DocsAccessAIPREF Content-Usage
AIPREF Content-Usage
The IETF's standards-track vocabulary for stating how content may be used by AI systems.
Why an agent cares
It is the one opt-out mechanism on a path to becoming an actual internet standard, which is what Content Signals and ai.txt are not. The half that defines the crawl-detectable surface is a Content-Usage HTTP structured-field dictionary plus a content-usage: robots.txt rule, and it updates RFC 9309.
Why we do not score it
attach draft reaches RFC. The vocab half advancing does not move that condition: only attach defines the header and robots.txt rule a crawl could read. Note what the draft does not give us yet: it states no MUST or SHOULD governing how a site expresses a preference, so today's defect surface is syntactic only: a malformed structured-field dictionary, including uppercase keys, which fail parsing. vocab-08 sections 6.4 and 6.5 make unknown labels and non-y/n values "not errors", and attach-05 section 3.2 reads a value without a leading / as having no path, so none of those is a defect.Adoption
Specification only. A published specification or draft. Deployments are limited to the authors' own demos.
Direction of travel: ▲ rising, as at .
Drafts only, with no deployment, but both halves are moving again in lockstep: attach reached -05 on 2026-08-18 and is an active WG document targeting Proposed Standard, having been lapsed at -04 since 2025-10-28, and vocab reached -07 the same day and -08 on 2026-09-14. Neither is in IESG processing yet.
Specifications
| Document | Revision | Kind |
|---|---|---|
| IETF AIPREF working group | Living document | guidance |
| AI Preferences: attaching usage preferences (IETF draft) | Draft, 19 August 2026 | specification |
| AI Preferences: the usage vocabulary (IETF draft) | Draft, 14 September 2026 | specification |
Last re-read against the published documents: .