The state of government website accessibility, 2026

Of the 94 federal government homepages we tested, the homepage alone showed:

• 67% had at least one serious accessibility failure (a critical or serious WCAG issue), two in three.
• 18% had at least one critical issue, under one in five, the lowest critical rate of any sector we have scanned.
• 12%, about one in eight, passed every automated check.
• The median site had just 6 detectable issues on its homepage, though the worst had 118.

These are detectable failures from an automated scan. They are the floor, not the ceiling (more on that below). And to be clear: this is 67% of the government sites we tested, a curated sample of federal agency and service homepages, not a census of every public-sector site.

We have now run the identical scan over three sectors, which lets us line them up on the same yardstick. Federal sites came out ahead of both:

• At least one critical issue: 18% of government sites, versus 35% of hospitals and 49% of online stores.
• Median issues per homepage: 6, versus 8 for hospitals and 17 for stores.
• Median items needing human review: just 2, versus 15 for hospitals and 24 for stores.

That last number is the tell. A low "needs review" count means a page is not just cleaner but less ambiguous: semantic, predictable markup that an automated test can actually reason about. The most likely reason is the US Web Design System, the shared, accessibility-first component library that federal agencies are steered toward. When teams build from accessible components instead of hand-rolling everything, it shows up in the data. This is the strongest argument in the series so far for a shared design system over bespoke front-ends.

Here is the other half of the story. Government is the one sector with a black-letter legal duty to get this right, Section 508 has required accessible federal websites for two decades, and its standard now points at WCAG 2.0 Level AA. And yet two in three of these homepages still shipped a serious failure, and a serious failure was about as common here as on hospital sites.

So the lesson cuts both ways. A mandate and an accessible design system measurably narrow the gap, government pages are cleaner than anything else we have scanned. But they do not close it. A component library only helps on the parts of the page that use it, and the failures we found tend to sit in the custom bits bolted on around the edges.

Ranked by the share of tested sites affected, the recurring WCAG problems were:

• Colour contrast (22%) — text that does not meet the 4.5:1 ratio, WCAG 1.4.3. As in every sector we have tested, this was the most common WCAG failure, and our figure includes text on CSS gradients that most scanners skip.
• Touch target size (21%) — controls smaller than the 24×24px minimum, WCAG 2.5.8.
• Images missing alt text (10%) — WCAG 1.1.1.
• Links with no discernible text (9%) — usually icon-only links, WCAG 2.4.4 / 4.1.2.

The top two are the same top two we found on stores and hospitals: contrast and target size are universal problems, not sector-specific ones. Structural best-practice issues were more widespread than any single WCAG failure, content not contained in landmarks showed up on 50% of sites and non-unique landmarks on 34%, but those are advisory rather than WCAG success-criterion failures, so we have kept them separate from the figures above.

The honest caveat that runs through this whole series applies here too, even though the numbers are smaller. The median government site had 2 items flagged as "needs human review", things an automated test cannot decide on its own, like whether alt text is accurate or whether a custom form is actually operable by keyboard.

Automated testing only covers the machine-checkable part of WCAG, which independent research puts at roughly a third of the success criteria. So the failure rates above are almost certainly an undercount of the real barriers, not an exaggeration. Government sites tripping fewer automated traps does not prove they are fully usable, only that the machine-visible layer is in better shape than elsewhere.

The takeaway is not that these agencies are breaking the law, automated failures are not a legal compliance verdict, and we are naming no one. The takeaway is more useful than a scolding: the sector with a mandate and a shared accessible design system produced the cleanest results we have measured, which is real evidence that both of those things work. And it still left a serious failure on two-thirds of homepages, which is evidence that neither is a finish line.

The failures that remain are mostly ordinary front-end work: contrast, alt text, link and button names, on the custom pieces a component library does not cover. Those belong in the code itself, not in a layer added on top of it. You can see where your own site stands with the free scan, or wire it into CI with the CLI and GitHub Action so it does not regress after the next release.

In the interest of not overclaiming:

• Sample: homepages of 94 US federal government sites, scanned in July 2026. We attempted 110; 16 were excluded because they returned bot-protection blocks (HTTP 403), protocol errors, or did not resolve, rather than a real homepage. Federal sites sit behind aggressive bot mitigation more often than commercial ones, which is the main reason for the exclusions. This is a curated sample of recognisable federal agencies and services, not a random or traffic-ranked one, and it does not include state or local government.
• Engine: the axe-core rules for WCAG 2.x Level A and AA, plus our own resolution of colour contrast over CSS gradients. The same engine and settings we used for the stores and hospitals reports, so all three are directly comparable.
• Scope: homepages only. Application flows behind them, benefits enrolment, tax filing, form wizards, often fare worse and are exactly where the custom, non-design-system code tends to live.
• As loaded: pages were scanned as they rendered, including any cookie or consent overlays, which can both add and mask issues.
• Not a compliance audit: detectable WCAG failures are not a Section 508 determination, and automated testing does not cover the criteria that need a human. We have reported aggregate figures only and named no individual agency.