Automated scanning catches the obvious issues fast

npx axe-cli https://prantatech.eu.org

Tools like axe-core catch missing alt text, insufficient colour contrast, and missing form labels reliably and fast - a good first pass, run automatically in CI on every deploy.

But automated tools have a real, known ceiling

Research consistently puts automated coverage at roughly a third of real WCAG issues. Anything requiring judgment - does this alt text actually describe the image's purpose, does this focus order actually make sense - needs a human.

The manual checks that matter most

  • Tab through the entire page with a keyboard alone, mouse untouched
  • Turn on a real screen reader (VoiceOver or NVDA) for at least the critical user flows
  • Zoom the page to 200% and check nothing breaks or gets cut off

Our current process

Automated scanning on every deploy, a full manual pass before any major release, and a screen-reader walkthrough of any newly built interactive component before it ships to a client.

Automated tools catch the easy 30 percent

Axe, Lighthouse and WAVE will reliably flag missing alt text, insufficient color contrast, and unlabeled form fields - the mechanical violations that a rule engine can check without understanding what the page is for. That's a real and worthwhile 30 percent or so of accessibility issues, and running one of them in CI on every pull request catches regressions before they ship. But treating a clean automated scan as "done" is the single most common accessibility mistake we see in client codebases we inherit.

npx @axe-core/cli https://example.com --exit

The other 70 percent needs a keyboard and a screen reader

Tab through your own interface with the mouse unplugged. Does focus land somewhere visible at every step, in an order that makes sense? Then turn on VoiceOver (Mac) or NVDA (Windows, free) and listen to the same flow - a modal that traps focus correctly for a sighted keyboard user can still announce itself as an unlabeled group to a screen reader. No automated tool catches that gap, because it requires understanding intent, not just markup structure.