AI Manifests Coverage
stableCategory: ai-signals · Methodology v4.5
The scanner hunts for AI manifests along several discovery paths: - /.well-known/ai/manifest (or its .json variant) - /.well-known/ai-plugin.json or /ai-plugin.json - A <link rel="ai-plugin" href="...
Signal Source
- Source
https://{domain}/.well-known/ai/manifest (and homepage HTML, headers)- Kind
- http_response
Score Bands
| Verdict | Condition |
|---|---|
| Pass | 2 or more distinct valid AI manifest sources are detected, OR a .well-known AI manifest path (/.well-known/ai/manifest or /.well-known/ai-plugin.json) is present and non-empty |
| Partial | exactly one valid AI manifest source is detected |
| Fail | no valid AI manifest source is detected in the homepage HTML, headers, or prefetched well-known paths |
Description
AI Manifests Coverage checks whether your site publishes AI-specific manifest files that tell AI agents what it does and how to interact with it programmatically. friendly4AI scans your homepage HTML, response headers, and well-known paths for these manifests and scores the coverage from 0 to 100. One non-empty .well-known AI manifest is enough to pass.
What does this parameter check?
The scanner hunts for AI manifests along several discovery paths:
/.well-known/ai/manifest(or its.jsonvariant)/.well-known/ai-plugin.jsonor/ai-plugin.json- A
<link rel="ai-plugin" href="...">tag in the homepage<head> Link:response headers carrying anai-pluginorai-manifestrelation- A generic
/manifest.json— but only when its contents include AI-specific fields such asname_for_model,description_for_model, orschema_version
Why does it matter for AI-readiness?
A web page tells a human what you offer. An AI manifest tells an autonomous agent how to use it. Manifests at /.well-known/ai/manifest and ai-plugin.json describe your capabilities, API endpoints, and contact details in machine-readable form, so an agent can discover and invoke your services instead of guessing from rendered HTML.
More than one discovery path means wider reach. Some agents check the well-known directory, others read the <head> link, and a few follow Link: headers. The standard is young, so publishing one now gets your site ready for agent-driven workflows before they go mainstream. This signal pairs with the UCP manifest, llms.txt presence, and AI crawler access control parameters.
How is it scored?
This AI-Specific Signals parameter is graded across three tiers under the v4.5 methodology, and the scanner follows the published rubric exactly:
- Pass (100): two or more distinct valid manifest sources are found, OR a
.well-knownAI manifest path is present and non-empty. A single well-known file alone reaches full score. - Partial (50): exactly one valid source is detected.
- Fail (0): no valid source is found across the HTML, headers, and prefetched well-known paths.
A generic /manifest.json counts as a source only after it passes the AI-specific field check, so an ordinary PWA manifest earns nothing here.
How do you fix it?
- Publish a manifest at
/.well-known/ai/manifestthat describes your capabilities, API endpoints, and contact info. That one well-known file reaches the pass tier on its own. - Add a
<link rel="ai-plugin" href="/.well-known/ai-plugin.json">tag to your homepage<head>so agents have a second discovery path. - Follow the ChatGPT Plugin or MCP specification for the manifest structure. Include
name_for_model,description_for_model, andschema_versionso a generic/manifest.jsonreads as AI-specific. - Keep the file non-empty and valid JSON. An empty or blank well-known response does not count as a source.
- Re-scan after publishing to confirm the scanner detects your manifest sources.
Version History
- Introduced
- v1
- Last changed
- v4.5
Key takeaways
- Signal: https://{domain}/.well-known/ai/manifest (and homepage HTML, headers)
- Category: AI-Specific Signals
- Passes when: 2 or more distinct valid AI manifest sources are detected, OR a .well-known A…