Research/Society
Most big US local-government PDFs fail basic accessibility checks
Of 5,412 PDFs read from the websites of large US cities and counties, 22.0% pass four basic machine checks that screen readers rely on. Files made since the 2024 ADA Title II rule do better, and the software that made a file decides most of the outcome.
Oct 17, 2026 · 6 min read

By 26 April 2027, US cities and counties serving 50,000 people or more must make their web content, PDFs included, meet the WCAG 2.1 AA accessibility standard under the Justice Department's ADA Title II rule. We read the PDFs linked from the websites of those governments six months ahead of the deadline. Of 5,412 PDFs from 926 government sites, 60.5% are tagged with the structure screen readers use, and 22.0% pass all four of the basic checks we ran. PDFs created since the rule was published do better, 25.5% against 18.2%. The biggest single factor is the software: PDFs saved from Word are tagged 94.2% of the time, scanner output 4.5%.
22.0%
of 5,412 PDFs pass all four basics (tagged, language, title, has text)
60.5%
are tagged with a structure tree
25.5% vs 18.2%
pass all four, created since the April 2024 rule against before
94.2% vs 4.5%
tagged, Word-saved PDFs against scanner output
The question
Local governments post agendas, minutes, forms and notices as PDFs, and a PDF without structure is hard or impossible to use with a screen reader. How accessible are those files today, and have documents made since the rule improved? A vendor index has reported that almost all government PDFs are inaccessible, but its data are proprietary and it gives no local-government breakdown. Other large studies test web pages and exclude PDFs. We found no open, per-government measurement for the governments facing the 2027 deadline.
What we did
We took every US city, county, consolidated city-county and town of 50,000 or more with an official website on Wikidata: 1,657 governments. For each site we honoured robots.txt, read the home page and up to 8 pages linked from it, and up to 8 PDFs linked from those pages, one request at a time. On each PDF we checked four basics: whether it is tagged (has a structure tree), whether a document language is set, whether a title is set, and whether it has extractable text rather than being an image only. We also recorded its creation date and the software that made it. We store structural facts only, no document text. This is the third full run, crawled on 16-17 October 2026, and the cleanest of the three; all three agree within about a point.
What we found
Of the 1,657 sites, 1,169 home pages were read and 267 disallow crawlers. From 926 sites we read 5,412 PDFs. Missing titles are the most common failure: only 41.9% of PDFs set one. About one in seven is an image with no text at all, which a screen reader cannot read.
Share of local-government PDFs passing each basic check
Newer files do better, except on titles
PDFs created on or after 24 April 2024, when the rule was published, are tagged more often than older ones (66.2% against 56.0%) and set a language more often (63.9% against 50.3%). The title barely moved. The comparison also holds within the same government: on the 632 sites with PDFs from both periods, the share tagged rose by 6.7 points on average, rising at 267 sites and falling at 212 (sign test p = 0.01).
PDFs created before and after the Title II rule
| Created | n | Tagged | Language | Title | Image only | All four |
|---|---|---|---|---|---|---|
| On or after 2024-04-24 | 2,776 | 66.2% | 63.9% | 42.5% | 11.6% | 25.5% |
| Before 2024-04-24 | 2,482 | 56.0% | 50.3% | 41.7% | 14.0% | 18.2% |
| Undated | 154 | 28.6% | 26.6% | 33.8% | 38.3% | 21.4% |
Share of PDFs that are tagged, by year created
Four pre-registered follow-ups looked at files created around April 2026, when the Justice Department moved the deadline back a year. Pooled over 12 months on each side, files after the extension were tagged as often as before (67.2% against 67.3%). Small drops in narrower windows looked mostly like a change in which governments posted, not governments changing their own practice.
The software decides most of it
How a PDF was made predicts its structure far better than when it was made. Word and other Microsoft Office apps tag almost everything. Print-to-PDF output and scanners almost never do, and 72.4% of scanner PDFs have no text at all. Word files still often lack a title, which is why only 32.2% of them pass all four basics.
Share of PDFs that are tagged, by the software that made them
Images inside PDFs are rarely described. Of 71,226 figures tagged in 2,021 PDFs, 13.8% have alt text. On the other hand, 72.6% of the home pages we read link to a page about accessibility or the ADA.
What it means
With six months to go, most of these PDFs would need work before they could meet the standard, and passing our four checks is only the start. The cheapest gains are upstream: export from the source document instead of printing or scanning, and set a title and language. The census is reproducible, so the same governments can be measured again after the deadline.


