axe® Advanced Rules
Advanced Rules is a powerful set of automated accessibility tests that runs alongside axe-core in the axe DevTools Extension and the axe MCP Server. These innovative rules help developers and testers save significant time by automatically identifying issues that previously required manual testing. By detecting accessibility problems earlier in the development process, advanced rules helps teams build more accessible products with less effort.
Advanced Rules is available to all axe DevTools Pro and axe DevTools HTML users.
Getting Started
From the start screen
To run advanced Rules from the start page, make sure you are logged in with a Pro account. From there, click either the "Full page Scan", or "Partial page scan" and select which element to test. This opens up the "Choose your scan type" modal. There you are given two options: You are given two options:
- "Standard scan" runs with axe-core, but without the advanced automated rules.
- "Advanced scan" runs both axe-core and the advanced automated rules.
The "Don't show again" checkbox will store the answer for future scans and update the Advanced rules setting to match.
From the settings
To enable Advanced Rules in the axe DevTools Extension:
- Navigate to Settings
- Enable the "Advanced Rules" option
After enabling Advanced Rules, you'll see them in your analysis results alongside standard axe-core rules:
Running Advanced Rules may increase scan time compared to standard axe-core tests. You can disable Advanced Rules at any time if you need faster scanning.
Try It Out
Ready to see Advanced Rules in action? Visit our workshop demo page and run the axe DevTools Extension to see examples of each advanced rule.
Overview of Advanced Rules
The initial Advanced Rules release includes four powerful rules:
- advanced/css-focus-visible: Check if CSS was used to hide focus indicators
- advanced/heading-markup: Find headings that are missing heading markup
- advanced/image-informative-has-alt: Test if informative images have a non-empty alternative text
- advanced/text-contrast: Check for complex text contrast issues
Additional rules are in development and will be added in the future. These Advanced Rules use innovative technologies not available to traditional axe-core tests, including:
- Screenshots and image analysis
- Large language models
- Computer vision models
- Chrome debugger protocols
- And more
These technologies allow us to automatically detect issues that otherwise would require running Intelligent Guided Testing (IGT) or have expert human review.
AI Credits
Some Advanced Rules use large language models (LLMs) for analysis, which incur costs. Organizations are provided with a monthly credit allocation, and additional credits can be purchased as needed. When credits are exhausted, the rules that use them will be automatically skipped.
The following rule uses AI credits:
image-informative-has-alt: Uses credits for AI-powered image analysis
For more information see AI credits.
Confidence Controls
New in Advanced Rules is the concept of confidence. The heading-markup and image-informative-has-alt rules use artificial intelligence (AI) for testing. These rules give a confidence rating for each issue they find, rather than a definitive pass or fail. To decide whether to report an issue, Advanced Rules applies a confidence threshold. This puts you as the tester in control. A high confidence setting makes it likely that everything that is reported a fail is truly an accessibility issue, but the rules are more likely to miss issues. A lower confidence makes it less likely you'll miss issues, but means you may see more false issues.
Confidence presets
The threshold is chosen from four presets:
| Preset | Reports findings with confidence of… | Trade-off |
|---|---|---|
| Precise (90%+ confidence) | 90% or higher | Fewest false positives, but misses the most real issues |
| Balanced (70%+ confidence) | 70% or higher | A middle ground between precision and coverage |
| Thorough (50%+ confidence) | 50% or higher | Finds the most issues, but reports the most false positives |
| Disabled | — | Advanced rules do not run; only the standard axe-core ruleset is used |
Balanced is the default. The same presets and thresholds apply in every product that runs Advanced Rules, so a scan at Thorough from the axe DevTools Extension and a scan at Thorough from the axe MCP Server apply the same bar.
Where the threshold is set
- Your organization's default is set by an administrator under Default Advanced Rules in axe Configuration. Administrators can also lock it so users cannot change it.
- axe DevTools Extension — when your administrator permits it, change the threshold in the extension's settings.
- axe MCP Server — when your administrator permits it, override the threshold per server or per scan. See Advanced Rules in the axe MCP Server documentation.
Test reliability
Because these rules use AI, there can be slight variations between runs on the same page. This is known as test reliability. For example, an issue reported with 75% confidence may on a subsequent scan have 65% confidence. This can then change whether that issue is reported as a violation or not too.
Limitations and Considerations
When using Advanced Rules, keep in mind:
- Results may vary slightly between different devices and environments. This happens because operating systems, graphics cards, even what monitor is used changes changes little details in the page
- Some rules use AI, which may produce slightly different results when run multiple times on the same content
- Advanced Rules complement (but don't replace) standard axe-core, IGT, and manual testing
- Scan times will be longer when Advanced Rules are enabled, usually 15 to 20 seconds
- Confidence thresholds can be adjusted to balance between finding more issues and reducing false positives
As always, we believe in being transparent about the capabilities and limitations of these technologies, and we're committed to putting accessibility professionals in control of how they're used.
Note: It is never necessary to use Deque's software or services in a way that collects, stores, or shares personal data. Do not use the software or services in a manner that collects, stores, or shares personal data.