The accessibility tools universities recommend are quietly rotting

Of 66 accessibility "testing tools" pages we attempted, 60 returned a real page. Across those 60:

• 16 pages (27%) linked at least one tool that no longer exists.
• Those 60 pages carried 1,279 outbound links to 782 distinct destinations.
• 16 of those links were dead — a 404, a delisted browser extension, a parked domain, or a document moved into a vendor archive.
• Every affected page had exactly one dead link. This is slow decay, not neglect.

27% of pages but only 1.3% of links is the shape worth noticing. These lists are overwhelmingly accurate. They just each carry one entry that has quietly stopped working, and a reader has no way to tell which one without clicking every link.

The well-known tools are fine. Across the sample, every link to WebAIM and WAVE (178 links between them), Deque, NVDA, Siteimprove, Accessibility Insights, the PDF Accessibility Checker and Chris Pederick's Web Developer toolbar resolved normally.

Two things rot instead, and neither looks broken from the outside:

• Browser extensions get delisted. Five links pointed at Chrome Web Store items that no longer exist. A removed listing does not 404 — the store returns a normal page with no extension name, no developer and no install button, so an automated link checker sees HTTP 200 and reports nothing wrong.
• Vendors get acquired, and the label stops matching the destination. This is the larger effect by far.

The single biggest finding in the sample is not a dead link at all. Every one of the 27 links to TPGi, and 23 of the 25 links to Freedom Scientific, now redirect to Vispero.

Both acquisitions are real and public, and the products still exist — the Colour Contrast Analyser and JAWS are both alive and maintained. Nothing is broken. But a page that says "download the Colour Contrast Analyser from TPGi" now sends a reader to a different company's site, and a procurement officer checking a vendor list finds a name that does not match.

That is 50 of 1,279 links, on pages that pass every automated link check, because a 301 to a live page is exactly what a healthy link looks like. No link checker will ever flag these. Only a human who knows the acquisition happened will.

We drew the universities from the same sample as our Title II deadline report, so the list was chosen for an unrelated reason rather than assembled from search results. Of 73 public universities, we could find a published accessibility tools page at 13 — about 18%.

That number deserves a caveat rather than a headline. It means "no page found by following accessibility entry points and their internal links", not "no page exists". An institution that keeps its tool list behind a search box, inside a PDF, or on a login-gated intranet reads as a miss here. Treat 18% as a floor.

One pattern did hold consistently: state and city government portals publish accessibility policy, universities publish tool lists. Of 41 state portals we checked before narrowing the sample, not one carried a developer-facing list of testing tools.

A week after the scan we re-checked the pages with findings. Three institutions had already corrected theirs.

We had emailed one of them about three delisted extensions. All three were gone within hours — and not only from the page we mentioned, but from two further pages on the same site that we had never raised. Two other institutions fixed their broken links with no contact from us at all.

That is worth stating plainly because it cuts against the framing this kind of audit usually invites. These lists are not abandoned. They are maintained by people who fix things when they find out. The failure is not negligence, it is that nothing tells a maintainer when a third-party tool disappears underneath a link that still returns 200.

Every figure above comes from a single run on 2026-09-16, because a rate assembled across runs measures a moving target — the web changed underneath this audit twice in a week.

Three rules kept the numbers conservative:

• A bot wall is not a dead link. 100 links returned a Cloudflare JavaScript challenge. A spoofed user agent does not pass one and neither does an automated checker, but the resource is usually fine. All 100 are excluded from the dead count.
• A login page is not a dead link. 39 links resolved to SSO endpoints — an institution's own Siteimprove or ServiceNow sign-in. Excluded.
• A delisted extension needs a control. Because a removed listing returns HTTP 200, every one was confirmed against a known-live extension fetched at the same moment. Without that check you are measuring the store being slow, not a delisting.

The 6 pages that did not return a real page — one 404 and five bot-blocked — are excluded from the denominator rather than counted as clean, which would have flattered the result.

Three checks catch almost everything we found, and none of them need a tool:

• Open every browser-extension link. A delisted item looks like a working page until you notice there is no install button. This was 5 of our 16 dead links.
• Check whether any vendor on your list has been acquired. The link will still work. The company name on your page is what goes stale, and it is what a procurement reader notices.
• Re-check anything added more than two years ago. Every dead link we found was an older entry; nothing added recently had rotted yet.

If it is useful, our own gradient and image contrast checker and focus-state contrast checker are free, need no signup and no email address, and cover two things most of the tools on these lists do not: text over a gradient or photograph, and contrast that only fails once a control takes keyboard focus.