Skip to main content

Accessibility Testing With AI

5 min read
Ux Eng

Ux Eng

AI can automate checks. It can't replicate the experience of a blind user. Combine both.

Accessibility Testing With AI

TL;DR

  • AI and automation can catch many a11y issues: missing labels, contrast, semantic structure. They can't catch everything.
  • Use automated tools for regression and coverage. Use human testing (including assistive tech) for the rest.
  • AI can suggest fixes. Verify them. Some fixes are wrong or incomplete.

Accessibility is non-negotiable. It's also time-consuming. AI-powered tools promise to speed up testing and fixing. They deliver — for a subset of issues. Your job: use automation where it helps, and know where it falls short.

What Automated and AI Tools Catch

Structural and semantic:

  • Missing alt text, empty links, improper heading hierarchy, form labels. Rule-based tools (axe, Lighthouse) and AI-assisted tools catch these well.
  • Run them in CI. Fix what they find. Don't stop there.

Color and contrast:

  • WCAG contrast ratios. Automated tools can measure. AI can suggest color adjustments.
  • Verify. "Suggest a darker shade" might break your design system. Check in context.

Keyboard and focus:

  • Some tools can simulate keyboard nav and flag focus issues. Not all. And they can't tell you if the order makes sense to a user.
  • Manual keyboard test. Always.

What They Miss

Context and meaning:

  • Alt text can be "present" but wrong. "Image of chart" vs. "Bar chart showing Q3 revenue up 15%." AI might suggest generic alt. You need descriptive alt. Human judgment.

Complex interactions:

  • Modals, drag-and-drop, custom widgets — automation often fails here. Screen reader testing by a human (or with real assistive tech) is irreplaceable.
  • Test with VoiceOver, NVDA, or similar. Regularly.

Cognitive accessibility:

  • Clear language, predictable flow, not overwhelming. No tool measures this well. You do.
  • Read your UI. Could someone with cognitive differences use it? Revise.

Real-world usage:

  • Automation checks technical compliance. It doesn't replicate "can a blind user complete this task?" You need user testing. Preferably with people who use assistive tech.

The Workflow

  1. Automate the basics — axe, Lighthouse, or similar. Run on every PR. Fix blockers.
  2. AI-assisted fixes — When a tool flags an issue, AI can suggest a fix. Review before applying. Some suggestions are wrong.
  3. Manual keyboard test — Tab through. Can you do everything? Is focus order logical?
  4. Assistive tech test — At least for critical flows. Screen reader, magnification if relevant.
  5. User testing — When possible, include people with disabilities. They'll find issues no tool can.

Integrating Into Your Process

  • Shift left. Catch a11y issues in design and development, not QA. AI can help designers check contrast and structure early.
  • Document exceptions. Sometimes you have a valid reason to break a rule. Document it. Don't let automation "fix" things that would make the experience worse.
  • Training. Your team needs a11y literacy. AI can help with docs and examples. You own the standards.

Manual process. Repetitive tasks. Limited scale.

Click "With AI" to see the difference →

Quick Check

What remains human when AI automates more of this role?

Do This Next

  1. Run your main flow through axe and a screen reader — List what automation caught vs. what you found manually. Fill the gaps.
  2. Add a11y to your component checklist — Every new component: keyboard, labels, contrast. AI can help verify. You own the bar.
  3. Schedule one assistive tech session — Use VoiceOver or NVDA for 30 minutes. Navigate your product. Fix what's broken.