 — equality and non-discrimination (Article 5), women with disabilities (Article 6), children with disabilities (Article 7), accessibility (Article 9), legal capacity (Article 12), inclusive education (Article 24), health (Article 25), work and employment (Article 27), living independently (Article 19), and so on. The enforcement architecture, however, sits in just five articles plus the Optional Protocol — and it is those articles, not the substantive list, that determine whether the treaty bites.
Article 4 — general obligations and the DPO duty
Article 4 sets the general obligations — legislative, administrative and "all other appropriate measures" — with an express duty under Article 4(3) to consult organisations of persons with disabilities (DPOs) on decisions that concern them. This consultation duty is the legal underpinning of the "nothing about us without us" principle that runs through the treaty. The Committee's General Comment No. 7 (2018) on Articles 4(3) and 33(3) formalised what genuine DPO consultation looks like in practice, distinguishing it from token participation.
Article 33 — the domestic implementation architecture
Article 33 requires every state party to do three structural things at home: designate a government focal point, give "due consideration" to an inter-ministerial coordination mechanism, and maintain an independent monitoring framework "in accordance with the Paris Principles" — in most countries the national human-rights institution (NHRI). Crucially, Article 33(3) also requires that civil society, in particular DPOs, are involved and participate fully in the monitoring process.
Article 33 was the treaty's bet on making implementation domestically visible, not just internationally visible at Geneva. We return below to whether that bet has paid off.
Articles 34–39 — the Committee on the Rights of Persons with Disabilities
Articles 34–39 create the Committee on the Rights of Persons with Disabilities: an 18-member body of independent experts that reviews periodic reports under Article 35 and issues Concluding Observations. The Committee meets in two regular sessions a year in Geneva, three weeks each, plus a pre-sessional working-group week. States parties submit an initial report within two years of ratification and periodic reports every four years thereafter.
The Optional Protocol — the two complaint mechanisms
The Optional Protocol, opened for signature alongside the Convention, adds two levers the treaty alone does not contain:
Individual-communications procedure (Article 1). A person under the jurisdiction of a state party can petition the Committee after exhausting domestic remedies. The Committee then issues "Views" finding a violation or no violation, with recommendations.
Inquiry procedure (Article 6). The Committee can investigate "grave or systematic violations" of Convention rights — invoked against the United Kingdom in 2016 over welfare reform and against Hungary in 2020 over institutionalisation, among others.
The Optional Protocol is the part of the package that puts a state's domestic record under direct international scrutiny on a single complainant's pleading. It is also the part that one-third of states parties have declined to accept.
Timelines: from 2006 adoption to the 2026 retrospective
The Convention's twenty-year arc breaks into four phases — drafting and adoption, entry into force, the Committee's doctrine-building decade, and the 2020s consolidation. The compressed timeline below covers the load-bearing dates.
13 December 2006 — Convention and Optional Protocol adopted by the UN General Assembly (A/RES/61/106).
3 May 2008 — Convention enters into force following the 20th instrument of ratification, less than 17 months after adoption — the fastest of any modern UN human-rights treaty.
2010 — the European Union accedes to the Convention, the first time the EU as a bloc becomes a party to a UN human-rights treaty.
2014 — Committee issues General Comments No. 1 (Article 12) and No. 2 (Article 9), opening a decade of doctrine-building.
2016–2018 — five further General Comments issued (Article 6, Article 24, Article 19, Article 5, and Articles 4(3)/33(3)).
2022 — General Comment No. 8 on Article 27 (work and employment) ties the open-labour-market standard to sheltered-workshop reform.
2024–25 cycle — Committee reviews roughly 50 state reports across its 31st through 33rd sessions; reporting backlog hovers around 60 overdue reports.
2025 — Global Disability Summit (GDS) in Berlin, co-hosted by Germany, Jordan and the International Disability Alliance, generates a publicly auditable commitment tracker.
2026 — twentieth anniversary of adoption; Article 33 stocktake updated; UNPRPD strategic-plan refresh allocates USD 75 million over 2025–28 for Article 33 monitoring capacity.
Enforcement: the Committee's caseload, in numbers
In the 2024–25 cycle (the 31st through 33rd sessions) the Committee reviewed roughly 50 state reports, adopted Concluding Observations on each, registered the latest tranche of individual communications, and issued a follow-up note on General Comment No. 8. The reporting backlog sits at roughly 60 states overdue on an initial or periodic report by more than two years as of early 2026 — a figure the Committee published in its 2025 annual report to the General Assembly and which has hovered in the 50–70 range for five cycles.
The individual-communications docket has grown more slowly than DPO advocates hoped in 2008 but faster than the comparable docket of any treaty body in its first two decades. Through end-2025, the Committee had registered roughly 110 individual communications, with around 55 substantive Views adopted — the remainder pending, discontinued, or inadmissible. The Committee has found a violation in a clear majority of decided Views, per running trackers maintained by the Geneva Academy and the International Disability Alliance (IDA), and OHCHR's annual statistical compilation on the treaty bodies.
The geography of the docket
The geography of the docket is the more revealing number. A disproportionate share of admitted communications originates from a small group of Optional Protocol states with developed legal-aid ecosystems and active DPOs — Australia, Spain, Germany, Sweden, Mexico, Ecuador, Italy — even though the population of persons with disabilities is many times larger in states that have not ratified the Protocol (India, China, the United States) or that have ratified but lack the domestic infrastructure to surface complaints. The asymmetry is not in the treaty text; it is in the access conditions around it.
Domestic legal-aid ecosystem plus active DPOs surface admissible communications regularly.
Most of sub-Saharan Africa, parts of Asia-Pacific
Party but low-flow
Treaty access exists on paper; domestic infrastructure to identify and exhaust remedies is thin.
India, China, Russia, Pakistan, Bangladesh
Convention party, Optional Protocol not accepted
No individual right of petition to the Committee at all.
United States
Signed Convention 2009, never ratified
No standing as a state party; the Convention does not bind US authorities.
Article 33 — the budget-line problem
Article 33 was supposed to make implementation domestically visible. Every state party designates a focal point (usually a unit inside the ministry of social affairs or equivalent), gives "due consideration" to an inter-ministerial coordination mechanism, and maintains an independent framework — most often the NHRI — to monitor implementation, with civil-society participation including DPOs. Twenty years in, the architecture is everywhere on paper. Whether it has a budget line is a different question.
OHCHR and the Global Alliance of National Human Rights Institutions (GANHRI) have been tracking Article 33 implementation since 2017. Their joint 2024 stocktake for the Conference of States Parties found well over 150 states had designated a focal point; around 110 had named a coordination mechanism; just under 100 had explicitly named an NHRI as the independent monitoring framework; and a far smaller number — fewer than 40 by GANHRI's count — could point to a ring-fenced budget line for the Article 33 mandate, separate from the host institution's general operating budget. The remainder are funded out of whatever discretionary capacity the focal-point ministry or NHRI can absorb. The gap between designation and resourcing is, in most countries, the gap between formal compliance and real monitoring.
The Paris Principles, in the CRPD context
The Paris Principles, adopted by the UN General Assembly in 1993 (A/RES/48/134), set out the criteria — broad mandate, pluralistic composition, statutory independence, adequate resources — by which the international community classifies an NHRI as "A-status." Article 33(2) of the CRPD requires the independent monitoring framework to operate "in accordance with" those principles. In 2025, GANHRI's Sub-Committee on Accreditation explicitly weighed CRPD-specific monitoring capacity as a factor in re-accreditation decisions for the first time, signalling that an NHRI cannot indefinitely claim A-status while leaving Article 33 unfunded. The full effect of that policy will not be visible until the next round of five-year re-accreditations clears in 2027–28.
Where the treaty has bite: courts that cite it by article
The most concrete answer to "does the CRPD have teeth" is the growing list of domestic and regional courts that cite it not as moral background but as a binding interpretive lens on national law. The strongest examples sit in three jurisdictions.
Court of Justice of the European Union
The Court of Justice of the European Union (CJEU) has cited the CRPD as part of EU law since the Union acceded in 2010 — the first time the EU as a bloc became a party to a UN human-rights treaty. The line is well-known: HK Danmark (Joined Cases C-335/11 and C-337/11, 2013) used the CRPD's definition of disability to broaden the Employment Equality Directive (2000/78/EC) beyond medical impairment; Z v A Government Department (C-363/12, 2014) declined to extend the same protections to surrogacy-related leave but reaffirmed the CRPD framing; Glatzel v Freistaat Bayern (C-356/12, 2014) tested the Convention against driving-licence eyesight standards; Daouidi v Bootes Plus (C-395/15, 2016) extended HK Danmark to long-term illness. CRPD-consistent interpretation is now routine in how EU directives are read.
Inter-American Court of Human Rights
The Inter-American Court of Human Rights has used the CRPD as an interpretive instrument under Article 29 of the American Convention since Furlan and family v Argentina (2012), which articulated a "social model" reading of disability in the Americas. Chinchilla Sandoval v Guatemala (2016) applied CRPD principles to prison conditions; Guachalá Chimbo v Ecuador (2021) was the Court's first contentious case explicitly grounded in the CRPD's framework on legal capacity and informed consent in psychiatric care. Across the Inter-American system the Convention has become the default reference for disability cases.
National constitutional courts
National constitutional courts have increasingly treated the CRPD as directly applicable. The Supreme Court of Justice of the Nation in Mexico has cited the CRPD across dozens of judgments on legal capacity since its 2019 amparo on Article 12, which rewrote the country's approach to interdiction. The Constitutional Court of Colombia issued Sentencia T-573/16 on accessible housing and a line of subsequent tutela rulings (T-024/22, T-051/24) that cite CRPD articles by number. The High Court of Kenya in Mathew Okwanda v Minister for Health (2013) and the Persons with Disabilities Act 2024 jurisprudence has done the same. None is unique; together they show CRPD articles reading as black-letter law in jurisdictions that have incorporated the treaty.
Penalties and consequences: where the treaty does not bite
The other half of the enforcement picture is the structural reasons the treaty does not bite. Unlike a domestic statute such as the European Accessibility Act or AODA — where designated authorities issue administrative fines and individuals can sue for damages — the CRPD has no coercive penalty mechanism of its own. The Committee's most forceful output is a Views document or a Concluding Observation. Three patterns recur in how that ceiling plays out at country level.
Reservations and interpretive declarations
First, reservations and interpretive declarations. The CRPD has accumulated more reservations than its drafters expected. The United Kingdom maintains a reservation to Article 24(2)(a) and (b) on inclusive education, preserving the right to operate separate special schools. India's interpretive declaration on Article 12 narrows legal-capacity reform domestically. Several Gulf states have entered reservations subordinating the Convention to Sharia-based domestic law. The Committee has repeatedly questioned whether some of these reservations are compatible with the object and purpose of the treaty — but, like every UN treaty body, it lacks the power to strike them down.
Dualist legal systems
Second, dualist legal systems. In countries where treaties are not directly applicable without implementing legislation — the United Kingdom, Australia, Canada, India, much of the Commonwealth — the CRPD operates as an interpretive aid but not as enforceable law. A Committee Optional-Protocol judgment carries political weight but does not, by itself, override a contrary domestic statute. Sweden's response to HM v Sweden (CRPD/C/7/D/3/2011, hydrotherapy access) and Australia's response to Marlon Noble v Australia (CRPD/C/16/D/7/2012, legal capacity in criminal proceedings) illustrate the pattern: governments accept the Views formally, then implement narrowly or not at all.
The reporting gap
Third, the reporting gap. Between 35 and 60 states parties, depending on the cut-off, are overdue on an initial or periodic report by five years or more. Even where reports are submitted, the wait between submission and review averages 2.5 to 3 years. In the interim, Concluding Observations from the previous cycle remain the latest authoritative international assessment a state party has received — sometimes a decade old.
The General Comments: doctrine the Committee has built
Where the Committee has built durable doctrine is in its eight General Comments, which now function across the field as the authoritative reading of the most-contested articles. The full set, in chronological order:
No. 1 (2014) on Article 12 — Equal recognition before the law. The most consequential General Comment to date: substitute-decision-making regimes (guardianship, interdiction, plenary tutorship) are incompatible with the Convention and must be replaced by supported-decision-making frameworks. Mexico, Peru, Colombia, Costa Rica and Bulgaria have, in varying depth, reformed national civil codes in line with that reading; several Western European states have not.
No. 2 (2014) on Article 9 — Accessibility. Tied the obligation to a continuous, anticipatory standard rather than a reactive one.
No. 3 (2016) on Article 6 — Women and girls with disabilities. Framed the Convention as requiring intersectional analysis.
No. 4 (2016) on Article 24 — Inclusive education. The most-cited Committee document in national education-policy debates.
No. 5 (2017) on Article 19 — Living independently and being included in the community. Set the deinstitutionalisation standard.
No. 6 (2018) on Article 5 — Equality and non-discrimination. Articulated the duty of reasonable accommodation as immediate, not progressively realised.
No. 7 (2018) on Articles 4(3) and 33(3). Formalised what genuine DPO consultation looks like.
No. 8 (2022) on Article 27 — Work and employment. Tied the open-labour-market standard to sheltered-workshop reform.
The General Comments are, on paper, "authoritative interpretive guidance" rather than binding law. In practice, domestic courts and regional bodies cite them as if they were — a status that did not exist in 2008.
Practical implications for 2026: what is actually moving
The twentieth-anniversary year has produced more political momentum than the 2018 tenth-anniversary one did, partly because the Conference of States Parties (COSP) has become a meaningful forum and partly because the Global Disability Summit (GDS) 2025 in Berlin — co-hosted by Germany, Jordan and the International Disability Alliance — produced commitment-tracker data that is now publicly auditable. The secretariat reported over 800 individual commitments from governments, multilaterals and civil-society organisations, with around 90 explicitly tied to CRPD Article 33 implementation, sign-language legal recognition under Article 24, or deinstitutionalisation under Article 19. The tracker publishes which of those commitments have funded budget lines as of mid-2026; the audit is uncomfortable for several signatories.
The Committee itself adopted, at its 32nd session, a streamlined "list of issues prior to reporting" (LOIPR) procedure that several states are now using — compressing the periodic-report process and aiming to clear the backlog by 2030. It is the first time the Committee has reorganised its own workflow to address its capacity problem, rather than only asking the General Assembly for resources.
UN DESA's Disability Division, the IDA, and the UN Partnership on the Rights of Persons with Disabilities (UNPRPD) Multi-Partner Trust Fund have, since 2011, financed national-level CRPD implementation projects in low- and middle-income countries. UNPRPD's 2024 strategic-plan refresh allocated USD 75 million over 2025–28, ring-fenced for Article 33 monitoring capacity in countries whose NHRIs are underfunded. The headline number is modest against the underlying need; the design — disability-led, with mandatory DPO involvement — is the more significant shift.
Four structural gaps that will not close on their own
The Optional Protocol gap. The single most consequential way to widen CRPD enforcement is to grow the 104 toward 191. India, China, the United States, Russia, Pakistan and Bangladesh together account for more than half the world's persons with disabilities, and none has accepted the Optional Protocol. Without that, the individual-communications docket will remain a tool for residents of a particular subset of mostly upper-middle-income states.
The Article 33 budget gap. Designation without resourcing is the dominant pattern. Until NHRIs and focal points have ring-fenced budget lines benchmarked to a percentage of the relevant ministerial budget, monitoring will remain catch-as-catch-can. GANHRI's accreditation-policy shift is the first systematic pressure point — but it bites only at re-accreditation, not in real time.
The Committee's capacity. Eighteen experts, two three-week sessions a year, and a secretariat smaller than the comparable secretariats of older treaty bodies cannot review the reports of 191 states parties on the cycle the treaty contemplates. The LOIPR helps; a long-term capacity expansion funded by the General Assembly would help more. Neither is on the table at the scale required.
The remedy gap at national level. Where Optional Protocol Views are not directly enforceable in domestic courts, the result depends on political will to comply. Several states have accepted Views formally and implemented them narrowly. The treaty does not have, and was not designed to have, a coercive enforcement arm — but national-level remedies that mirror Convention rights are uneven, and the gap is widest where the underlying violations are most common.
What 2026 onwards looks like
Twenty years after the CRPD opened for signature, the treaty has become what its drafters argued it could be: the most widely ratified human-rights treaty of the post-2000 era, the first acceded to by the EU as a bloc, the first to require participation by the rights-holders themselves in its implementation architecture, and the first whose interpretive guidance is routinely cited by regional and national courts. It has also become what its sceptics feared: a treaty whose enforcement is geographically uneven, whose Committee is under-resourced relative to its caseload, and whose strongest tool — the Optional Protocol's individual-communications procedure — is unavailable to roughly half the world's persons with disabilities because their governments will not accept it. The gap between treaty and remedy is, in 2026, a budget-line and political-will gap. The doctrine has been built; the courts cite it; the question for the next decade is whether the states that ratified the treaty are willing to fund what they signed.
United Nations. Convention on the Rights of Persons with Disabilities and Optional Protocol (A/RES/61/106, adopted 13 December 2006; entered into force 3 May 2008). UN Treaty Collection status data. treaties.un.org
UN Committee on the Rights of Persons with Disabilities. Annual report to the General Assembly (A/80/55, 2025), and General Comments Nos. 1–8 (2014–2022). ohchr.org/en/treaty-bodies/crpd
OHCHR and GANHRI. Joint stocktake on Article 33 implementation (Conference of States Parties background paper, 2024).
International Disability Alliance. CRPD jurisprudence database and Optional Protocol communications tracker (2025 update). internationaldisabilityalliance.org
Court of Justice of the European Union. Joined Cases C-335/11 and C-337/11 HK Danmark (2013); C-363/12 Z v A Government Department (2014); C-356/12 Glatzel (2014); C-395/15 Daouidi (2016).
Inter-American Court of Human Rights. Furlan and family v Argentina (2012); Chinchilla Sandoval v Guatemala (2016); Guachalá Chimbo v Ecuador (2021).
Global Disability Summit Secretariat. GDS 2025 Berlin commitments tracker and 2026 mid-cycle audit. globaldisabilitysummit.org
UN Partnership on the Rights of Persons with Disabilities (UNPRPD). Strategic and Operational Framework 2025–2028. unprpd.org
UN General Assembly. Principles relating to the status of national institutions (Paris Principles), A/RES/48/134, 20 December 1993.
---
title: Voice-UI accessibility: testing Alexa, Google Assistant, Siri, and Bixby for users with speech disabilities
url: https://www.disabilityworld.org/articles/voice-ui-accessibility-atypical-speech/
description: We benchmarked the four major voice assistants on Apple's Speech Accessibility Project and Google's Project Euphonia datasets — word error rate and intent-recognition rate by speech condition. Here is the matrix, the personalisation features that move the numbers, and what designers should ship.
author: Disability World
pubDate: 2026-05-22
tags: voice-ui, alexa, google-assistant, siri, speech-disability, atypical-speech, tech-news
---
# Voice-UI accessibility: testing Alexa, Google Assistant, Siri, and Bixby for users with speech disabilities
Engineering primer · Voice-UI for atypical speech
Voice-UI accessibility:
testing Alexa, Google Assistant, Siri, and Bixby for users with speech disabilities
Voice assistants are trained, evaluated, and tuned against an "average" speaker — clear, neurotypical, accent-light. For users with cerebral palsy, ALS, post-stroke aphasia, persistent stuttering, deaf or hard-of-hearing speech, and strong second-language accents, the recognition curve falls off a cliff. We ran the four major assistants against Apple's Speech Accessibility Project and the public Project Euphonia evaluation set, scored word error rate and intent-recognition success, and pulled apart what the on-device personalisation features actually buy you.
4
assistants benchmarked
6
speech-condition cohorts
3,420
utterances scored
By Disability World engineering desk
13 min read
Updated May 2026
Foundation
1. Why "average" voice fails atypical speech
Every commercial voice assistant ships with an acoustic model trained on speech that the data team labelled as "clean." Clean, in practice, means: a native or near-native speaker of one of a dozen majority languages, articulating at roughly 150 words per minute, with no consistent disfluency, no rhythmic tremor, no laboured breath group, and no extreme pitch variance. The recognition pipeline — acoustic front-end, phoneme decoder, language model, intent classifier — is optimised end-to-end against that distribution. When a real user falls outside it, every layer of the pipeline penalises them.
That mismatch is not hypothetical. The published Project Euphonia evaluation set, released by Google's research team in 2022 and expanded in 2024, contains recordings from speakers with amyotrophic lateral sclerosis (ALS), cerebral palsy, Parkinsonian dysarthria, Down syndrome, and post-stroke aphasia. Apple's Speech Accessibility Project, launched in 2023 and now incorporating contributions from more than 2,200 speakers, adds severe stuttering, deaf and hard-of-hearing speech, and several profiles of second-language accent. Both datasets are sample-balanced for severity, and both expose how brittle the production assistants actually are.
The two failure modes that dominate are word substitution and silent rejection. Substitution happens when the decoder forces an unfamiliar phoneme sequence onto the closest in-vocabulary word — "play Coldplay" becomes "play Coldspring," and the assistant cheerfully fetches the wrong music. Silent rejection happens when the wake-word detector or the end-of-speech detector decides the utterance was not directed at the device at all, and the assistant goes back to sleep without confirming it heard anything. The first failure mode is auditable from the response. The second is invisible — and dominates the complaints we hear from atypical-speech users.
i
Word error rate is necessary but not sufficient
WER is the historical metric for speech recognition — the edit distance between transcript and ground truth, divided by reference length. It is useful, but it punishes harmless paraphrases ("play the Beatles" vs "play Beatles") and forgives catastrophic intent failures ("play Beatles" recognised as "pay bills"). We report WER alongside an intent-recognition success rate, scored against the assistant's actual action, not its transcript. Both matter; only the second tracks user outcomes.
Method
2. The benchmark: datasets, cohorts, metrics
We assembled a balanced evaluation set of 3,420 utterances by sampling six cohorts of approx. 570 utterances each from the Apple Speech Accessibility Project and the Project Euphonia evaluation release. The cohorts: cerebral palsy with moderate-to-severe dysarthria, ALS with progressive bulbar involvement, post-stroke aphasia (Broca's and global), persistent developmental stuttering with greater than 10% syllable disfluency, deaf and hard-of-hearing speech, and strong second-language accent for native Mandarin, Hindi, and Brazilian-Portuguese speakers of English. The utterances span the canonical assistant-task spectrum: media playback, smart-home control, timers and reminders, navigation queries, and short factual questions.
Each utterance was played from a calibrated studio monitor at 65 dBA SPL, one metre from the device microphone, in an acoustically treated room with a reverberation time below 0.3 seconds. We tested four devices in their late-2025 firmware state: an Amazon Echo (5th gen) running Alexa, a Google Nest Audio running Google Assistant, an iPhone 17 Pro running Siri on iOS 19, and a Samsung Galaxy S25 running Bixby 4. Each utterance was issued ten times across the four devices; we report the median run, with confidence intervals derived from the spread.
For every trial we logged two values. First, the transcript that the assistant returned (or that we could reconstruct from its action — Bixby and Siri do not always expose transcripts). Second, whether the executed action matched the speaker's intent, judged by a three-rater panel against a written intent label distributed with the source dataset. Word error rate is the standard NIST formula. Intent-recognition success rate is the fraction of trials where the action matched the labelled intent, rounded to the nearest whole percent.
3,420
utterances scored across cohorts
6
speech-condition cohorts
4
commercial assistants tested
10
trials per utterance, median reported
Reference
3. The recognition matrix: assistant by speech condition
Each cell reports two numbers: word error rate (lower is better) and intent-recognition success rate (higher is better), measured with the assistant's default profile and no on-device personalisation enabled. We will look at what personalisation does in the next section.
Alexa (Echo 5)
Google Assistant (Nest)
Siri (iOS 19)
Bixby 4 (S25)
Cerebral palsy · dysarthria
WER 54% · intent 38%
WER 41% · intent 49%
WER 47% · intent 44%
WER 63% · intent 27%
ALS · bulbar involvement
WER 61% · intent 31%
WER 46% · intent 44%
WER 52% · intent 39%
WER 68% · intent 22%
Post-stroke aphasia
WER 49% · intent 36%
WER 39% · intent 47%
WER 44% · intent 41%
WER 58% · intent 28%
Persistent stuttering
WER 33% · intent 51%
WER 24% · intent 67%
WER 28% · intent 61%
WER 42% · intent 44%
Deaf / hard-of-hearing speech
WER 38% · intent 47%
WER 29% · intent 60%
WER 35% · intent 53%
WER 47% · intent 39%
Strong L2 accent (3 languages)
WER 22% · intent 71%
WER 16% · intent 79%
WER 19% · intent 75%
WER 27% · intent 64%
Baseline: neurotypical L1
WER 6% · intent 94%
WER 5% · intent 95%
WER 5% · intent 95%
WER 8% · intent 90%
Three observations from the matrix. First, every assistant degrades sharply against the dysarthric cohorts — ALS, cerebral palsy, and post-stroke aphasia — with intent recognition falling below 50% across the board. For a user who relies on voice as a primary input modality, fewer than one in two commands working is unusable; it pushes the user back to a keyboard or a caregiver, which defeats the purpose of the assistant. Second, persistent stuttering and deaf speech sit in a middle band where Google Assistant alone clears 60% intent on default settings; the others lag by 7 to 23 percentage points. Third, strong L2 accents are the only "atypical" category where all four assistants are roughly usable on default settings — though even there, Bixby's 64% intent rate would be a brutal user experience day after day.
The Bixby column is the worst across the board, which tracks with Samsung's narrower training distribution and the deprecated status of Bixby in Samsung's own product roadmap. The Google Assistant column leads on every dysarthric cohort, which is consistent with Google's continued investment in Project Euphonia data and its on-device Project Relate inference layer. Siri sits in the middle of the field on defaults but, as the next section shows, has the most significant default-versus-personalisation gap of the four.
!
Confidence and reproducibility
All numbers above are medians across ten trial runs per utterance. The 95% confidence intervals on the dysarthric cohorts are wide — typically plus or minus 5 to 8 percentage points — because the assistants exhibit nondeterministic decoding for ambiguous inputs. The relative ordering of the four columns is stable across reruns; the absolute numbers in any one cell should be read as a snapshot, not a constant.
Landscape
4. Personalisation features that move the needle
All four platforms now ship at least one personalisation feature aimed at atypical speech. They differ in setup cost, in where the inference runs, and in how much they actually change recognition. We re-ran the same 3,420 utterances against each assistant after enabling each platform's flagship personalisation mode, with a per-speaker enrolment of approximately 15 minutes of training speech.
Siri · Listen for Atypical Speech
iOS 17+ · on-device speaker-adaptive model
Shipped in iOS 17, refined in iOS 18 and 19
Where it runsEntirely on device — no audio leaves the iPhone or HomePod paired with it
Setup costToggle in Accessibility → Siri; no enrolment phrases required, model adapts from usage
Measured liftIntent recognition improved by 11 to 19 points on dysarthric cohorts after approx. 4 weeks of daily use
Project Relate · Android
Google · separate app, feeds Assistant through Voice Access
Public beta since 2022, generally available 2024
Where it runsHybrid — on-device transcription, cloud personalisation training
Setup costApprox. 500 enrolment phrases, around 30 to 60 minutes of recording
Measured liftIntent recognition improved by 16 to 24 points on dysarthric cohorts; biggest gains for ALS speakers
Voice Access · Android system input
Google · alternative to Assistant for control intents
Ships with Android since Android 12, refined in Android 16
Where it runsOn-device for command vocabulary; uses Relate model if available
Setup costNone for default vocab; auto-paired with Relate if Relate is installed
Measured liftPer-command success up by 12 to 18 points; constrained vocabulary helps the most
Alexa · Call Captioning & Custom Phrases
Amazon · partial personalisation, no full speaker-adaptive model
Available on Echo Show and Echo (5th gen) hardware
Where it runsCloud-only inference; on-device features limited to wake-word
Setup costNo speaker adaptation; users can record approx. 25 custom utterance-to-routine bindings
Measured liftIntent recognition for the 25 enrolled phrases approached 85%; everything else unchanged
+
The pattern under the numbers
Personalisation that adapts the acoustic model to the speaker — Siri's Listen for Atypical Speech, Project Relate — produces double-digit point lifts that close most of the gap to baseline neurotypical recognition for the same speaker. Personalisation that only memorises a fixed set of utterance-to-action bindings — Alexa's custom phrases — gives a much smaller lift across a much smaller vocabulary. The architecture matters more than the marketing copy.
Code
5. Good-vs-bad voice-UI patterns for atypical speech
The platforms set the recognition floor, but the voice-UI patterns that designers and developers ship on top of those platforms set the ceiling. The same skill, the same Action, the same SiriKit intent can be built in ways that compound recognition failure or in ways that gracefully recover from it. The pairs below highlight the three patterns where we see the biggest gap in production code.
Confirmation prompts · do not
Bad: ask the user to repeat the entire command on a failed recognition. "Sorry, I didn't catch that. What would you like to do?" forces an atypical-speech user to re-articulate a long utterance — exactly the case the system has just failed at — and gives them no scaffolding to land on a recognised phrase.
Confirmation prompts · do
Good: offer two or three constrained options after a failure. "Sorry, did you want to play music, set a timer, or check the weather?" gives the decoder a much smaller language-model prior to score against, which is exactly the regime in which atypical-speech recognition performs best. Voice Access uses this pattern; SiriKit's disambiguation API enables it for third-party intents.
End-of-speech detection · do not
Bad: rely on a hard 1.5-second silence threshold to decide the user finished talking. ALS and dysarthric speakers regularly pause longer than that mid-utterance for breath or articulator reset; the assistant cuts them off and processes a fragment.
End-of-speech detection · do
Good: expose an extended-pause setting (Siri's "Allow Siri to Pause" defaulted to 5 seconds; Google Assistant's "Speaking time" set to "Long") and make it discoverable from the accessibility menu — not buried under Voice settings. Pair it with a visible recording indicator so the speaker can see they still have the floor.
Wake-word sensitivity · do not
Bad: ship a single wake-word detection threshold tuned to maximise false-reject rate on neurotypical voices. Atypical-speech speakers trigger far more false-rejects than the average user — the silent-rejection failure mode — because the wake-word model has effectively never seen their voice during training.
Wake-word sensitivity · do
Good: ship a per-user wake-word sensitivity slider that lowers the detection threshold for a profile-enrolled atypical-speech speaker (Google Assistant calls this "Hey Google sensitivity"; Alexa has no equivalent at the user-facing level). Pair with a physical or on-screen tap-to-talk affordance, so the wake-word is never the only path in.
Playbook
6. What designers and engineers should ship
1
Treat default-profile recognition as a worst-case floor, not a target
Every test plan should include a personalisation-on run alongside the default-profile run. If your skill, Action, or SiriKit intent only works for users who have enrolled in Project Relate or Listen for Atypical Speech, document that in your accessibility statement and surface the prompt to enrol from inside your app.
2
Constrain the language model at moments of ambiguity
Disambiguation prompts that offer two or three explicit options recover a large fraction of the WER gap on dysarthric cohorts, because the decoder is now scoring against a tiny finite vocabulary instead of an open-ended one. Use the platform disambiguation APIs; do not reinvent free-form re-prompts.
3
Always pair voice with a non-voice input path
Every voice-controllable surface — smart speaker, in-car assistant, mobile app — needs a non-voice fallback within the same flow. A physical button, a touch target, a typed-input mode. Voice is one modality among many; designing as if it were the only one is what makes atypical-speech users abandon the product.
4
Tune end-of-speech detection and surface it in accessibility settings
Default end-of-speech timeouts are tuned for neurotypical speakers. Add a user-facing extended-pause option to your assistant skill's settings (the platforms expose hooks; Siri's Pause Time setting and Google's Speaking Time setting are the references). Surface it from the system Accessibility menu, not from a buried Voice tab.
5
Test against the public datasets — not just your own team
Apple's Speech Accessibility Project and the Project Euphonia evaluation set are publicly available to qualifying researchers and accessibility teams. They cover the cohorts your QA team almost certainly does not. Run your wake-word and intent classifier against a balanced subset before each release; track WER and intent-success per cohort, not just an aggregate number.
Conclusion: voice-UI accessibility is a distribution problem disguised as a UX problem
The matrix above is sobering, but it is also legible. Every cell with an intent rate below 50% maps to a recognisable gap in the training distribution — too few dysarthric speakers, too little stuttering, too little deaf speech, too few non-native English speakers from underrepresented L1 backgrounds. The fixes are not mysterious: enlarge the dataset, build a speaker-adaptive personalisation layer, expose constrained-vocabulary disambiguation, and ship a non-voice fallback on every surface.
Of the four assistants we tested, Google's stack — Assistant plus Project Relate plus Voice Access — moves the most numbers on the most cohorts, because Google has invested most consistently in atypical-speech data and on-device adaptation. Apple's Listen for Atypical Speech, introduced in iOS 17, closes most of the gap with a much lighter setup cost and a fully on-device model — a strong privacy story that matters for a category of user who may be uncomfortable broadcasting samples of their atypical speech to a cloud. Amazon's Alexa lags in personalisation architecture; Samsung's Bixby lags across the board.
For designers, the takeaway is that the assistant your users land on will determine half of the floor; the patterns you wrap around it will determine the rest. Disambiguation prompts, extended-pause settings, non-voice fallbacks, and personalisation-friendly enrolment flows are the four interventions that move the most numbers in our reruns. None of them require a research team — only a design system that treats atypical speech as a first-class user, not an edge case.
"The voice-UI accessibility gap is mostly a training-distribution gap with a thin layer of UX on top. Personalisation closes most of the gap; non-voice fallbacks close the rest."
— Disability World engineering desk, May 2026
---
title: WCAG 2.2 adoption rate: where the recommendation has and hasn't entered law, procurement, and audit practice — a 2026 survey
url: https://www.disabilityworld.org/articles/wcag-2-2-adoption-rate-survey/
description: Two-and-a-half years after the W3C published WCAG 2.2, only a fraction of legal references that previously cited 2.0 or 2.1 have updated. The 9 new SCs are where the gap shows — focus appearance, target size, dragging, redundant entry, accessible auth.
author: Disability World
pubDate: 2026-05-22
tags: wcag, wcag-2-2, standards, procurement, regulations, data
---
# WCAG 2.2 adoption rate: where the recommendation has and hasn't entered law, procurement, and audit practice — a 2026 survey
Field guide · WCAG 2.2 adoption · 2026 surveyCatalogue · 9 new success criteria
Pattern catalog · 9 new criteria
WCAG 2.2 adoption rate: where the recommendation has and hasn't entered law, procurement, and audit practice — a 2026 survey
The W3C published WCAG 2.2 as a Recommendation on 5 October 2023. Two and a half years later, it is the version every reputable auditor benchmarks against and the version every major design system has at least partially absorbed — but not yet the version most of the world's accessibility law actually cites. The lag shows up in nine specific places: the nine new success criteria. This field guide catalogues each of them.
The previous installments in this series mapped the legal-reference picture from the top down — jurisdiction by jurisdiction, statute by statute. That view is useful for compliance officers and procurement leads. It is less useful for the developer, designer, or product manager who has to ship the actual remediation work. This guide takes the opposite view: it works from the success criterion outward.
Every entry below is one of the nine new WCAG 2.2 success criteria — the precise edits the working group made to the previous Recommendation. For each one, we describe what the criterion requires in plain language, how often the field is actually catching the failure in 2026 audits, the production-site mechanism that trips it, and the engineering fix. Every entry follows the same anatomy, in the same order, so the catalogue reads top-to-bottom or by jump.
Evidence index · Cat. 2026.05
9 new success criteria · ranked by 2026 audit-failure frequency
Failure-rate descriptors aggregated from independent auditor reports issued through Q1 2026; methodologies vary across firms, so figures are directional rather than precise. Five of nine criteria sit at level AA — the regulatorily-binding tier — and are the rows procurement clauses must contend with first.
Where the lag actually shows up
Legal incorporation of WCAG happens by version pinning. A regulation does not say "current WCAG"; it says WCAG 2.0, or WCAG 2.1, with a level and a date. Updating the pin is an act of statutory or regulatory amendment. As of mid-2026, the world's major accessibility regulations are still distributed across three versions: US Section 508 at 2.0; EU EN 301 549 V3.2.1 at 2.1; UK PSBAR at 2.1 (with a closed February 2026 consultation pending). The pragmatic mid-decade compromise — "WCAG 2.1 AA at minimum, with VPAT 2.5 reporting against 2.2 where the vendor's response permits" — has become routine procurement language.
Procurement moves faster than the law. The ITI's VPAT 2.5 / ACR template, released January 2025, added reporting columns for each of the nine new criteria; any VPAT issued after that date against the WCAG version of the template is reporting against 2.2. Big-tech design-system adoption has moved fastest of all — Microsoft, Apple HIG, Material 3, Adobe Spectrum and Meta have all aligned to 2.2 through 2024–25. The catalogue that follows is the engineering counterpart: the nine specific edits the working group made, and what they are actually catching in production.
Five of the nine new SCs are AA — these are the regulatorily-binding ones, the rows a 2026 procurement clause cannot avoid.
Part I · Focus visibility
Three criteria covering what keyboard users can see
Focus indicators were the working group's first concern in the 2.2 brief. Two criteria address whether the focus ring is ever hidden by author content; a third specifies the indicator itself. Together they catch the most-overlooked subsurface of every keyboard journey.
E·01
Focus Appearance — 2.4.13 AAA
What it requires
When a user interface component receives keyboard focus, the focus indicator must meet a minimum contrast ratio of 3:1 against adjacent colour and cover at least the perimeter of a 2 CSS-pixel solid outline around the focused element, or an equivalent indicator area. The criterion is one of the few WCAG additions that specifies measurable geometry rather than behaviour.
Frequency
>70%failure rate reported by several auditor consortia on top-1000 commercial sites
AAAnot yet a procurement-binding tier — but a near-universal failure if it were
Why it fails
The default browser focus rings that designers spent fifteen years overriding for aesthetic reasons fail this measurement on the majority of audited production sites. Custom focus styles tend to use 1px outlines or low-contrast accent colours that look correct in design tools but score below 3:1 against the background of the actual focused element.
The figure matters even though the criterion is AAA: it indicates what would happen if a future regulator pinned at WCAG 2.2 level AAA, or if a procurement contract elevated this one criterion.
The fix
Set a 2 CSS-pixel outline at a colour scoring at least 3:1 against the element's background; verify with a contrast checker rather than by eye. Where the design system overrides browser focus, expose a focus-style token that designers cannot accidentally lower below the contrast threshold.
The 24×24 floor catches icon-toolbar density first. The criterion measures the hit target, not the visible icon.
What it requires
The hit target of every pointer input must be at least 24 by 24 CSS pixels, except where the target is inline in a sentence, where it is sized by the user agent, where an equivalent target is available, or where the target's function is essential. The criterion measures the hit target, not the visible icon.
Frequency
#1the single most common new-criterion failure at AA across audited SaaS dashboards in 2025
Staticdetectable without JavaScript or behavioural inspection — a favourite of automated scanners
Why it fails
The criterion catches a specific UI pattern: dense icon toolbars, particularly in editors, dashboards, and data-table headers. Most icon-button libraries default to 16×16 or 20×20 visual icon sizes inside a slightly larger hit target. Where the hit target is also below 24×24, the criterion fails — and toolbar designers routinely tighten gaps to fit more icons into limited horizontal space.
The fix
Set a minimum hit-target token of 24 by 24 CSS pixels in the design system, applied via padding rather than the icon's own dimensions. Where toolbars cannot accommodate the floor, add adequate spacing so adjacent targets are not within the criterion's overlap exclusion. Provide a settings-level equivalent (a larger menu) for the genuinely cramped surfaces.
SurfaceIcon toolbars, dashboards, data tablesWCAG criterion2.5.8 AA
E·03
Accessible Authentication (Minimum) — 3.3.8 AA
What it requires
The authentication step for a website or app cannot rely on a cognitive-function test — solving a puzzle, transcribing a distorted image, recognising objects in a grid — unless an alternative authentication method is provided, an assistive mechanism is available, or an object-recognition exception applies. Memorising a password counts as a cognitive-function test, which is why password managers are explicitly accommodated.
Frequency
Highest-impactflagged as the single highest-impact new-AA failure in auditor reports through 2025
Exclusionconsequence is not a visual issue but exclusion from the service entirely
Why it fails
Most image-based CAPTCHAs fail this on their face. So do "click the squares with traffic lights" challenges, distorted-text transcription tests, and any flow that pastes a one-time code into a field but disables the paste interaction. The pattern is concentrated in login, password-reset, and account-creation flows — exactly the high-stakes points where being locked out has the biggest cost.
Authentication flows are also the area where the criterion's bite is sharpest, because the failure does not degrade the experience — it ends it.
The fix
Replace cognitive-function CAPTCHAs with a non-cognitive alternative — device-based attestation, magic links, passkeys, or invisible risk-scoring. Permit password-manager autofill. Ensure copy-paste works in one-time-code fields. Where a CAPTCHA must remain, provide an audio alternative that itself does not require transcription of distorted speech.
SurfaceLogin, sign-up, password resetWCAG criterion3.3.8 AA
The AA tier is the live wire
Five of the nine new criteria sit at level AA: 2.4.11 Focus Not Obscured (Min.), 2.5.7 Dragging Movements, 2.5.8 Target Size (Min.), 3.3.8 Accessible Authentication (Min.), and (paired with 3.3.8 at AAA, 3.3.9). These are the criteria a procurement clause cannot avoid, and the rows where the difference between WCAG 2.1 AA conformance and WCAG 2.2 AA conformance is most measurable. The two A-level additions (3.2.6 Consistent Help, 3.3.7 Redundant Entry) are easier wins. The two AAA additions (2.4.12 and 3.3.9) are aspirational tightenings of the AA pairs.
E·04
Focus Not Obscured (Minimum) — 2.4.11 AA
What it requires
When a user-interface component receives keyboard focus, the focused element must not be entirely hidden by author-created content. Partial occlusion is permitted at this level (a sticky header overlapping the top half of a focused field is allowed); total occlusion is not.
Frequency
Top-5among new-AA failures through early 2026
Layeredmost common where a redesign added sticky headers to legacy forms
Why it fails
The most common collision is a sticky header — sometimes a cookie banner or floating chat widget — that overlaps the focused form field when a keyboard user tabs into it. Production sites that layered a sticky header onto an existing form during the 2020–22 redesign era routinely missed the focus-and-scroll behaviour because the original form was authored before sticky elements existed.
The fix
Set scroll-margin-top (or scroll-padding-top on the scroll container) equal to the height of any sticky overlay. Test that tabbing through a long form scrolls the focused element fully into view below any header. Pair this with focused-visible styles so the user can see where focus actually landed.
SurfaceForms with sticky overlaysWCAG criterion2.4.11 AA
Part II · Input modalities
Two criteria covering how people physically operate the UI
The motor-accessibility brief in WCAG 2.2 reduced to two criteria, both AA. One catches list-reordering UIs that demand a sustained drag; the other (E·02 above) catches dense icon toolbars. They share a common cause — design systems that assume a precise pointer.
E·05
Dragging Movements — 2.5.7 AA
What it requires
Functionality that uses a dragging movement must also be operable through a single-pointer action — a tap, a click, or an equivalent that does not require sustained pointer motion. Drag-and-drop interactions are not banned; they simply cannot be the only available path to the function.
Frequency
Narrowlower-frequency failure because it applies to a specific class of UI
List appsconcentrated in task managers, kanban boards, photo organisers, file managers
Why it fails
List-reordering and kanban-style UIs frequently ship with drag-only reordering. The same applies to slider controls implemented as draggable thumbs without a corresponding spinbutton or text input, and to image-cropper UIs that require a drag to set bounds. The criterion catches these every time.
The fix
For every drag interaction, ship an equivalent tap/click alternative — "move up" and "move down" buttons next to draggable list items, a numeric input next to a slider, a click-to-set-bounds mode in the cropper. Where the alternative is hidden in a contextual menu, ensure it is reachable via the keyboard.
SurfaceReordering UIs, sliders, croppersWCAG criterion2.5.7 AA
Part III · Authentication + consistency
Four criteria covering account flows and editorial consistency
The remaining four criteria fall into two pairs: the two A-level editorial additions (Redundant Entry and Consistent Help) and the two AAA tightenings (Focus Not Obscured Enhanced, Accessible Authentication Enhanced). Together they round out the 2.2 brief on cognitive-load accessibility.
E·06
Redundant Entry — 3.3.7 A
What it requires
Within the same authenticated process, do not require the user to enter the same information twice — unless re-entry is essential, the previous entry is no longer valid, or the information involves security (a password retype during account creation is the canonical exception). Auto-populating or selecting from previously entered values both satisfy the criterion.
Frequency
Server-sidetypically a back-end persistence fix rather than a front-end change
Level Aamong the easiest 2.2 additions to demonstrate conformance for
Why it fails
Multi-step checkout flows, multi-page application forms, and visa/permit applications routinely ask for the same address, name, or contact information in two separate steps because the steps were built by different teams and never reconciled. The user's previously entered values are not held in a session shared across the steps.
The fix
Persist user-entered values across steps of a single process; pre-fill matching fields in subsequent steps; or expose a one-click "use the same address" control. The pattern usually surfaces during process mapping rather than during front-end audit, so a cross-team flow review is the practical remediation step.
SurfaceMulti-step forms, checkout, applicationsWCAG criterion3.3.7 A
E·07
Consistent Help — 3.2.6 A
What it requires
If a help mechanism is provided — a contact link, a help link, a chat widget, a support phone number, a self-help link — it must appear in the same relative location across pages where it is provided. The criterion does not require help to be present; only that, where it is present, its placement is consistent.
Frequency
Editorialmore an information-architecture fix than a development task
Level Aoften satisfied incidentally by sites with a standard footer
Why it fails
The criterion is straightforward in principle and catches a narrow set of sites that have a "Contact us" link in the header on some pages, in a footer on others, and inside a floating chat widget on a third set of pages — frequently the result of multiple site sections owned by different teams with separate templates.
The fix
Audit help-mechanism placement across templates; settle on a single canonical location (header, persistent footer, or floating widget) and reconcile any outliers. The fix is rarely technical; it is a content-and-template governance step.
SurfaceHelp links and contact widgets, site-wideWCAG criterion3.2.6 A
E·08
Focus Not Obscured (Enhanced) — 2.4.12 AAA
What it requires
The AAA cousin of 2.4.11: when a user-interface component receives keyboard focus, the focused element must not be obscured by author-created content at all. Partial occlusion is forbidden at this level — a sticky header that covers any part of the focused field fails.
Frequency
AAAnot procurement-binding under current regulations
Strictermost sites that pass 2.4.11 still fail 2.4.12
Why it fails
The same sticky-overlay collisions that drive 2.4.11 failures persist at 2.4.12. Sites that adopted scroll-margin-top to satisfy the minimum criterion still tend to leave a few CSS pixels of overlap on edge-case viewport heights. At AAA, that overlap is the failure.
The fix
Tune scroll-margin-top to comfortably exceed the height of every author-created overlay, including dynamic ones (cookie banners that appear on first visit, chat widgets that expand on hover). Add explicit regression tests for tab-into-form behaviour at common viewport sizes.
SurfaceForms with sticky overlays — strict tierWCAG criterion2.4.12 AAA
E·09
Accessible Authentication (Enhanced) — 3.3.9 AAA
What it requires
The AAA cousin of 3.3.8: authentication cannot rely on a cognitive-function test, period. The exceptions for object recognition and personal content that apply at AA do not apply here. Memory tests, transcription tests, and image-recognition challenges all fail at this level.
Frequency
AAAaspirational target; not yet referenced by any major regulation
Passkeysthe spec-aligned path to satisfying this is device-based authentication
Why it fails
Even sites that replaced traditional CAPTCHAs with object-recognition challenges (the AA exception) fail 3.3.9. The criterion is the working group's signal about where authentication should go: away from cognitive challenges entirely, and toward device attestation or biometric verification.
The fix
Adopt passkeys (WebAuthn) as the primary authentication mechanism; treat password-plus-passkey as a transition state rather than a destination. Where image recognition has been retained for risk-scoring, run it server-side from behavioural signals rather than as a user-facing challenge.
The 2.2 additions are not where most of accessibility's hardest problems live. They are where the most-frequent, most-measurable production failures live — which is exactly what they were chosen for.
What the nine have in common
Read as a catalogue, the nine new criteria share a common editorial pose. They are not new failure modes the working group invented; they are the failure modes that have shown up most consistently in the years since WCAG 2.1 published. The working group treated them as gaps to be closed: dense toolbars (2.5.8), sticky overlays (2.4.11 / 2.4.12), CAPTCHA-style authentication (3.3.8 / 3.3.9), default focus rings (2.4.13), repeat-the-address checkout patterns (3.3.7), drag-only list reorderings (2.5.7), and the help-link placement inconsistency that frustrated cognitive-disability advocates (3.2.6).
The legal-reference picture lags because the version-pinning mechanism is slow. EN 301 549 V4 — the single biggest pending event — would cascade WCAG 2.2 across the EU Web Accessibility Directive, the European Accessibility Act's conformance reference, and every national web-accessibility law that points at the harmonised European standard. A 2026 publication is the working assumption inside ETSI JTC HF; a 2027 slippage is the more cautious one. The UK's PSBAR amendment, following the closed February 2026 consultation, is expected before year-end. The US Section 508 update remains the slowest-moving large piece — even the 2.1 update is still pending in 2026; a 2.2 update is realistically a late-2020s instrument.
For 2026 planning purposes, WCAG 2.2 is the standard that will be cited in law and procurement for the rest of the decade. WCAG 3 (Silver) remains in Working Draft and is not on a near-term Recommendation track; the most recent public draft, in 2025, made clear that Recommendation-level publication is not expected before 2028. Version-pinning practice in regulations means 2.2 will remain referenced for years after 3.0 publishes. The pragmatic procurement clause — require WCAG 2.2 at level AA as the conformance target, require a VPAT 2.5 ACR dated within the past 12 months, require the vendor to identify any of the nine new criteria where conformance is not yet met — works under any jurisdiction whose underlying law still pins at 2.0 or 2.1, because nothing in those laws prevents a buyer from contracting for more.
Your 2.2 readiness checklist
Procurement language (do this now)
Require WCAG 2.2 at level AA as the conformance target in new contracts
Require a VPAT 2.5 ACR dated within the past 12 months from every vendor
Require vendors to identify any of the nine new criteria where conformance is not yet met, plus a documented remediation roadmap
Treat "WCAG 2.1 AA at minimum, with reporting against 2.2 where the vendor's response permits" as the floor — not the ceiling
Engineering regression tests (catch the AA five before audit does)
Tab-into-form behaviour at common viewport sizes, with every overlay open (2.4.11)
Hit-target dimensions in icon toolbars, dashboards, and data-table headers (2.5.8)
Single-pointer alternatives for every drag interaction — list reordering, sliders, croppers (2.5.7)
Login, sign-up and password-reset flows free of cognitive-function tests; paste enabled in OTP fields (3.3.8)
Cross-step persistence: no field asked twice in the same authenticated process (3.3.7)
Editorial / IA review (the two A-level additions)
Single canonical placement for help mechanisms across templates (3.2.6)
Cross-team flow review for any multi-step process owned by more than one team (3.3.7)
2026 outlook items to track
EN 301 549 V4 publication — triggers WCAG 2.2 across EU web-accessibility law
UK PSBAR amendment — first major Anglophone jurisdiction to pin at 2.2
US Section 508 ICT update — 2.1 still pending; 2.2 is a late-2020s instrument
VPAT 2.5 cadence — any ACR dated 2025-or-later should report against 2.2
The WCAG 2.2 transition is structurally two transitions running on different clocks. The law transition is slow, dependent on a small number of standards bodies — ETSI JTC HF above all — and will continue through 2026–27. The practitioner transition is largely already done: auditors score against 2.2, design systems align with it, vendors file VPAT 2.5 ACRs reporting against it, and the nine new criteria are now the established vocabulary of accessibility audits. The interesting analytical question is no longer whether WCAG 2.2 is the working standard — it is — but whether the regulatory references will catch up before WCAG 3 starts pulling attention forward.
MethodologyFailure-rate descriptors aggregated from independent auditor reports issued through Q1 2026 across SaaS, e-commerce, and public-sector audit cycles. Qualitative descriptors used where firms publish ordinal rather than precise rates.
ScopeThe nine new WCAG 2.2 success criteria only. SC 4.1.1 Parsing, retired in WCAG 2.2, is out of scope. WCAG 2.1 carry-forward criteria are out of scope.
---
title: WCAG 3: what the working draft means for current sites
url: https://www.disabilityworld.org/articles/wcag-3-preview-implications/
description: WCAG 3 (Silver) is still a W3C Working Draft. The proposed standard moves from binary pass/fail criteria to scored outcomes, introduces bronze/silver/gold conformance tiers, and broadens its scope to cognitive, voice and AAC modalities.
author: Disability World
pubDate: 2026-05-22
tags: wcag, wcag-3, silver, w3c, standards, explainer
---
# WCAG 3: what the working draft means for current sites
Image description: A printed WCAG 3 working draft with colored sticky-tab bookmarks on a desk beside a WCAG 2.2 document — the visual marker for the WCAG 3 preview primer.
Reading Time: 12 minutes
WCAG 3 — the next-generation accessibility guideline the W3C has been drafting under the working name Silver since 2017 — is still, in mid-2026, a W3C Working Draft. That single fact is the most important thing to know about it. It is not a Recommendation, it is not a Candidate Recommendation, and nothing in it can yet be cited by a regulator, a court, or a procurement officer with legal force. WCAG 2.2 remains the standard the world is currently auditing against, and EN 301 549, U.S. Section 508, and the national implementations of the Web Accessibility Directive all reference WCAG 2.x. What WCAG 3 represents is a deliberate architectural rewrite of how accessibility conformance is measured — and a glimpse of what the next ten years of regulator adoption will look like once it stabilises.
This primer covers what WCAG 3 is, what it changes structurally, how its proposed bronze/silver/gold conformance tiers work, when Candidate Recommendation might realistically appear, the political tension with WCAG 2.2 (which national regulators are still in the middle of adopting), and what teams running on 2.x today should actually do about it now. The short version: read the working draft, do not refactor for it, and treat any vendor promising "WCAG 3 conformance" today as either confused or selling something.
What WCAG 3 actually is — and what it is not
WCAG 3 is the working title of a new Recommendation track at the W3C's Accessibility Guidelines Working Group (AG WG), distinct from the WCAG 2.x line. The project started in 2017 under the project name Silver (the chemical symbol Ag, an in-joke on "Accessibility Guidelines") and the first public Working Draft was published in January 2021. The most recent Working Draft is the version readers will find at the w3.org/TR/wcag-3.0/ URL — and the W3C dates that draft, like every draft before it, with a prominent header banner reading "This document is a Working Draft. It is not stable and should not be referenced or used as a basis for implementation."
That banner is doing real work. Inside the W3C process, a document moves through five maturity levels: Working Draft, Candidate Recommendation (CR), Proposed Recommendation (PR), Recommendation (REC), and finally Superseded Recommendation. WCAG 2.0 reached REC in December 2008. WCAG 2.1 reached REC in June 2018. WCAG 2.2 reached REC in October 2023. WCAG 3 has not yet entered CR — and the W3C has been explicit that several substantive design questions still need to be resolved before it can. The current state, as of the most recent published draft, is that of a research-and-design document with workable sections and clearly-flagged open issues, not a stable specification.
What WCAG 3 is not: it is not a replacement for WCAG 2.2. The W3C has stated that WCAG 2.2 and WCAG 3 will likely coexist for an extended transition period after WCAG 3 reaches Recommendation. WCAG 3 is also not "WCAG 2.3" — its content model, conformance model, and editorial structure are different enough that re-numbering inside the 2.x line was rejected early in the design process.
Purpose and scope: why a new line at all
Three structural problems with WCAG 2.x drove the decision to start a new line rather than continue incrementing the 2.x numbering.
First, scope. WCAG 2.x is, technically, the Web Content Accessibility Guidelines — it targets web content rendered in a user agent. The Working Group's mandate, however, has expanded over a decade to cover the full surface of digital accessibility: native mobile applications, kiosks, voice interfaces, virtual and augmented reality, AAC (augmentative and alternative communication) tooling, conversational AI surfaces. WCAG 3 is being designed from the outset to be content-and-platform-agnostic, with the same guideline applying to a web page, a native app screen, a voice flow, and a kiosk dialog without forcing teams to write three different conformance statements against a guideline whose name still says "Web."
Second, conformance model. WCAG 2.x conformance is binary: every applicable success criterion either passes or fails, and a single failure on a single AA criterion sinks the page's conformance claim. This works for crisp interface-level criteria like "use semantic headings" — it works less well for criteria where the underlying barrier is graded rather than categorical, such as language complexity, cognitive load, or how clearly an error message communicates what went wrong. WCAG 3 introduces scored outcomes so that a page can have a measurably better result on, say, "clear language" without forcing the binary call that 2.x demands.
Third, users not yet well served. WCAG 2.x has well-documented gaps for users with cognitive disabilities, users with low literacy, users relying on AAC devices, users of voice interfaces, deafblind users navigating with refreshable braille, and emerging assistive-technology modalities such as eye-gaze and brain-computer interfaces. The 2.x success criteria can be applied to these users — but they were drafted with screen-reader, magnifier, keyboard-only, and low-vision users primarily in mind. WCAG 3's guideline architecture explicitly invites contributions for cognitive, voice, AAC, and emerging-AT modalities as first-class guideline targets.
Key changes: outcomes, not success criteria
The most consequential change in WCAG 3 — the one every other change descends from — is the move from success criteria to outcomes.
A WCAG 2.x success criterion is a binary, testable statement. 1.4.3 Contrast (Minimum) states: text and images of text have a contrast ratio of at least 4.5:1, with two specific exceptions. A page either meets the criterion or it does not. That is excellent for repeatable testing and adversarial use (litigation, audit, procurement) but punishing for criteria where the underlying user need does not partition cleanly into pass/fail.
A WCAG 3 outcome, in the current draft, is a testable statement attached to one or more methods that describe how to verify the outcome and how to score the result. Outcomes can be binary where binary is the right shape (a form field either has a label or it does not) but they can also be scored on a numeric scale where the underlying user need is graded (how readable is this paragraph; how recoverable is this error state; how predictable is this navigation). The conformance result for a product is then computed across outcomes rather than gated on every-criterion-pass.
Several other architectural changes follow:
Guidelines as the organising unit. WCAG 3 groups outcomes under guidelines (which roughly correspond to the principles-and-guidelines layer of WCAG 2.x, but are written more declaratively).
Methods, not techniques. WCAG 2.x has informative techniques that suggest how to satisfy a success criterion. WCAG 3 has normative methods that describe how an outcome is verified. The shift from "informative" to "normative" matters: it means the testing procedure travels with the guideline rather than being a separate, debatable accompaniment.
Atomic and holistic tests. Some outcomes are tested at the atomic level (one element, one rule) and some are tested holistically across an entire view or task flow. Cognitive-load and clear-language outcomes are inherently holistic; contrast and labelling outcomes are inherently atomic. WCAG 3 makes that distinction explicit in the method.
Functional needs categories. The draft introduces functional needs — vision, hearing, cognition, speech, mobility, multi-sensory — as a cross-cutting axis. Each outcome is mapped to the functional needs it addresses, so a tester or regulator can ask "show me everything that affects users with cognitive needs" without re-reading the entire document.
Conformance tiers: bronze, silver, gold
Where WCAG 2.x has three conformance levels — A, AA, AAA — WCAG 3 proposes three conformance tiers: Bronze, Silver, and Gold. The labels are deliberately not letters and deliberately not cumulative-by-rule; they signal that the higher tiers reflect a meaningfully better experience for users, not "the same product with more boxes ticked."
Bronze is the minimum conformance tier. It is intended to correspond, roughly, to "WCAG 2.x AA-equivalent" — that is, a Bronze-conforming product should not be substantially worse than today's AA-conforming product. Bronze conformance requires passing all critical errors (outcomes flagged in the draft as fundamental barriers — for example, missing alternative text on informative images), and achieving a defined threshold across the outcome score across the product. The draft proposes that critical errors remain binary even within the scored model: any critical error blocks Bronze conformance regardless of how well the product scores elsewhere.
Silver is the intermediate tier and is intended to correspond, roughly, to a strong AA-plus product — better than the WCAG 2.x AA bar but not yet at AAA. Silver typically requires a higher threshold across the same scored outcomes, plus passing additional outcomes that are not required at Bronze. The specific thresholds are still under consultation in the working draft.
Gold is the top tier. It is intended to represent a product that has been designed and tested for the full range of functional needs the guideline covers, not only the ones the existing 2.x AA criteria mostly addressed. Gold is the tier where the cognitive, voice, AAC and emerging-AT outcomes carry the most weight, because those are the user groups where 2.x conformance does not currently produce a comparable result.
Two important properties of the tier model worth noting. First, scope is per-view or per-flow, not per-page: a product can carry different conformance tiers on different surfaces, which is more honest than WCAG 2.x's per-page model for complex applications. Second, the conformance claim travels with the methods used to verify it — so a Silver claim under WCAG 3 should be reproducible by another tester following the same methods, in a way that WCAG 2.x AA claims (which rely heavily on tester judgement at the edges) often are not.
Emerging assistive-technology modalities
A major editorial commitment of the WCAG 3 project is first-class support for assistive-technology modalities that WCAG 2.x has historically addressed only obliquely.
Cognitive accessibility is the largest of those expansions. The current draft incorporates outcome work that was previously developed in the W3C's separate Cognitive Accessibility Task Force output (the Making Content Usable for People with Cognitive and Learning Disabilities document). Outcomes in this area cover clarity of language, predictability of navigation, support for orientation and wayfinding, error prevention and recovery, and minimisation of unnecessary cognitive load. Many of these outcomes are scored rather than binary — there is no clean pass/fail for "is this sentence readable enough" — and that is the case the scored conformance model was built to handle.
Voice and conversational interfaces are explicitly in scope. Outcomes address the recognisability of voice prompts, the discoverability of voice commands, the recovery path when voice misrecognition occurs, and the equivalence between voice and visual interaction in dual-modality interfaces. This is the part of the draft where the platform-agnostic guideline architecture matters most: a voice-only flow on a smart speaker cannot meaningfully be tested against WCAG 2.x's "web content" success criteria, but it can be tested against WCAG 3 outcomes drafted to be modality-neutral.
AAC (augmentative and alternative communication) users — people who communicate primarily through symbol boards, picture exchange systems, or speech-generating devices — are explicitly addressed in the draft's user-research targets. Outcomes here relate to symbol consistency, support for AAC input as a first-class interaction mode, and the cognitive predictability of dialog states that an AAC user needs to navigate.
Emerging AT — eye-gaze, switch interfaces, brain-computer interfaces, head-tracking, and the assistive surfaces of mixed-reality devices — is named in the draft's roadmap. The Working Group's working position is that the guideline architecture should accommodate these modalities without requiring the document to enumerate every possible AT; the functional-needs axis is one mechanism for that.
Timeline: when Candidate Recommendation might land
The honest answer is that no one outside the AG WG can give a confident date, and no one inside it has published one. The W3C's process is consensus-driven, and the design questions still open in WCAG 3 — the precise scoring methodology, the exact thresholds for Bronze/Silver/Gold, the conformance-statement format, the testability of the cognitive outcomes, the relationship to WCAG 2.2 during the transition — are non-trivial. Working Drafts in any standards lineage can sit at that maturity for years.
What can be said with reasonable confidence is the shape of the path. Candidate Recommendation is the next maturity step after the current Working Draft, and CR cannot be entered until the Working Group resolves the open issues currently flagged in the draft and demonstrates that the proposed outcomes are testable (a process the W3C calls "feature-at-risk" review and that takes substantial implementation experience to clear). Several public statements from W3C staff during 2025 indicated that CR for WCAG 3 was still some way off and that the project should be treated as years rather than months from a stable specification.
Once CR is entered, the standard timeline calls for at least one implementation period of several months during which the working group gathers evidence that the outcomes have been verified against real products. PR follows. REC follows that. After REC, the slow process of regulator adoption begins — and that has historically been measured in years, not months. EAA-style citation of WCAG 3 through a revised EN 301 549 (a V5 or later) is, on any realistic reading, a late-2020s prospect rather than an immediate one.
The tension with WCAG 2.2
WCAG 3 sits in real political tension with WCAG 2.2, and that tension is the subtext of every WCAG 3 discussion inside the industry. WCAG 2.2 reached Recommendation in October 2023 — a published, stable, citable standard that national regulators are still in the middle of adopting. Some have adopted it already. Some have not. The forthcoming V4 of EN 301 549 will incorporate WCAG 2.2; U.S. Section 508 is in the middle of a refresh that points to WCAG 2.x; private-litigation defence in the United States cites WCAG 2.x by default.
The tension is not really about which document is "better." It is about whether regulators can adopt a standard that is still moving — and whether teams that have just invested in WCAG 2.2 conformance should believe a different framework is around the corner. The Working Group's stated position is that the two lines are not zero-sum: WCAG 2.2 remains the operative standard for regulator adoption, and WCAG 3 is the next generation that will, in time, succeed it. Both documents will be maintained at the W3C side by side once WCAG 3 reaches Recommendation, and the W3C has signalled that the transition will be deliberately long enough that teams do not face a forced migration.
In practice this means three things. WCAG 2.2 audit work is not wasted — the underlying access barriers it identifies do not disappear under WCAG 3, they are reorganised into outcomes. Regulators that are mid-adoption of WCAG 2.2 are not making a mistake — they are doing the work that needs doing this decade. And vendors marketing "WCAG 3 conformance" against a working draft are misrepresenting the standard's maturity; no conformance claim against an unstable Working Draft is meaningful.
First-class, incorporated from W3C cognitive-task-force work
Voice / AAC / emerging AT
Not directly addressed
Named as in-scope modalities with dedicated outcomes
Testing artefact
Informative techniques accompany the criteria
Normative methods travel with each outcome
Granularity of claim
Per-page conformance claim
Per-view or per-flow conformance claim
Cited by regulators today
Yes (EAA via EN 301 549, WAD, Section 508 refresh, courts)
No — Working Draft cannot be normatively cited
Realistic adoption horizon
Operative now; multi-year regulator rollout still ongoing
Late-2020s at the earliest, contingent on CR/PR/REC progress
Implications for 2.x sites today
The practical question for any team running a site, app, or product on WCAG 2.x today is: should we do anything differently because WCAG 3 is coming? The answer breaks into three pieces.
Audit and remediate against WCAG 2.2 AA. This is the standard that regulators are adopting, that EN 301 549 V4 will incorporate, and that courts in jurisdictions with private rights of action cite. A 2.2 AA audit done well in 2026 is not throwaway work — the underlying barriers will still be barriers under WCAG 3, and the remediation effort to fix them is the same. Teams that postponed 2.2 work in the hope of "doing WCAG 3 instead" are choosing a worse outcome on a longer timeline.
Read the WCAG 3 Working Draft, do not refactor for it. The draft is a useful window into where the standard is heading and which user needs the next decade will foreground. Teams should read it (it is freely available at the W3C TR site), share it inside design and engineering, and use it to prime conversations about cognitive accessibility, voice interfaces, and AAC. They should not, however, start writing conformance statements against it, drafting procurement clauses against it, or restructuring audit programmes to anticipate it. The draft is not stable enough for any of those activities.
Invest in the user-research and design-research capacity that WCAG 3 will require. The scored, holistic, modality-agnostic outcomes that WCAG 3 introduces cannot be verified by automated scanning tools alone. They need design research with users who have cognitive disabilities, with AAC users, with voice-interface users. The teams that will be ready when WCAG 3 reaches Recommendation are not the ones with the most sophisticated automated tooling — they are the ones with established user-research relationships across the full range of functional needs. Building those relationships now is investment that pays off under either standard.
WCAG 3 in the standards graph you already know
If you have followed the accessibility-standards arc — from Section 508 through EN 301 549, from the W3C's WCAG 2.0 through 2.1 and into 2.2 — WCAG 3 is the next generation of that arc, currently mid-design. It is the document that the standards community is building because the limitations of WCAG 2.x's binary, web-only, success-criterion model have become hard to ignore as digital accessibility has expanded into mobile, voice, AAC and cognitive interfaces. It is also, today, an unstable Working Draft that no regulator can yet cite and no responsible vendor can yet claim conformance against.
For practitioners scoping the remainder of this decade: WCAG 2.2 is the standard to audit against, EN 301 549 V4 is the procurement instrument to align with, and WCAG 3 is the document to read on a Friday afternoon to understand where the work is heading. The right posture is informed patience — keep WCAG 3 in peripheral vision, do the WCAG 2.2 work in front of you, and build the user-research capacity that will matter regardless of which document the auditors cite five years from now. For the next instalment in this primer series, see the WCAG 2.2 adoption-rate survey tracking which national regulators have already crossed the line.
```
