Automated vs manual accessibility testing: what each can establish

Use automation for repeatable evidence

A supported rule can detect a missing programmatic label or a particular structural problem on a rendered page. Repeating these checks helps identify regressions and shared-template patterns. Save the tested URL, state, rule, and available element evidence so someone can reproduce the observation.

Tools vary in coverage and classification. Different content types need different checks. A best-practice warning is not automatically a WCAG failure, and a check that could not finish does not count as a pass. Evaluate the tool’s output rather than assuming every reported count has the same meaning.

Use human evaluation for meaning and task completion

An image can have an alt attribute whose text is misleading. A form can have labels but leave keyboard users unable to recover from an error. A caption track can exist but be inaccurate. These examples show why presence checks do not establish practical accessibility.

Manual evaluation should include keyboard operation, focus behavior, screen-reader output, meaningful alternatives, media quality, and representative end-to-end tasks. Include dynamic states: expanded menus, dialogs, validation errors, loading results, and success messages.

Build a representative scope

List public and authenticated services, domains, page templates, content types, and critical tasks. Sample within that inventory deliberately. A random collection of URLs can miss a unique application form or an externally hosted service that residents depend on.

Record what was excluded and why. If a result changes between runs, examine redirects, 404s, interactive states, scanner version, and rule mappings before claiming the site changed. Coverage evidence is part of the result, not an optional appendix.

Connect findings to an acceptance process

Give each finding an accountable owner and a concrete repair. Group shared components where the evidence supports it, while keeping occurrence lists for verification. Distinguish detected failures from recommendations and unanswered questions.

• Reproduce the problem in the relevant page or task state.
• Correct the source and check adjacent states or pages using that component.
• Rerun the relevant automated rules.
• Repeat the human task using keyboard and appropriate assistive technology.
• Record unresolved limits instead of converting them into a passing status.

What your scan results mean

An automated scan can identify certain accessibility barriers on the pages it tests. It cannot find every barrier or determine on its own whether everyone can use your website. A scan with no detected errors does not establish full accessibility or legal compliance.

A useful report explains what was tested, shows the findings and supporting evidence, and identifies what still needs manual review. Use those results to prioritize improvements, then combine automated checks with human testing to understand whether people can complete important tasks on your website.