Automated scanners catch ~30% — claim

Accessibility testing starts with code checks, like a spellchecker for websites, but many barriers still need a human to use the page and judge what works. WebAIM says automated tools cannot detect every Web Content Accessibility Guidelines failure, and WAVE says “only a human can determine true accessibility.” (webaim.org) That gap surfaced again in a post by accessibility researcher Deepika Singh, who wrote that tools such as Lighthouse catch only about 30% of Web Content Accessibility Guidelines issues and called for systems that audit more like humans do. Google’s Lighthouse is one of the most widely used browser-based accessibility scanners. (x.com, developer.chrome.com) Lighthouse runs automated accessibility audits inside Chrome DevTools and scores pages on pass-fail checks. Google says the score is a weighted average of accessibility audits, and some manual audits are excluded from the score entirely. (developer.chrome.com) Those automated checks are good at machine-readable problems such as missing alternative text, low color contrast, and buttons without accessible names. They are weaker at tasks that depend on meaning, sequence, and behavior, such as whether link text makes sense, focus moves in a logical order, or a screen reader user can complete a purchase flow. (developer.chrome.com, webaim.org) Web Content Accessibility Guidelines, usually shortened to Web Content Accessibility Guidelines, are the technical rules used in many accessibility laws and procurement standards. Deque says those guidelines underpin laws including Section 508, the Americans with Disabilities Act, the Accessibility for Ontarians with Disabilities Act, the European Union Web Accessibility Directive, and the European Accessibility Act. (accessibility.deque.com) The 30% figure is widely repeated in accessibility practice, but it is also disputed. Deque published a report based on more than 13,000 pages or page states and nearly 300,000 issues arguing that the common “20 to 30%” claim uses the wrong yardstick when it counts only how many guideline criteria can be tested by automation. (deque.com, accessibility.deque.com) Even groups that build automated tools still warn against treating a clean scan as a pass. WebAIM’s 2026 Million report, based on one million home pages tested in February 2026, says all automated tools have limits and that the absence of detected errors does not mean a page is accessible or conformant. (webaim.org) That warning matters because the web still fails basic checks at scale. WebAIM found 56,114,377 distinct accessibility errors across the one million home pages it tested, an average of 56.1 errors per page, and said the total was up 10.1% from 2025. (webaim.org) Lighthouse also is not a full accessibility lab on its own. Deque says axe-core powers Google Lighthouse, and third-party comparisons note that Lighthouse runs a subset of axe-core rules rather than the full rule set used by dedicated accessibility tools. (deque.com, inclly.com) The practical result is that teams still pair automated scans with manual testing, keyboard checks, and screen reader use. WebAIM puts it plainly: WAVE can help a human evaluate a page, but it cannot tell you that the page is accessible. (webaim.org)