Accessibility annotation is one of the most tedious tasks in the design workflow. Checking contrast ratios for every text-background combination. Annotating heading hierarchy for screen readers. Verifying touch targets meet the 44×44px minimum. Suggesting alt text for every image. Documenting ARIA labels for every interactive component. For a typical app with 50 screens, manual annotation takes 20-40 hours.
Figma's AI Design Agent, combined with the accessibility Skills launched in May 2026, automates the mechanical parts of this work. It can analyze your designs, check mathematical compliance (contrast ratios, touch target sizes), and generate annotation layers — cutting annotation time from hours to minutes per screen.
Key Takeaway
Figma AI doesn't replace accessibility expertise — it automates the tedious annotation work. The agent generates a first draft of accessibility specs that an a11y expert reviews and refines, cutting spec creation time from 20-40 hours (for a 50-screen app) to 2-4 hours of review and refinement. The mathematical checks (contrast, dimensions) are highly accurate; the contextual checks (alt text, focus order) need human review.
What Can Figma AI Check Automatically?
| Check Type | What the Agent Does | Accuracy | Human Review Needed? |
|---|---|---|---|
| Color contrast | Checks text/background pairs against WCAG AA/AAA | Very high (mathematical) | No (unless edge cases) |
| Touch target sizes | Flags interactive elements below 44×44px | Very high (dimensional) | No |
| Heading hierarchy | Annotates H1→H6 structure for screen readers | High (structural) | Quick verification |
| Alt text suggestions | Generates descriptive alt text for images | Medium (needs context) | Yes — review purpose |
| Focus order | Suggests keyboard tab order based on layout | Medium (layout-dependent) | Yes — verify logic |
| ARIA labels | Suggests ARIA roles and labels for components | Medium (contextual) | Yes — verify semantics |
| Font size minimums | Flags text below readable size thresholds | Very high | No |
| Color-only information | Identifies states communicated only by color | High | Quick verification |
The pattern: mathematical and dimensional checks are highly reliable and can run unsupervised. Contextual and semantic checks are starting points that need expert review. This is the right division: AI handles the tedious measurement work (which is most of the annotation time), humans handle the judgment calls (which require understanding user needs).
How to Run an Accessibility Check with the Agent
Step 1: Select the frame you want to check.
Step 2: Tell the agent: "Run an accessibility check on this screen. Check WCAG AA compliance for contrast ratios, verify touch targets are at least 44×44px, annotate the heading hierarchy, suggest alt text for images, and flag any information communicated only by color."
Step 3: The agent creates an annotation layer on your canvas with results: green checks for passing items, red flags for failures, orange warnings for items needing review.
Step 4: Review the medium-accuracy items: alt text suggestions, focus order, ARIA labels. Adjust based on your knowledge of the feature's purpose and user context.
For more consistent results, create a custom /accessibility-check Skill that encodes your organization's specific accessibility requirements (which may exceed standard WCAG). Your skill might include: "Our minimum font size is 14px (not WCAG's 12px)," or "All interactive elements must have a visible focus indicator using our brand blue outline."
📬 Getting value from this?
One actionable AI insight per week. Plus a free prompt pack when you subscribe.
Subscribe free →Frequently Asked Questions
Does AI-generated a11y spec replace accessibility testing?
No — AI-generated specs are a starting point, not a final audit. Real accessibility testing requires screen reader testing (VoiceOver, NVDA), keyboard navigation testing, and ideally user testing with people who have disabilities. The agent handles annotation; humans handle validation. The spec is the documentation layer; testing is the verification layer.
Which WCAG level does the agent check?
The contrast checks support both WCAG AA (4.5:1 ratio for normal text, 3:1 for large text) and AAA (7:1 ratio). Specify your target level in your prompt or Skill. Most organizations target AA; AAA is recommended for government and healthcare applications.
Can I create a custom accessibility Skill for my organization?
Yes — and you should if your organization has requirements beyond standard WCAG. Many companies have brand-specific accessibility guidelines: higher contrast ratios for brand colors, specific minimum font sizes, required focus indicator styles, and touch target sizes that exceed WCAG minimums. Encode these in a custom Skill so the agent checks your specific standards automatically.
How accurate is the AI's alt text?
Moderate — the agent describes what it visually recognizes in the image. But good alt text requires understanding the image's PURPOSE in context. A decorative hero image needs different alt text (or none) than an informational chart. A product photo needs descriptive text; a decorative background doesn't. Always review and refine AI-generated alt text based on the image's role in the user experience.
Does this work on existing designs or only new ones?
Both — and retroactive checks may be the highest-value use case. Point the agent at any existing frame and ask for an accessibility review. It analyzes what's already there and generates annotations. This is especially powerful for auditing screens designed before your team adopted accessibility standards, providing a quick baseline assessment that guides remediation priorities.
Disclosure: Some links in this article are affiliate links. We only recommend tools we've personally tested and use regularly. See our full disclosure policy.