WCAG

Perceivable: information must be available to the senses a person has. Text alternatives for images, captions for video, sufficient contrast, content that does not rely on colour alone.

Operable: the interface must be usable by whatever a person uses. Everything reachable by keyboard, no time limits that cannot be extended, no content that flashes in a way that can trigger seizures.

Understandable: content and operation must be comprehensible. Language declared, navigation consistent, errors identified and explained.

Robust: content must work with current and future assistive technology, which in practice means valid markup and correct use of names, roles and values.

Level A is the minimum. Failing it makes content unusable for some groups entirely.

Level AA is the practical target. It is what European accessibility legislation, the United States Section 508 rules and most procurement requirements reference.

Level AAA includes criteria that cannot be met by all content — sign language interpretation for all recorded audio, for instance — and the guidelines themselves state that AAA conformance is not recommended as a general policy for whole sites.

Levels are cumulative: AA conformance requires meeting every A criterion as well.

Published analyses consistently find that automated tools detect a minority of WCAG issues — the figure most often cited is around a third, and it varies with the content.

What automates well: missing alternative text, missing form labels, contrast ratios of plain text, missing language declarations, duplicate identifiers, invalid ARIA attributes, heading level skips.

What does not: whether alternative text is accurate, whether the reading order makes sense, whether a custom control behaves as its role claims, whether an error message explains anything.

So a page with no automated findings is a page that has cleared the mechanical checks, which is the beginning of an accessibility review rather than the end of one.

Each success criterion has an Understanding document explaining its intent, with examples of sufficient techniques and documented failures. These are far more useful than the criterion text alone.

Test with a keyboard first: unplug the mouse and use the site. This finds a large share of operable failures in minutes and requires no tooling.

Then test with a screen reader on the flows that matter, which is what reveals whether names and roles are correct.

Version 2.2 added criteria on focus appearance, dragging alternatives, target size and accessible authentication, and it is backwards compatible with 2.1 and 2.0.