Can an automated accessibility test find everything?

Automated scans lack human judgment

No, an automated accessibility test does not find everything. Such a scan runs through a page's code and flags what can be detected mechanically: insufficient color contrast, images without alt attributes, form fields without labels, missing page titles, incorrect language settings, heading structures that are incorrect. This represents a substantial portion of the requirements in EN 301 549, and a scan processes this faster and more consistently than a human would.

What a scan cannot assess is meaning. An image may have alt text yet fail to truly describe what it shows. A checkout process may be technically operable with a keyboard and still feel illogical, with focus jumping in all directions. Whether an error message on a form is actually understandable, or whether a video needs captions that match what is being said — that requires someone who looks and thinks critically, not a script that runs through the HTML. An accessibility overlay does not solve this either: it places a layer over the page while the underlying structure remains unchanged, and screen readers often struggle with it. No report, from any vendor, should claim that a passing scan proves a site is fully compliant with the law.

Where this comes from: WCAG 2.1 and EN 301 549

EN 301 549, the European standard referenced by the accessibility directive, comes down in practice to WCAG 2.1 level AA. Some of the success criteria within it are objective and machine-testable — contrast ratios, for example, can be measured precisely. Another part of those same criteria is phrased in terms like "sufficient" or "logical", and that by definition requires an assessor who understands the context. That distinction is not in a disclaimer somewhere at the bottom; it is built into the standard itself.