I scanned every US local government website. 72% fail the accessibility standard they are about to be held to.

There is a federal deadline coming that most American local governments are going to miss, and until this week nobody could tell you how many.

So I measured it. Every city, county and special district in the United States holding a .gov domain — 11,720 entities — taken from CISA's public .gov registry rather than from any list I assembled. Each homepage loaded in a real browser and tested against WCAG 2.1 Level A and AA, the standard the ADA Title II web rule points to.

8,844 answered. 6,384 of those fail at least one Level AA check. That is 72.2%.

You can look up any government free. No account, no email.

The deadline is two deadlines, and almost everyone gets this wrong

I got it wrong myself, in writing, before I checked the source — which is exactly why I am putting it near the top.

The Interim Final Rule of 20 April 2026 split the compliance date in two:

• 26 April 2027 — public entities with a total population of 50,000 or more
• 26 April 2028 — entities under 50,000, and all special district governments

Most coverage says "April 2027" and stops. For the majority of American municipalities that is simply not the date. But it is not a reprieve either: remediation, document cleanup and vendor contracts routinely run 12 to 18 months, which means the 2028 cohort is roughly now, and the 2027 cohort is already late.

I joined the scan to Census 2024 population estimates to work out which entities sit on which side. 729 failing governments fall under the 2027 date. Those are the ones out of time.

The finding I did not expect

I assumed failure would track resources — that small, poor towns would fail and large, well-staffed cities would pass. That is not what the data says.

Failure tracks the platform the site is built on.

| Platform | Sites measured | Failing | Rate | | --- | ---: | ---: | ---: | | Weebly | 117 | 110 | 94.0% | | Revize | 637 | 557 | 87.4% | | WordPress | 1,884 | 1,632 | 86.6% | | Wix | 300 | 224 | 74.7% | | Joomla | 374 | 274 | 73.3% | | CivicPlus | 353 | 244 | 69.1% | | Drupal | 165 | 93 | 56.4% | | Granicus / GovAccess | 2,150 | 1,165 | 54.2% | | Municipal Impact / MuniCode | 382 | 203 | 53.1% |

A town on Weebly is nearly twice as likely to fail as a town on MuniCode. Same deadline, same statute, same public.

And when you look at what fails, the picture sharpens. Across Granicus-hosted sites, color-contrast fails on 631 of them and link-name on 561. Those are not 631 independent mistakes by 631 different clerks. That is a shared template.

Which means the fix is not where the liability is

Here is the part I think matters, and it is an opinion rather than a finding.

A town bought a website in good faith. It had no way to evaluate the accessibility posture of what it was buying — there was no published measurement to consult, which is precisely the gap this study fills. It inherited defects it did not create and cannot see.

When a complaint lands, it names the town. Not the vendor.

Meanwhile the vendor is the only party who can fix it at scale. One correction to a shared template moves hundreds of customers at once. No individual municipality has that leverage; every platform vendor does.

I am not arguing anyone acted badly. I am arguing the incentives are pointed at the wrong party, and that eight months before a federal deadline is a strange time for that to still be true.

What this does not show

Three limits, and I would rather state them than have them found.

Homepage only. One page per entity. A site can pass its homepage and fail comprehensively behind it. Every number here is a floor.

Automated only. Automated testing catches roughly a third of real accessibility problems. It cannot judge whether alt text is meaningful, whether a form can be completed with a screen reader, or whether a keyboard user can escape a menu. Passing every automated check is not compliance, and any vendor telling a city otherwise is selling something.

Entities that did not answer are counted separately. 2,876 did not respond when scanned. They have not been shown to pass and have not been shown to fail, and folding them into either side of the ratio would be a claim the data cannot support.

The data is free

Any government can have its own row, free, on request. So can any journalist, any state municipal league, and any researcher who wants to check the method rather than take my word for it. If a result looks wrong, tell me and I will re-run it — corrections are welcome, and a measurement nobody can challenge is not worth publishing.

I do not sell remediation and I am not going to. I measure. Selling both halves would make the measurement worth less, and the measurement is the whole point.

David A. Thompson is the founder of KnightByrd Tech LLC, a software company in South Carolina. The scan was run on 10 September 2026 and no vendor commissioned, reviewed or paid for it.