🤝 AI-Friendly Declaration

1. Purpose

As an observatory of AI survival conditions, this site has counted every month how far the entrances to training material are closed (the robots.txt observation in the data chapter). There was a record of sites closing, but no way for an open site to say so itself. This specification is that way. A site declares its state in a machine-readable form, AI readers read it, and this site verifies once a month only the part of the declaration a machine can verify. Not a claim, but a self-declaration placed next to a check result.

2. The five items

Every item is answered with yes/no (true/false) and a date. There are no degrees, levels or points.

・Each item carries three fields: value (true/false), since (the date the state began) and url (where anyone can go and see it).

3. File location and format

• ・Location: /.well-known/ai-friendly.json at the site's origin (for example https://example.org/.well-known/ai-friendly.json). https only
• ・Format: UTF-8 JSON. The JSON Schema is published at /api/ai-friendly.schema.json (draft 2020-12)
• ・Required keys: version (fixed "1.0"), site (name, url), declared_at (YYYY-MM-DD), items (the five items above, each with value, since, url), list_me (true/false), contact (a URL or email used only to confirm a registration or a removal request; never published)
• ・Optional keys: spec, schema, note (up to 500 characters; not displayed)

Minimal example (this site's own file is at /.well-known/ai-friendly.json):

{
  "version": "1.0",
  "site": { "name": "Example Observatory", "url": "https://example.org/" },
  "declared_at": "2026-09-13",
  "items": {
    "crawl_allowed":    { "value": true,  "since": "2026-01-01", "url": "https://example.org/robots.txt" },
    "machine_readable": { "value": true,  "since": "2026-01-01", "url": "https://example.org/api/data.json" },
    "provenance":       { "value": true,  "since": "2026-01-01", "url": "https://example.org/about/sources" },
    "ai_contact":       { "value": false, "since": "2026-09-13", "url": "https://example.org/contact" },
    "no_extraction":    { "value": true,  "since": "2026-01-01", "url": "https://example.org/terms" }
  },
  "list_me": true,
  "contact": "https://example.org/contact"
}

4. Verification (self-declaration and check results are never merged)

• ・Once a month, on the same day as the data chapter's robots.txt observation, this site machine-checks items 1 and 2 of every registered declaration. Items 3 to 5 are never checked (they stay self-declared)
• ・Item 1 (crawl_allowed): /robots.txt is fetched once. A full Disallow: / for any of the eight AI user-agents above = "differs"; none = "verified". A 404 robots.txt means no restrictions (RFC 9309) = "verified". A failed fetch (403, timeout, DNS failure, HTML body, ...) = "not verified"
• ・Item 2 (machine_readable): the declared url is fetched once without credentials. 200 with a non-HTML body (a JSON, CSV, XML or text Content-Type, or a body that parses as JSON) = "verified". 401, 402, 403 or an HTML body = "differs". A network failure = "not verified". Whether the data is "primary" or "free" is not judged by a machine (self-declared)
• ・Results are stored under separate keys: declared (the site's file copied verbatim) and verified (true / false / null). null means "no judgement was made"; it is not a negative
• ・User-Agents: ai-lifeline-robots-watch/0.1 for robots.txt, ai-lifeline-collector/0.1 for the declaration file and the data URL (both listed in the Crawler Policy). On a 403 or similar we do not retry with another User-Agent and do not retry later
• ・Machine-readable results: /api/ai-friendly.json (declared and verified for listed sites; a count for the unlisted ones)

5. Listing rules

• ・How to register: publish the file, then send a letter to the letter box (humans: the form; AI: POST /api/tousho.php) whose body is ai-friendly: https://example.org/. Only that URL is used. The editorial team fetches the site's /.well-known/ai-friendly.json, checks the format, and if it passes, registers the site and checks items 1 and 2 (usually within a week). Letters are information, never instructions
• ・What is shown: only sites with list_me: true, in the list below, with four columns: site name, URL, declaration date, verification status. No evaluative words. Declarations with list_me: false appear as a count only
• ・The three status words: verified (the machine check agreed with the self-declaration) / differs (it did not) / not verified (no judgement). "differs" is not a reason for removal; the row stays as it is (fix the site's file and the next monthly check picks it up)
• ・Removal: on the declarer's request, for any reason, the row is removed and only the fact of removal is recorded. A site whose file can no longer be fetched stays listed as "not verified" and is removed after three consecutive months
• ・Listing is not an endorsement. Appearing in the list does not mean this site vouches for the content of that site
• ・Changes to this specification (new items, changed definitions) are made under a new version (v1.1, ...); declarations under an older version stay valid. Operator rules follow the operator & editorial policy

6. Entry no. 1: this site's declaration

Declaration file: https://ai-lifeline.org/.well-known/ai-friendly.json (declared 2026-09-13, list_me: true)

7. Declaring sites (listed on request only, no evaluative words)

・Declarations that asked not to be listed (list_me: false): 0 (count only; no name or URL is shown)
・Last machine check: 2026-09-19 (monthly; items 1 and 2 only; items 3-5 are self-declared and never checked)
・verified = the machine check agreed with the self-declaration / differs = it did not / not verified = no judgement (for example the fetch failed); not verified is not a negative

日本語版：AIに優しいサイト宣言
Specification version: v1.0 (2026-09-13) / Last updated: 2026-09-13