What the service checks – and what it doesn’t

Web accessibility is measured against the WCAG success criteria. Some of them can be measured, a larger part can only be judged. The service keeps the two apart – and tells you in the result how far the scan reaches for each criterion.

What is checked automatically

A real browser loads every selected page as a visitor would see it. What is checked is the rendered page, not the source code. For every page, the scan runs:

• Rule checks for missing alternative texts, insufficient contrast, unlabelled form fields, faulty markup and much more.
• Rendering at 320 pixels wide: do people have to scroll sideways to read content? (WCAG 1.4.10)
• Keyboard operation: the scan presses real Tab keys and checks whether the focus is visible, never gets stuck and is not hidden.
• Evidence: occurrences come with screenshots, the selector and the measured value – for example the actual contrast ratio.

Some requirements only show when several pages are compared, for example whether the navigation appears in the same order everywhere. The service checks that too.

The scan closes cookie consent layers cautiously, so it checks your pages rather than the same banner thirty times. Findings from the consent layer itself count and are grouped separately in the report.

How far the scan reaches per criterion

Not every criterion can be measured equally well. The service distinguishes three levels:

Of the 50 success criteria of WCAG 2.1 AA, the service checks 4 fully and 23 largely or partially. 23 cannot be measured at all.

Anything not fully checked never counts as passed through measurement alone. For each of these WCAG 2.1 AA criteria there is a question that lets you assess it yourself. If it stays open, the report and the statement list the criterion as not evaluated.

Where the scan is uncertain

Some spots cannot be decided reliably even by a browser – for example text over a background image. These occurrences are listed separately in the report and count neither as a defect nor as passed until you assess them: Assessing uncertain results.

What only you can judge

Whether an alternative text describes the image accurately, whether the focus order makes sense, whether videos have captions or whether form error messages are understandable is not something a machine decides. For these, the service asks you manual checks: 45 questions in 8 groups, each with instructions on how to check. Preliminary questions such as “Is there video or audio?” spare you everything that does not concern your website. Only a few questions have to be answered before publishing; the others are voluntary, and what stays open counts as not evaluated: Answering the manual checks.

Your answers are a self-assessment – your own statement, not an audit by a third party. The report and the statement mark them as such.

What the service does not do

• It does not change your website. It reads your pages like a visitor and writes nothing back. There is no script to install and no overlay – overlays do not remove barriers and do not count as proof.
• It does not replace legal advice or a full test by people. You get scan data and a draft text, not a legal assessment of your individual case.
• It does not invent conformance. The statement says explicitly that it is based on a self-assessment and names the scope and limits of the scan – including how many success criteria were not evaluated.