Every year, WebAIM runs an automated accessibility check on the home pages of the top one million websites. The 2026 results: "95.9% of home pages had detected WCAG 2 failures," up from 94.8 percent in 2025, with an average of 56.1 errors per page. WCAG is the main set of accessibility guidelines for the web, covering things like readable contrast and labels that screen readers can announce.
If you built your app with Claude Code or another AI tool and never asked about accessibility, it's very likely in that 95.9 percent. The models learned to build websites from the web, and the web mostly fails these checks. Unless you ask for something different, you get the average.
The six errors that cover almost everything
WebAIM found that "96% of all errors detected fall into these six categories," and they're the same handful every year:
- Low contrast text: 83.9 percent of pages
- Missing alternative text for images: 53.1 percent
- Missing form input labels: 51 percent
- Empty links: 46.3 percent
- Empty buttons: 30.6 percent
- Missing document language: 13.5 percent
Each one shows up in AI-built apps in a predictable way. The light gray text on white that looks modern fails contrast. An icon-only button, like a trash can or an X to close a dialog, has nothing for a screen reader to announce, which counts as an empty button. Forms that use placeholder text instead of a real label lose the label as soon as someone starts typing. Images get dropped in with no description. And the page's lang attribute, one line at the top of the HTML, often just isn't there.
All six are small and quick to fix, and once they're fixed they stay fixed.
How to get Claude to fix it
Start by asking. A prompt like "audit this app for WCAG 2 accessibility problems, starting with contrast, alt text, form labels, empty links and buttons, and the document language, then fix them" will catch most of the list above.
Then give Claude a check it can run, so it's measuring and not guessing. Lighthouse, built into Chrome's developer tools, has an accessibility score, and axe is a widely used open-source checker that can run in a browser or in tests. Automated tools don't catch everything, but they catch the six errors above, which is most of what's out there.
To keep new pages from slipping back, put a short rule in your CLAUDE.md: every image gets alt text, every input gets a visible label, every icon button gets an accessible name, and text meets WCAG contrast. That's four lines, and it applies to every page Claude builds afterward.
Last, use the app the way some of your users will. Unplug the mouse and try to get through your signup flow with the keyboard alone. If you can't reach a button or can't tell where focus is, nobody using a keyboard or a screen reader can either.
Accessibility is part of what separates a real product from slop, and it's one of the cheapest upgrades an AI-built app can get.
Sources
WebAIM: The WebAIM Million (2026) - Automated accessibility results for the top one million home pages: 95.9 percent with detected WCAG 2 failures versus 94.8 percent in 2025, an average of 56.1 errors per page, and the six most common error types with their rates, which account for 96 percent of detected errors.
