Aria
axe-core · GPT-4o Vision · WCAG 2.2

Accessibility testing that sees like a human.

Two analysis engines run on every scan — automated WCAG rule checks plus GPT-4o visual reasoning over a real screenshot. Catch what scanners alone can't see.

Try:
See how it works
  • Powered by GPT-4o Vision · OpenAI
  • axe-core engine · Industry standard
  • WCAG 2.2 criteria · Full coverage
  • Real screenshot · Not a simulation
  • Two analysis layers · Rules + AI vision
  • Zero data stored · Scan, read, done

How it works

Two engines, one report

Paste a URL. Both engines run automatically and the results arrive together.

Automated WCAG Analysis

axe-core runs against the live DOM and flags structural violations — missing alt text, ARIA misuse, landmark issues, invalid semantics — mapped to exact WCAG 2.2 success criteria.

Visual AI Review

GPT-4o receives a full-page screenshot and reasons about what a human evaluator would notice: touch target sizing, visual hierarchy, color-only cues, and contrast over complex backgrounds.

Unified Intelligence

Both engines report to a single accessibility score with issues grouped by severity. Every finding includes a description, affected element, and a suggested fix.

Under the hood

What happens during a scan

Both engines run in parallel — results arrive in under 30 seconds.

Scan pipeline
  1. Navigating to URL
  2. Analyzing DOM structure
  3. Running WCAG rule checks
  4. Capturing full-page screenshot
  5. Sending to GPT-4o Vision
  6. Generating accessibility report

The report

Every scan delivers a full picture

A sample of what your accessibility report looks like — score, severity breakdown, and per-issue fix guidance.

Sample report · example.com

Accessibility Score: 62 / 100

26 issues found across 2 analysis engines

3

Critical

7

Serious

12

Moderate

4

Minor

Automated

Image missing alt attribute

7 <img> elements have no alt text — screen readers will skip them.

AI Vision

Insufficient touch target size

Primary CTA button is 28×28 px — WCAG 2.5.5 requires 44×44 px minimum.

Automated

Heading order skips level

<h1> followed directly by <h3> — assistive tech users lose document structure.

+ 23 more issues in the full report

AI Vision layer

AI sees what rules can't

A real screenshot is sent to GPT-4o — not a description, the actual pixels. It reasons about the visual experience the way a human evaluator would.

AI Vision finding

Touch target too small

Severity: Critical

The primary action button is 28×28 px. WCAG 2.5.5 requires at minimum 44×44 px for interactive controls.

Who is affected

Users with motor impairments, tremors, or those using touch devices with limited precision.

Suggested fix

Increase button dimensions to at least 44×44 px, or add min-h-[44px] min-w-[44px] with appropriate padding.

This kind of finding — measured from pixel dimensions in a real screenshot — cannot be produced by DOM-only rule-based scanners.

The gap

What rule-based scanners miss

Traditional scanners analyze the DOM. Aria analyzes the DOM and the visual experience.

CapabilityTraditionalAria
Structural HTML/ARIA violations
Color contrast failures
Missing alt text (DOM)
Keyboard focus order
Visual touch target sizing
Color-only information conveyance
Visual hierarchy problems
Cognitive complexity & layout issues
Contrast over gradients / images
Inconsistent interaction patterns

Beyond automated rules

What Aria finds that others miss

The barriers that require visual reasoning — the kind only a human evaluator or a vision model can catch.

  • Visual hierarchy problems

    Unclear reading order, buried CTAs, and structural confusion that disorients screen-magnifier and low-vision users — invisible to DOM scanners, visible in pixels.

  • Touch target sizing

    Interactive elements measured from their rendered pixel dimensions. Rules can't know if a button's padding makes it large enough; GPT-4o can.

  • Cognitive complexity

    Dense text walls, inconsistent layouts, and unpredictable patterns that raise the cognitive bar for users with ADHD, dyslexia, or processing differences.

  • Color as sole signal

    Error states, status indicators, and data encoded only in color. Caught visually — both the element and the surrounding context matter.

  • Contrast over backgrounds

    Text rendered over gradients, images, or layered colors where computed contrast ratios don't tell the full story. GPT-4o evaluates the actual composite.

  • Ambiguous interactive cues

    Hover-dependent affordances, non-obvious clickable areas, and missing focus indicators that a DOM walk can't surface without seeing the rendered output.

Find what your scanner misses.

Paste any URL — no account required. Results arrive in under 30 seconds, combining automated WCAG checks with AI visual analysis.