What the accessibility checker tells you
Around one in five people has a disability, and many of them use a screen reader, keyboard, voice control or zoom to get around the web. Some of the most common barriers are also the easiest to detect: images with no text alternative, form fields with no label, links and buttons that a screen reader can only announce as "link" or "button", and pages that stop people from zooming in.
The accessibility checker reads a page and runs a set of basic automated checks against it. For each problem it finds, you get the rule, a count, and example elements with their position in the page and their start tag, so you can find them in your code. Automated checks find only some accessibility problems; this is a starting point, not a compliance certificate.
What we check
- Page basics. A language on the html element with a valid language tag, and a page title.
- Images. Images without alt text (an empty alt is fine for decoration), image buttons and image map areas without a text alternative, and elements marked as images without a label.
- Forms. Fields without a label, whether from a label element, a wrapping label, aria-label, aria-labelledby or a title. A placeholder alone doesn't count.
- Links and buttons. Links and buttons that have no text, no image with alt text, and no label.
- Structure. Heading levels that skip, a page with no headings, and a page with no main landmark.
- Keyboard and zoom. Positive tabindex values that scramble the tab order, frames without a title, duplicate ids that labels refer to, and viewport settings that block zooming.
Colour contrast
Contrast can only be measured once styles and scripts have run, so it needs a browser. With the contrast option, we load the page in our sandboxed Chromium browser, run the basic checks on the rendered page, and run axe-core, the open-source accessibility engine. Text that is too close to its background colour is reported with its foreground colour, background colour and measured ratio. Normal text needs at least 4.5 to 1 and large text 3 to 1.
axe-core also runs many more rules. We list every rule it reports that our basic checks don't already cover, marked serious or critical where axe says so, and the rules it couldn't decide, which need a person to look at. The browser has no network of its own: every request is checked and fetched by our safe fetcher.
Reading the result
The result opens with the disclaimer, then four counts: images, form fields, links and buttons, each with how many have a problem. Below them, each rule that found problems opens to show up to 50 examples, with a CSS path and the element's start tag shown as text. The heading outline shows the page's structure at a glance. In contrast mode, the axe-core section lists each failing rule with its examples and contrast ratios.
Every finding links to a fix guide that explains who is affected, how to fix it and the WCAG criterion behind it. After fixing, run the check again, then test the page yourself with the keyboard and, if you can, with a screen reader: automated checks are where testing starts, not where it ends.
What the checker doesn't do
It doesn't log in, fill in forms or follow links, so it checks one page as a first-time visitor sees it. It can't judge whether alt text actually describes an image, whether instructions make sense, or whether a custom widget can be used with the keyboard. It never stores the page itself: the result keeps only the counts, the example elements and the heading outline. If the page is larger than we read, the result says some elements weren't checked.