Duke Accessible-U+35.8 pp
Known accessibility training page
AccessAudit recall92.9%
axe-core recall57.1%
F1 +23.6 ppcoverage +85%
AI-assisted accessibility evaluation
Understand and fix accessibility issues faster.
AccessAudit combines automated WCAG scanning, visual evidence, AI explanations and guided verification so teams can detect, understand and resolve accessibility problems with more confidence.
- WCAG 2.0, 2.1 and 2.2 AA mapping
- Visual markers and list-only issue handling
- AI guidance with human verification workflow
Same scan pipeline as authenticated reports
Run the full audit inside the app
Full scans are created after sign in and use the same worker, rule set, screenshot capture and scoring model as dashboard reports.
Current stagePublic validation prototype
Core methodaxe-core + custom rules + AI guidance
Benchmark typePreliminary criterion-level evaluation
Preliminary benchmark evidence
Measured against pages with known accessibility issues
These results are not a final claim of superiority. They are an early validation layer built from pages with known or manually verified issues, used to compare AccessAudit with an axe-core baseline.
Deque Mars Demo+23.0 pp
Accessibility demo with known issues
AccessAudit recall61.5%
axe-core recall38.5%
F1 +16.7 ppcoverage +69%
W3C BAD Demo+11.1 pp
Before/after WAI demo website
AccessAudit recall38.9%
axe-core recall27.8%
F1 +10.2 ppcoverage +88%
Why this matters
Read methodology The benchmark layer is designed to answer a practical research question: how many known accessibility problems are detected by the tool, and how many are missed or incorrectly reported.
Product workflow
Detection is only the first layer
The platform is designed around a practical accessibility workflow: detect the issue, locate it visually, explain it clearly, verify uncertain cases and produce a report.
Hybrid WCAG detection
axe-core is used as a baseline and extended with custom checks for focus visibility, reflow, text spacing, touch targets, landmarks and other WCAG areas.
Visual evidence layer
Reports include a full-page screenshot with approximate markers, list-only findings and grouped issue explanations when multiple rules affect the same element.
AI explanations and fixes
Each finding can be explained in plain language with user impact, likely cause and a practical remediation direction for developers.
Guided verification snippets
For findings that depend on browser state, focus or interaction, AccessAudit can provide a DevTools snippet that helps classify the issue more consistently.
Reports for clients and teams
Generate shareable reports, PDF exports, executive summaries and issue lists that map findings to WCAG criteria and severity.
Monitoring and re-scan workflow
Track scan history, compare changes over time and move findings through a practical remediation workflow.
Evidence pipeline
From scan result to decision
Accessibility work is rarely solved by a single score. AccessAudit keeps each finding connected to evidence, explanation and review status.
Scan
Enter a URL and run automated WCAG checks.
Locate
Review issue list, visual markers and affected elements.
Understand
Use plain-language explanations and WCAG mapping.
Verify
Use guided checks where automation is uncertain.
Report
Export a clear report for teams, clients or reviewers.
Research direction
Built as a product, evaluated as a research platform
AccessAudit is being developed as a usable SaaS product and as a platform for evaluating AI-assisted accessibility workflows. The focus is measurable user support, not only raw issue counts.
Open research pageDetection value
How much additional verified WCAG coverage can a hybrid scanner provide compared with an axe-core baseline?
Remediation support
Do AI explanations and fix suggestions help users understand and resolve accessibility issues faster and more accurately?
Human verification
Can guided verification reduce uncertainty for findings that require browser state, keyboard focus or manual judgement?
AI transparency
AI helps explain findings. It does not certify compliance.
The platform separates deterministic scanning, AI-assisted interpretation and human verification so each layer can be reviewed more clearly.
What may be sent to AI
Relevant issue metadata can include rule ID, WCAG criterion, selector, short HTML snippet, element text, nearby context and screenshot crops for selected visual checks.
What AI is used for
AI explains findings, drafts remediation guidance and supports selected semantic checks such as whether existing alt text appears meaningful.
What AI does not decide
AI does not certify legal compliance, replace expert review or make final conformance decisions for complex user flows.
What users should avoid
Users should not intentionally submit passwords, API keys, private customer data or sensitive internal content into scans or AI prompts.
Model usage
Where AI is used in the workflow
Model use is intentionally scoped. Primary detection comes from deterministic rules, while AI supports explanations, draft remediation and selected semantic review.
Vision-assisted alt text assessment
Used for selected image-based checks where text alone is not enough. Missing alt attributes are detected by rules; poor or misleading alt text requires interpretation.
Explanations and fix suggestions
Used to generate plain-language explanations, affected-user impact, draft fixes, WCAG context and developer caveats.
Primary detection layer
Most findings come from axe-core and AccessAudit custom rules. AI is an assistance layer, not the main scanner.
Roadmap
Clear stage, clear next steps
The tool is live now, but the research and product validation are still evolving. This is the public roadmap for the current direction.
1
Available now
Current: Research prototype in production
Single-page scans, multi-page scanning, visual overlay, AI explanations, guided checks, reports and preliminary benchmark validation.
2
In progress
Validation: Better evidence and user metrics
Larger benchmark set, expert-confirmed ground truth input, per-finding feedback and measurement of how useful AI guidance is in real workflows.
3
Planned
Product: Team-ready accessibility workflow
Team spaces, client portals, CI checks, monitoring improvements, API workflows and clear paid packages after validation.
Access and pricing
Available during validation
AccessAudit is currently open to users while the product and evaluation methodology are being validated. Paid packages will be introduced later with clear notice before billing is enabled.
FAQ
Important limitations, clearly stated
Is AccessAudit a legal certification tool?
No. It is a technical evaluation and reporting tool. Automated checks can find many barriers, but final WCAG conformance still requires human review.
Why compare with axe-core?
axe-core is a widely used automated accessibility baseline. AccessAudit uses it as one layer and adds additional checks, visual evidence, AI explanations and guided verification.
Are the benchmark results final?
No. Current reports are preliminary and based on pages with known or manually verified issues. They are useful as proof of concept, not as a final scientific claim.
Does AI detect accessibility issues?
AI is not the main detector. axe-core and custom rules detect most technical WCAG issues, while AI helps with explanations, fix suggestions and selected semantic checks.
Can I use the tool now?
Yes. The platform is currently open to users during validation. Pricing will be introduced later with clear notice before billing is enabled.