Q13 of 24 · Accessibility

Why does automated accessibility testing only catch around 30-40% of real issues?

AccessibilityMidaccessibilityautomationaxe-corecoveragemanual-testing

Short answer

Short answer: Automation can verify structure — presence of alt text, label associations, contrast ratios, ARIA syntax — but it cannot evaluate meaning. Whether alt text is meaningful, whether the reading order makes sense, and whether a custom widget is usable with a screen reader all require human judgment.

Detail

The 30-40% figure comes from research by WebAIM and Deque. Automated tools like axe-core evaluate rules that have deterministic answers: is this contrast ratio above 4.5:1? Is this input labelled? Does this element have a valid role? These checks pass or fail without needing to understand the content.

What automation cannot evaluate:

  • Alt text quality: an image of a customer signing a contract with alt text "image123" passes the axe check (alt attribute is present and non-empty) but is meaningless to a screen reader user.
  • Keyboard interaction flow: axe can check that elements are focusable, but it can't verify whether the Tab order makes logical sense, whether arrow key navigation within a custom widget behaves correctly, or whether focus is returned to the right place after closing a modal.
  • Screen reader announcement quality: axe can verify that a button has an accessible name, but it can't determine whether that name is clear, distinct from other buttons, and contextually meaningful.
  • Reading order: the visual layout and the DOM order may diverge — visually logical but confusing when read linearly. Automated tools rarely catch this.
  • Cognitive accessibility: plain language, clear instructions, consistent navigation patterns. These require human judgment about meaning, not structure.

The right way to present this in an interview: automated testing is a regression net — it catches regressions in structural accessibility. It is not a substitute for a manual audit, keyboard testing, and screen reader testing.

// WHAT INTERVIEWERS LOOK FOR

Can articulate the 'structure vs meaning' distinction and give specific examples of each type of issue that automation misses. Presents automation as a regression net, not a compliance tool.