Design Accessibility Audits

Accessibility · accessibility, process, checklists

Updated 2026-07-28

Accept what automation can and cannot find. Automated checks catch missing labels, poor contrast, and broken structure, which is a real but small share of the problems people actually hit.
Audit flows, not pages. A checkout that passes every page check can still be impossible to finish with a keyboard because of one focus trap in the middle.
Run the three manual passes that find most issues:

Write findings as user consequences with a reproduction path. "Cannot complete signup with a keyboard, step 3, the country selector cannot be opened" gets fixed. "Missing ARIA" does not.
Rank by severity, not by count. One blocked task outranks fifty decorative images with imperfect descriptions.
Fix causes in shared components. A contrast failure in a button token is one fix and a hundred passing screens.
Re-test with the people who reported the problem where you can. Compliance is a floor, and passing a checklist is not the same as being usable.
Put the checks that keep regressing into the pipeline so the same defect cannot ship twice.

An audit is only useful if it produces a ranked list of fixable defects, each attached to a task someone could not complete.