A controlled comparison

Does an accessibility skill change the result?

Send the same prompt to the same free AI model twice. Side A gets a compact generation guide derived from wcag-skill and its references; Side B gets no WCAG guide. Compare each generated page and run axe-core separately to measure the difference. Results vary by model and run.

Before you start: Your prompt is sent to AI Horde, a public volunteer-run inference service. Do not include passwords, personal information, or confidential content. This demo shows generated examples; it does not certify WCAG conformance.

Set up the comparison

Free models and queue estimates are loaded from AI Horde. No API key is required.

512 tokens

This sets the requested output cap, not a model capability. Both sides request a 2048-token context; prompts that do not fit are rejected rather than truncated. Anonymous requests above 512 output tokens may require kudos or be rejected. Larger output caps may not fit the context or the selected worker.

The prompt, selected model, context size, output-token cap, and other generation settings are identical for both sides. Only Side A receives the WCAG generation guide.

Loading AI Horde text models…

Generated webpages

Previews run in isolated frames. A model may take several minutes if its free queue is busy.

Side A · With WCAG skill

Ready when you are.

Page preview

Generated HTML

No result yet.

axe-core results

No test run yet.

Side B · Without WCAG skill

Ready when you are.

Page preview

Generated HTML

No result yet.

axe-core results

No test run yet.

Checking the latest shared stable version.

Generate both pages before running the accessibility tests.

The latest stable release available on all three CDNs is selected by default. Each selected version is checked on all three CDNs before testing. The test runs that release’s rules tagged WCAG 2.0 A/AA/AAA, WCAG 2.1 A/AA, WCAG 2.2 AA, and best practice; older releases may not include newer tags. axe-core does not provide a WCAG 2.2 AAA tag, and automated results are not a conformance claim. Method: UUStreak axe test.