Last updated: May 13, 2026
Our scanner uses the open-source axe-core rules engine, the industry-standard accessibility testing library maintained by Deque Systems. axe-core is used inside Microsoft Accessibility Insights, Google's Lighthouse, US government accessibility tooling, and most enterprise accessibility scanners. By building on axe-core, our results align with what other widely deployed scanners would surface on the same page, rather than producing a proprietary classification that is difficult to cross-check.
We run axe-core's WCAG 2.1 Level A and Level AA rule set. This includes images-missing-alt, color contrast against background, form labels and associated controls, ARIA attribute validity and ARIA role validity, keyboard accessibility patterns that are detectable in static markup, heading hierarchy, document landmark structure, and accessible names on links and buttons. The current rule set is roughly 90 individual axe rules. Each scan returns the rules that fired, the elements that failed, and the WCAG success criterion each rule maps to.
Automated tools detect roughly 30 to 50 percent of real WCAG failures, depending on how the site is built. We cannot reliably test: whether alt text is actually meaningful for the image it describes, whether the heading hierarchy reflects content structure instead of just visual styling, whether ARIA is being used to describe behavior that actually exists in the JavaScript, screen-reader-specific output quality, focus order quality, or any criterion requiring human judgment. Pair automated scanning with manual keyboard testing and screen-reader testing for a real picture of accessibility.
Each violation is tagged Critical, Serious, Moderate, or Minor based on axe-core's impact classification. We derive a compliance score from zero to one hundred using the formula 100 - (critical × 15) - (serious × 8) - (moderate × 3) - (minor × 1), clamped to a non-negative result. The score is a quick comparative signal, not a pass-or-fail certification. A high score means automated checks surfaced few or low-impact issues; it does not mean the page is fully accessible.
For each violation, our system sends the offending element's HTML and the relevant WCAG rule context to an Anthropic Claude model and asks it to propose a targeted code suggestion. These suggestions are intended as starting points, not as verified drop-in solutions. Every suggestion should be reviewed by a developer, validated against the rest of the page's markup and styles, and tested before being deployed. We do not guarantee that any given AI suggestion is complete or correct for your codebase.
We update our scanner to the latest stable axe-core release on a regular cadence. Major axe-core releases occasionally change which rules are included, how rules are evaluated, or how impact is classified. When that happens, results on a previously scanned URL may shift even if the underlying page has not changed. That is expected behavior of staying current with the upstream engine.
Methodology questions, edge cases, and rule-coverage feedback can be sent to hello@adacodefix.com.