Reference

What an automated scanner can and cannot check

Every accessibility vendor is asked how much their tool covers, and most answer with a percentage and no method. This page shows the working.

How these numbers were produced

Two machine-readable sources, crossed against each other, regenerated whenever the engine is upgraded:

• The list of success criteria comes from W3C's own WCAG 2.2 data file — 31 at Level A, 24 at Level AA, 87 including AAA, which we exclude because almost nobody is asked to meet it.
• The rule list comes from the metadata inside axe-core 4.13.0, the engine we run: 70 rules carrying a WCAG tag.

Generated Tue, 25 Aug 2026 08:51:17 GMT. One rule tag (2.1.3) points at Level AAA criteria, and is excluded from the count above.

A worked example of the gap

Here are four text inputs. Three of them are a problem for somebody. Only one is reported by any automated rule.

<input type="text" placeholder="Your name">   ← passes
<input type="text">                            ← flagged: rule "label"
<input type="text" title="Phone">              ← flagged, but only as best practice
<input type="text" aria-label="City">          ← passes, correctly

The first one passes because a placeholder can supply an accessible name, so the rule's condition is satisfied. But a placeholder disappears the moment somebody starts typing, several screen readers do not announce it, and its default styling rarely reaches 4.5:1. It is one of the most common real defects on the web, and a scan will tell you your form is fine.

This is what the table below cannot show you, and it is the reason we say findings rather than conformance.

The 32 criteria with no automated rule at all

This is the honest half of the picture. Nothing in this list is checked by any scan, ours included.

Notice what is in there. Whether a keyboard can escape a dialog. Whether focus moves in an order that makes sense. Whether the focus indicator is visible. Whether an error message tells you what to do. Whether the navigation is in the same place on every page. These are among the defects that stop people using a site altogether, and no scanner finds them.

That is why we sell findings and a dated testing record, and why nothing we produce says “compliant”.