Blog
3 in 4 US government PDFs fail basic accessibility checks, with 6 months left before the ADA deadline
On 26 April 2027, US state governments, and cities and counties with 50,000 or more people, must make their web content meet WCAG 2.1 Level AA under Title II of the ADA. PDFs count. FileSeen downloaded 4,779 government PDFs and checked every one: only 23.5% passed five basic checks.
184 days to 26 April 2027, when states and large cities and counties must meet WCAG 2.1 AA (184 days as of 24 October 2026).
What the test found
4,779 public PDFs from 272 state agencies, state portals, cities, counties and public universities in all 50 states and DC, fetched on 24 October 2026.
- Only 23.5% pass five basic checks (tagged, language set, title shown, real text, alt text on images): 1,121 of 4,779 files.
- 32.7% have no tags at all, so a screen reader gets no headings, tables or reading order.
- 0.6% pass the PDF/UA-1 machine checks in veraPDF, the open-source PDF validator (28 of 4,767).
- Files dated since the April 2026 delay still fail 64.4% of the time.
- Forms and applications fail 74.7% of the time, and forms in use are not covered by the exception for older documents.
- California’s state sites pass 92.3%. South Carolina’s sampled files pass 0%.
These are minimum failures. A file that passes an automated check can still be hard to use; a file that fails one is missing a feature assistive technology relies on.
What the rule says, and what changed
The Department of Justice published the ADA Title II web rule on 24 April 2024. An interim final rule published on 20 April 2026 (91 FR 20902) moved the dates back by one year. The dates in force today, in 28 CFR 35.200:
| Who | Must meet WCAG 2.1 Level A and AA from |
|---|---|
| States, and other public entities with a total population of 50,000 or more (except special district governments) | 26 April 2027 |
| Public entities under 50,000, and all special district governments | 26 April 2028 |
PDFs are named in the rule: 28 CFR 35.104 defines “conventional electronic documents” as content in “portable document formats (“PDF”), word processor file formats, presentation file formats, and spreadsheet file formats.” Agencies and state universities count with their state’s population under the same section, so every site in this test is on the 2027 date.
Two exceptions in 28 CFR 35.201 matter for PDFs:
- Archived web content: made before the deadline, “retained exclusively for reference, research, or recordkeeping”, “not altered or updated after the date of archiving”, and stored in an area “clearly identified as being archived”.
- Preexisting documents posted before the deadline, “unless such documents are currently used to apply for, gain access to, or participate in the public entity’s services, programs, or activities.” A permit application posted in 2019 is still covered if people use it today.
So the useful question is not only how many old PDFs fail, but whether the documents made today, and the forms people still fill in, are any better. That is why the results are split by document date and by type.
The results
| Check | Pass rate |
|---|---|
| Tagged (has a structure tree) | 67.3% |
| Document language set | 68.2% |
| Title set and shown in the viewer | 35.7% |
| Real text (not a scan or picture of text) | 95.2% |
| Every image in a tagged file has alt text | 35.8% of files with images |
| All five basic checks | 23.5% |
| veraPDF PDF/UA-1 (machine-checkable rules) | 0.6% |
Of 57,333 images inside tagged PDFs, only 27.0% have a text alternative.
Newer files are better, but not good
| PDF internal date | Files | Basic pass | Untagged |
|---|---|---|---|
| Before the rule (before 24 Apr 2024) | 1,726 | 12.1% | 43.6% |
| Between the rule and the delay | 1,975 | 27.7% | 24.9% |
| Since the delay (20 Apr 2026 or later) | 1,006 | 35.6% | 26.8% |
The direction is right; the pace is not. At 35.6%, nearly two in three documents made since the delay would fail on day one, and anything posted after 26 April 2027 gets no “preexisting” exception at all.
Forms are no better. 1,166 files look like forms or applications: 74.7% fail the basic checks and 32.7% are untagged. Forms dated in the last 12 months do better (42.5% pass), but still mostly fail.
Cities trail everyone
| Entity type | Files | Basic pass | Untagged |
|---|---|---|---|
| Public universities | 644 | 35.4% | 28.0% |
| State agencies | 2,809 | 26.3% | 29.4% |
| State portals | 207 | 25.1% | 36.2% |
| Counties | 499 | 12.6% | 37.7% |
| Cities | 620 | 6.3% | 47.6% |
Of the 25 cities with at least 5 files tested, 12 had not a single passing PDF.
State by state
State government files only: the main portal plus the same five agencies in every state (public health, transportation, revenue, labor and environment). A state is ranked only with at least 25 files from at least 2 sites. Samples are small: treat a gap of a few points as noise.
| Rank | State | Basic pass | Untagged | Files | Sites |
|---|---|---|---|---|---|
| 1 | California | 92.3% | 1.1% | 91 | 5 |
| 2 | Minnesota | 78.0% | 7.3% | 41 | 3 |
| 3 | Colorado | 74.2% | 19.4% | 62 | 4 |
| 4 | Texas | 61.7% | 11.1% | 81 | 5 |
| 5 | Iowa | 49.4% | 11.1% | 81 | 5 |
| 6 | North Dakota | 46.5% | 16.8% | 101 | 6 |
| 7 | Illinois | 39.0% | 19.0% | 100 | 5 |
| 8 | Maryland | 38.3% | 31.7% | 60 | 3 |
| 9 | Wisconsin | 35.9% | 20.7% | 92 | 5 |
| 10 | Washington | 35.1% | 28.7% | 94 | 5 |
| 11 | Tennessee | 34.0% | 25.0% | 100 | 5 |
| 12 | Montana | 31.7% | 25.6% | 82 | 5 |
| 13 | Florida | 28.8% | 22.5% | 80 | 4 |
| 14 | Nebraska | 28.6% | 16.7% | 42 | 3 |
| 15 | Indiana | 27.5% | 20.0% | 120 | 6 |
| 16 | Maine | 27.0% | 32.0% | 100 | 5 |
| 17 | New York | 24.5% | 46.9% | 49 | 3 |
| 18 | South Dakota | 24.0% | 26.0% | 100 | 5 |
| 19 | Kentucky | 21.7% | 53.3% | 60 | 3 |
| 20 | Hawaii | 20.8% | 35.6% | 101 | 6 |
| 21 | Alabama | 18.8% | 42.5% | 80 | 4 |
| 22 | West Virginia | 17.2% | 29.7% | 64 | 4 |
| 23 | Missouri | 17.1% | 26.8% | 41 | 3 |
| 24 | North Carolina | 16.7% | 35.7% | 42 | 3 |
| 25 | Pennsylvania | 15.0% | 22.0% | 100 | 5 |
| 26 | Wyoming | 12.0% | 48.0% | 25 | 3 |
| 27 | Connecticut | 11.9% | 33.1% | 118 | 6 |
| 28 | Vermont | 11.8% | 23.7% | 93 | 6 |
| 29 | Ohio | 11.8% | 26.3% | 76 | 5 |
| 30 | Georgia | 11.1% | 37.8% | 45 | 4 |
| 31 | Virginia | 10.0% | 22.5% | 40 | 2 |
| 32 | Oklahoma | 9.1% | 52.7% | 110 | 6 |
| 33 | Delaware | 7.6% | 37.0% | 92 | 5 |
| 34 | Nevada | 7.5% | 25.0% | 40 | 2 |
| 35 | New Mexico | 7.1% | 69.0% | 42 | 3 |
| 36 | Alaska | 5.0% | 50.0% | 40 | 2 |
| 37 | Mississippi | 4.7% | 40.7% | 86 | 5 |
| 38 | Arkansas | 3.2% | 54.8% | 62 | 5 |
| 39 | New Jersey | 2.5% | 41.2% | 80 | 4 |
| 40 | South Carolina | 0.0% | 35.0% | 40 | 2 |
Not ranked. No state government site could be tested in Louisiana, Massachusetts, Michigan, New Hampshire, Oregon, Utah. Too few files to rank: Arizona, Idaho, Kansas, Rhode Island. Washington, DC is counted with the cities.
California’s lead is hard to miss. Its transportation, revenue and labor agencies passed on every file tested, and only 1 of its 91 sampled state files was untagged.
Why so many fail: the export, not the author
87.1% of PDFs exported from Microsoft Word or Office are tagged, yet only 28.4% pass, usually because the export skips the document title setting or the language. Files from Adobe Acrobat or Distiller workflows are tagged only 35.1% of the time. Scanned pages (4.2% of files) are the hardest case: 201 files are pictures of text, including meeting minutes, letters, signed documents and a Title VI complaint form.
The most common failures are settings, not rewrites: export with tags, set the document language, set a title and turn on “show document title”, and add alt text to images.
The coverage gap
Read this before the ranking.
What is missing
130 of the 455 sites listed could not be tested: 102 refused automated access, 27 could not be reached and 1 stalled and was stopped. A further 53 had no PDF a polite crawler could find.
The test ran from a server outside the US, and several states appear to block non-US networks. Refusals were respected, not worked around: no US proxy, no browser automation. Sites with heavy bot protection may differ from the rest.
No state government site at all could be tested in Louisiana, Massachusetts, Michigan, New Hampshire, Oregon, Utah.
| Type | Listed | Tested | Refused | Unreachable | No PDFs |
|---|---|---|---|---|---|
| Public universities | 50 | 44 | 0 | 0 | 6 |
| State agencies | 250 | 154 | 59 | 14 | 23 |
| State portals | 50 | 21 | 11 | 6 | 12 |
| Counties | 30 | 21 | 4 | 3 | 2 |
| Cities | 75 | 32 | 28 | 4 | 11 |
Every site’s outcome, with the reason, is in entities.csv.
Method
- Who: 455 public entities: each state’s portal; five agencies per state with the same functions everywhere; about the 75 largest cities; 30 of the largest counties; and each state’s flagship public university (four replaced by another large public university in the same state because the flagship’s site refused automated access). A purposive list, not a random sample.
- How: a crawler read each site’s robots.txt and obeyed it, made at most one request every 2 to 3 seconds per site, and never logged in or filled a form. It read sitemaps and up to 35 pages per site and downloaded up to 20 PDFs per state site and 30 per city, county or university, spread across document types and favouring recent files.
- Checks: the PDF catalog and structure tree were read directly (pikepdf, pypdf, pdfium), and every file was validated with veraPDF 1.30.3 against PDF/UA-1. A random 40 files were re-checked with separate code: 40 of 40 agreed on tags, language and text.
- When: 24 October 2026, 19:26 to 20:25 UTC, from first request to last check.
The method and checks are the same as in FileSeen’s EU bank PDF accessibility study, so the two can be compared: EU bank PDFs passed the same five checks 24.5% of the time.
Limitations
- Automated checks show the minimum. No tool can tell whether alt text is meaningful, whether reading order is right or whether a table makes sense read aloud. Real pass rates are lower than these.
- Coverage. 130 sites were blocked, unreachable or stopped, and 53 more had no findable PDF (see the coverage gap above).
- A sample, not an inventory. Up to 20 or 30 files per site, chosen by keyword and recency.
- Dates and types are estimates. Dates come from each PDF’s own metadata; “form” is a guess from the file name and link text.
- Not a legal finding. The deadline has not arrived, exceptions apply file by file, and this test does not say that any government breaks the law.
Download the dataset
Free to reuse under the Creative Commons Attribution 4.0 licence. Credit FileSeen.
- pdfs.csvOne row per PDF (4,779 rows): entity, state, source URL, document type, fetch time and every check result, including veraPDF. 4.5 MB
- entities.csvOne row per site (455 rows), including sites that could not be tested and why. 112 KB
- summary.jsonEvery aggregate in this post, with definitions. 153 KB
Licence: CC BY 4.0. Suggested credit: FileSeen, “US government PDFs and the ADA Title II deadline”, 24 October 2026, https://fileseen.com/blog/government-pdfs-ada-6-months-left/. The PDFs themselves are not redistributed: each row links to the original URL. Data last checked 24 October 2026.
Sources
- Interim final rule, 91 FR 20902 (Federal Register), checked 24 October 2026
- 28 CFR 35.200 (eCFR), checked 24 October 2026
- 28 CFR 35.201 (eCFR), checked 24 October 2026
- 28 CFR 35.104 (eCFR), checked 24 October 2026
FAQ
Questions about this post
When is the ADA Title II web deadline now?
26 April 2027 for states and for public entities with a total population of 50,000 or more; 26 April 2028 for smaller ones and special district governments (28 CFR 35.200, as amended by the interim final rule of 20 April 2026).
Do PDFs have to be accessible under Title II?
Yes. The rule names PDF as a conventional electronic document and requires WCAG 2.1 Level A and AA.
Are old PDFs exempt?
Some. PDFs posted before the deadline are exempt unless people still use them to apply for or access a service, such as forms. Archived files are exempt only if kept for reference, never updated and stored in a clearly labelled archive area.
What does basic pass mean here?
The file is tagged, declares its language, has a title set to display, contains real text rather than an image of text, and gives every tagged image a text alternative. It is a floor, not full WCAG conformance.
Which state did best?
California: 92.3% of its sampled state PDFs passed. Minnesota (78.0%) and Colorado (74.2%) were next.
Can the data be checked?
Yes. Every file is listed with its URL, fetch time and each check result, under CC BY 4.0, so anyone can re-download it and re-run the checks.