AI Sitemap Guide: What the File Does and Does Not Prove
An AI sitemap is a publisher-defined JSON index. It can give a known consumer richer content records, but the filename alone does not make it a standard or prove that an AI service uses it.
An AI sitemap is a publisher-defined JSON index of a website's content. It can give a known consumer more detail than an XML sitemap, but /ai/sitemap.json is not a universal web standard and its presence does not prove that an AI service reads or rewards it.
That limit does not make the file useless. It tells you how to use it honestly: define the contract, keep the records current, name the consumer when one exists, and measure retrieval separately from results.
AI Sitemap and XML Sitemap Have Different Jobs
| File | Status | Useful job | What it does not prove |
|---|---|---|---|
sitemap.xml | Established search discovery format | Lists canonical URLs and update hints for supporting crawlers | That every submitted URL will be indexed |
/ai/sitemap.json | Publisher or protocol defined | Groups richer content records for consumers that understand its schema | That OpenAI, Google, or another service uses the file |
Keep the XML sitemap even when you publish a JSON index. The XML file has documented support from major search engines. The custom JSON file has only the support that its contract and consumers establish.
What AItoAIWeb.com's AI Sitemap Contains
AItoAIWeb.com publishes its AI sitemap as one part of the Digital Karma Federation v8.2 contract. Each record carries a canonical URL, content type, title, description, update date, and publisher priority.
A shortened record looks like this:
{ "protocol_version": "8.2", "site": "https://www.aitoaiweb.com/", "pages": [{ "canonical_url": "https://www.aitoaiweb.com/what-is-the-ai-web/", "type": "page", "title": "What Is the AI Web?", "priority": "high" }] }The field names belong to this protocol. Another publisher can choose a different structure. A consumer must know which contract it is reading before those fields have dependable meaning.
The Manifest, Catalog, and Receipt Complete the Contract
The AI sitemap is not asked to do every job.
- The manifest declares the site, protocol version, endpoint locations, and portfolio relationships.
- The catalog publishes Schema.org DataCatalog and Dataset records for the site's real content families.
- The AI sitemap lists individual canonical content records in a predictable JSON shape.
- The rendered receipt records whether every sitemap URL exposed the required Dataset graph when checked.
- The LLM JSON file provides a compact generated guide built from the final manifest.
This is a first-party publishing system. It is useful because the files agree with one another and can be tested. It is not proof that an outside AI product adopted the protocol.
Build an AI Sitemap Only When It Has a Job
A custom JSON index makes sense when you control both sides of an integration, join a documented protocol, publish a reusable data product, or want one stable contract for your own sites and tools.
If no consumer is known, treat the file as an experiment. Keep it public, versioned, and measurable. Do not move established crawler controls or page-level structured data out of the way to make room for it.
AI Sitemap Quality Checks
- Use one canonical origin. Every record should point to the same preferred host and protocol.
- Generate from the live content inventory. Retired pages should leave the file when they leave the XML sitemap.
- Publish a protocol version. A consumer needs to know which field contract applies.
- Use stable content types. Pages, articles, glossary terms, and directory listings should not change names between runs.
- Validate the response. Return valid JSON with the correct content type and no private data.
- Test the final output. Compare counts, URLs, dates, and identities with the rendered website.
Measure Retrieval Without Inventing Adoption
A server log can show that a crawler requested the AI sitemap and received HTTP 200. That proves retrieval. It does not show that the service used the data in an answer, ranking, recommendation, or citation.
Record the requester, date, status code, protocol version, and response hash. Then look for a separate result such as a verified integration, referral visit, or reproducible citation. Keeping those stages separate turns the endpoint into useful data instead of a marketing claim.
For the larger distinction, read AI Endpoints Explained. For the short definition, use the AI Sitemap glossary entry.