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, 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.
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.
- 01How we read 100 statements
- 02Existence — half of the world’s biggest sites have nothing to show
- 03Standards — what the 22 honest citations look like
- 04Limitations — the disclosure no-one wants to make
- 05The feedback channel and the SLA gap
- 06The claim-versus-axe gap
- 07Honesty ranking — top and bottom of the 100
- 08Model statement language vs. worst boilerplate
- 09What an honest accessibility statement is for
01 — How we read 100 statements
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.
An axe-core scan is an automated-issue floor, not a ceiling. axe catches around 30-40% of WCAG violations a trained manual auditor 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%.
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.
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.
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.”
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.
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.
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.
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.
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.
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.
”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 runs the same axe-based baseline we used here; an ongoing-monitoring posture is covered in the accessibility monitoring buyer’s guide.