Check pasted HTML for missing alt text, unlabelled controls, heading jumps, duplicate ids and ARIA mistakes. Static markup only, never a compliance claim.
Waiting for markup
Reads the HTML as written. It does not run the page, so a clean result is not a pass.
The findings appear here
Images without alt, unlabelled controls, duplicate ids, invalid ARIA, heading jumps and landmark problems — each with how sure the tool is about it.
This reads markup and nothing else. It does not run the page, apply CSS, follow JavaScript or move focus — so it cannot see whether a control is visible, whether focus is managed when a dialog opens, what the contrast is, or whether a custom widget answers the keyboard. Automated checks of any kind catch a minority of real barriers. A clean result means nothing in this tool's list was found, which is a much weaker statement than “this is accessible”.
HTML Accessibility Inspector runs entirely on your device — nothing is uploaded.
You’re on the Free plan. Compare plans
This inspects the markup you paste for the problems that are visible in markup: images with no alt attribute, form controls with no label, heading levels that skip, duplicate ids, landmarks that are missing or duplicated, tables without header cells, ARIA roles that are misspelled or used on the wrong element, and a missing lang attribute.
What it cannot see is most of the picture, and saying so is more useful than a score. Colour contrast depends on the computed styles, not the markup. Focus order depends on the rendered layout and any JavaScript that moves focus. Whether alt text is *good* needs someone who can see the image — `alt="image"` passes a presence check and fails a person. Keyboard traps, live region behaviour, motion and reading order all need a real page and, in the end, a real person using it.
So the findings carry a confidence level rather than a pass or a fail. "Certain" means the markup is definitely wrong — an input with no accessible name has no accessible name regardless of context. "Likely" means it is probably wrong but context could justify it, such as a heading jump inside a fragment that continues an outline from elsewhere on the page.
No automated tool, including this one, can establish WCAG conformance. The best of them catch perhaps a third of real issues, and the proportion is lower for static markup with no styles and no interaction. Treat a clean result as having cleared a floor, not as having passed an audit — and the honest next steps are a browser extension on the live page, a keyboard-only pass, and a screen reader.
No, and this is the important limitation. The best automated checkers catch roughly a third of real accessibility issues, and static markup with no styles and no interaction is a narrower view still. A clean result means you have cleared a floor.
Anything needing computed styles or interaction: colour contrast, focus order, keyboard traps, live region behaviour, motion, and reading order. It also cannot judge whether alt text is useful — alt="image" passes a presence check and tells a blind reader nothing.
Certain means the markup is wrong on its own terms, such as an input with no accessible name. Likely means it is probably wrong but context could justify it — a heading jump inside a fragment may be continuing an outline from elsewhere on the page.
No tool can, and this one does not claim to. Conformance is a judgement about a whole experience, made partly by people using assistive technology. Automated checks find a subset of failures; they cannot establish conformance.
Because screen reader users navigate by heading level, and a jump from h2 to h4 implies a level that is not there. In a fragment the jump may be legitimate, which is why it is reported as likely rather than certain.
No. It is parsed in your browser and never leaves the page, which is what makes it safe for a component from an internal codebase.