Skip to main content :::
:::

Accesserty Observations

US accessibility statements, promise or proof? What 488 statements say

Using the W3C guidance and the EU model statement as a yardstick, we checked what 488 US website accessibility statements include: contact details, conformance status, assessment, dates, and the WCAG versions they cite.

Citation details

Published
Updated
Source snapshot date

Accesserty (2026-11-06). US accessibility statements, promise or proof? What 488 statements say. Source snapshot date 2026-11-04. https://accesserty.com/en/observations/us-accessibility-statements-what-they-say/

Suggested citation text

Summary

Nearly eight in ten statements have email or phone contact details. In this set of 488 statements, the program found email or phone contact details or wording in 379 (78%).

Most statements make a commitment but do not state conformance. Nearly nine in ten use commitment language such as “we are committed to,” but our process identified only 40 (8%) that clearly say the site fully or partially conforms to a standard.

Among the highest WCAG versions we identified, 2.1 is the most common, and 2.0 and 2.2 each account for about a fifth. Most statements in this set read more like “we are working on it, contact us if you have problems” than a conformance report.

Key figures

Statements analyzed
488
Email or phone detected
77.7%
379 statements
No clear conformance status identified
91.8%
448 statements
Mention WCAG
70.9%
346 statements

Background

The previous article explained how we collected the data: from 3,437 websites used in everyday life in the US, we collected 839 records related to website accessibility, or 796 pages after deduplicating by statement URL. 764 could be analyzed when fetched again, and by AI page-type labels, 488 of them are dedicated website accessibility statements. This article analyzes only those 488.

Three things to keep in mind:

  • We chose the candidate websites ourselves; they are not a random sample. Every share in this article describes only this set of statements and cannot be generalized to all US websites.
  • Some sites could not be reached or read by our program, so they are not in the data.
  • What a statement says is not what a site actually does. We analyzed statement content and did not test any website.

What a statement “should” include

Among the rules we found, only federal agencies and government bodies in some states have statement rules, and their content requirements differ. So we needed a yardstick, and used two authoritative documents.

W3C Web Accessibility Initiative (WAI), “Developing an Accessibility Statement”: guidance from W3C, the body that develops WCAG, also cited in the explanation of the US Department of Justice’s 2024 Title II rule. It recommends that a statement include at least:

  • A commitment to accessibility for people with disabilities
  • The accessibility standard applied, such as WCAG 2.2
  • Contact information in case users encounter problems

It also recommends adding known limitations, measures the organization takes, technical requirements such as supported browsers, the environments tested, and applicable laws.

The model statement in Commission Implementing Decision (EU) 2018/1523: statements for EU public sector websites must cover:

  • Conformance status: fully, partially, or not compliant
  • Content that does not conform, and why
  • When and how the statement was prepared
  • Feedback and contact details
  • The enforcement procedure to use if a response is unsatisfactory

The W3C statement generator also includes fields for typical response time, the publication date, and a complaints procedure. Every field we checked maps to an item in one of these documents; how we detect each one with text rules is our own method.

How we read the statements

We fetched the statement pages again on November 4, 2026, still following robots.txt, and used fixed text rules to check whether each page mentions the elements above. We checked whether something is written on the page, not whether it is correct. The shares in this article are what the program detected; conformance status was also checked by reading each candidate, as described in Finding 2.

To check the rules, we also randomly sampled 25 pages, had an AI model (Claude) label them independently page by page, and compared the labels with the program. These 25 pages were not used to tune the rules; 13 of them are dedicated statements. Fields with high agreement are reported as exact figures; fields with lower agreement are reported only as approximate shares, or not at all. Some fields appear on very few sample pages, so agreement does not mean missed cases have been fully tested.

Agreement for each field in the 25-page check sample
FieldAgreed / pages comparedSample pages labeled “yes”How we report it
Highest WCAG version identified25/2515Exact figures
Section 50825/258Exact figures
ADA24/248Exact figures
Date22/225Exact figures (few “yes” cases)
Response time23/233Exact figures (few “yes” cases)
Email22/2414Exact figures
Phone23/2513Exact figures
Commitment language21/2522Figures, noted as possibly low
Conformance status24/254Checked by reading each candidate
External assessment22/256Approximate shares only
Complaint procedure21/245Approximate shares only
Level AA22/2512Approximate shares only, possibly low
Any contact option21/2522Not used as a headline figure
Known limitations20/253Not reported
Contact form19/248Not reported

For a few long pages, some fields could not be fully labeled and were left out of the comparison for that page, so the number of pages compared is not always 25.

Finding 1: nearly eight in ten have email or phone contact details

In 379 of 488 statements (77.7%), the program found email or phone contact details. The rules also count obfuscated email addresses, wording such as “email us,” and TTY or 711 relay services; we did not check whether these channels actually work.

Email and phone in statements (488 statements)
Contact optionStatementsShare
Email or phone (either)37977.7%
Email31865.2%
Phone24149.4%

Another 74 have no email or phone detected, but match a contact form or general contact wording. The form field was not accurate enough, so these are not counted in the main figure.

Contact information is one of the three minimum items in the W3C guidance, and its purpose is practical: when a barrier stops a user, there is somewhere to ask for help. But few statements say how soon they will reply: only 27 (5.5%).

Finding 2: most make a commitment but do not state conformance

429 of 488 (87.9%) use commitment language, such as being committed to or continually working on making the site easier to use; the program occasionally misses this, so the real share may be higher. Very few clearly say how far the site currently conforms:

Conformance status clearly stated (488 statements)
Conformance statusStatementsShare
Clearly states full conformance173.5%
Clearly states partial conformance234.7%
No clear status identified44891.8%

How we read it: the program flagged 77 statements that might state conformance, and we searched the other statements with broader keywords for missed claims. An AI model (Claude) read the context of each candidate, and a different AI model (Codex) judged them again without seeing those results. Accesserty reviewed 9 by hand: the pages where the two disagreed in a way that changes the total, plus borderline cases from the keyword search. Statements that matched no keyword were not read one by one, so these are the clear claims this process identified, and some may still be missed.

  • Wording such as “we aim to meet WCAG 2.1 AA,” “we strive to conform,” or “designed to meet” was counted as not clearly stated, because it describes a goal or effort, not the current status.
  • Statements that only say some content, such as older PDFs or third-party content, may not conform, without stating the overall status, were also counted as not clearly stated.

Other information that could back up a conformance claim is also uncommon:

  • About one in ten use wording about an assessment or audit by an outside party.
  • In 111 (22.7%), the rules found date wording related to the page or statement, such as an update, publication, review, or effective date.
  • About 15% use wording about a complaint procedure or an enforcement body.

For comparison of the rules: the EU model statement for public sector bodies requires a status of fully, partially, or not compliant. US federal agencies must publish a statement under OMB M-24-08, which asks for the standard used and known limitations, but we did not find conformance status among its statement requirements; GSA’s Section508.gov guidance also says the part of the statement on known limitations should give users actionable information rather than document the agency’s conformance. Colorado’s minimum content for state and local government statements does not include conformance status either.

This does not necessarily mean the sites are doing badly; statements may be cautious for many reasons, which we cannot tell from the statements themselves. The statements in this set read more like a commitment plus a help desk, saying the site is working on it and whom to contact, than a conformance report that says how far the site has gone. We did not analyze EU statements, so this is a comparison of rules, not an empirical comparison of statements.

Finding 3: among the highest versions identified, WCAG 2.1 is most common; 2.0 and 2.2 are about a fifth each

346 of 488 (70.9%) mention WCAG. Among those 346, the highest version the program identified is:

Highest WCAG version identified, among statements that mention WCAG (346 statements)
Highest versionStatementsShare of statements mentioning WCAG
WCAG 2.116948.8%
WCAG 2.26819.7%
WCAG 2.06719.4%
No 2.0, 2.1, or 2.2 identified (a few cite the older WCAG 1.0)4212.1%

For context:

  • The latest W3C version is WCAG 2.2, last updated in December 2024. WCAG 2.2 builds on 2.0 and 2.1 but removes success criterion 4.1.1 (Parsing); where a rule names an earlier version, conformance should be checked against that version.
  • The Department of Justice Title II rule requires state and local governments to meet WCAG 2.1 AA by 2027 or 2028, depending on population and whether the government is a special district.
  • The Section 508 standards that apply to federal agencies currently reference WCAG 2.0.

The share whose highest version is 2.0 is similar across government, education, and other categories, at around one in five (15 of 72 government, 16 of 68 education, 36 of 206 other).

  • 129 (26.4%) mention Section 508 and 120 (24.6%) mention the ADA.
  • About half name Level AA; the program occasionally misses this, so the real share may be higher.

Again: a statement that cites WCAG 2.1 may describe a goal, a reference, or a claim of conformance. Even a claim is the site’s own; we did not test any website.

Conclusion: how to read a US accessibility statement

After reading this set of 488 statements, our conclusion is that most are a promise, not proof.

That is not a criticism. Without a common format, nearly eight in ten statements do the most basic thing: give email or phone contact details so users can reach someone when they run into problems. But if you want a statement to tell you how far a site has actually gone, more than nine in ten will not say clearly.

Three practical tips for readers:

  • Look for contact details first. That is the most practical use of a statement.
  • When you see a WCAG version, check whether it is a goal or a claim of conformance; even a claim is the site’s own.
  • Look for known limitations. W3C recommends listing them, and they tell you where on the site you may run into problems.

Method

  • Accesserty collected the data itself between 2026-11-03 and 2026-11-04 (Taipei time). The 3,437 candidate websites were organized by everyday-life category. The crawler followed robots.txt (RFC 9309), and pages the crawler could not reach were checked manually.
  • Scope: the 839 collected records became 796 pages after deduplicating by statement URL; 764 could be analyzed when fetched again on 2026-11-04 (readable, at least 300 characters of main text, in English); 488 of those were dedicated website accessibility statements by AI page-type labels. None of the 21 manually added records are in the analysis: when fetched again, 17 returned HTTP errors and 4 did not meet the main-text length or English threshold.
  • Page types were judged page by page by an AI model (Claude) against a fixed definition, reading each page’s title, main heading, and main text; for long pages, only about the first 2,800 and last 700 characters. A random sample of 60 pages was then judged again by a different AI model (OpenAI Codex) without seeing the original labels. The two agreed on whether the page is a dedicated statement for 54 pages; Accesserty reviewed the other 6 by hand, and the original label was right on 3. The 54 pages where the two models agreed were not checked by hand, and both could be wrong on the same page, so this check cannot be used to estimate the accuracy of the original labels.
  • Content reading: fixed text rules identified the elements in each statement, and a 25-page sample not used for tuning was compared with independent labels from an AI model (Claude). Fields with insufficient agreement are not reported. For conformance status, keyword candidates were read in context by two AI models separately, and disagreements that change the total, plus borderline cases, were reviewed by hand.
  • Sensitivity check: recalculating with only the 410 statements the AI labeled with high confidence changed the overall shares reported here by at most 3.2 percentage points (Level AA).
  • Correction (2026-11-07): The collection program checked only the page source. We later loaded these 488 statements in a browser and checked them again. Of the 449 that were checked, 7 are on sites that load an overlay (6 on sites with a commercial overlay, and 1 on a site where a website-management service adds an assistive toolbar); under the inclusion rules they should not have been included. The other 39 could not be checked, because they failed to load, robots.txt disallowed them, or the site refused the request. The figures here were not recalculated. Removing the 7 would change shares based on the 488 statements, or on the 346 that mention WCAG, by less than 1.5 percentage points.
  • This article publishes aggregate results only. It does not provide record-level data, name or rank individual websites, or quote individual statements.

Limits

  • The candidate websites are not a random sample. The shares describe only this set of statements and cannot be generalized to all US websites.
  • We checked whether something is written in the statement, not whether it is accurate, and not how accessible the site actually is.
  • Both the text rules and the AI judgments can be wrong. The agreement check covered only 25 pages, and some fields appear on very few sample pages.
  • Our data does not record which statement rules apply to each website, so this article does not compare required and voluntary statements.