In short. The IETF AIPREF working group develops content usage preference specifications, with vocabulary and HTTP attachment documents currently at Internet-Draft stage.

Definition: The IETF AI Preferences working group (AIPREF) develops technical building blocks for expressing preferences about content collection and processing for AI model development, deployment and use. Its charter covers a shared vocabulary and ways to associate preferences with content.

Key facts

The IETF's AIPREF charter includes a method for reconciling multiple expressions of preferences.
As of 17 October 2026, the vocabulary document's Datatracker record lists revision draft-ietf-aipref-vocab-08 as an active Internet-Draft with intended RFC status Proposed Standard.
The attachment document's Datatracker record likewise lists draft-ietf-aipref-attach-05 as an active Internet-Draft with intended RFC status Proposed Standard.
The attachment draft proposes a Content-Usage HTTP header field and a Content-Usage rule for robots.txt, the file used to communicate crawler access rules.

What does the charter cover?

The approved charter separates vocabulary from attachment: the vocabulary describes preferences independently of the mechanism that associates them with content. Attachment can use content metadata or the protocol delivering the content, including HTTP response headers and the Robots Exclusion Protocol.

The charter also calls for coordination with bodies responsible for other formats, including WHATWG or W3C for HTML. It excludes technical enforcement, client or crawler authentication and authorization protocols, preference registries, and auditing or transparency measures for AI training.

What is the status of the drafts?

An Internet-Draft is a working document. The vocabulary draft dated 14 September 2026 explicitly says its contents lack working group consensus and serve as a basis for further discussion. Its status text says Internet-Drafts can be updated, replaced or obsoleted at any time.

The Datatracker records distinguish intended publication status from current document type. Both the vocabulary record and the attachment record show the working group state as “WG Document”. Proposed Standard is their intended RFC status; both documents remain work in progress.

What would a website preference look like?

The attachment draft dated 19 August 2026 gives this robots.txt example:

User-Agent: *
Allow: /
Content-Usage: train-ai=n

In this proposed syntax, train-ai identifies the AI Training usage category and n expresses disallow. The vocabulary draft maps permission to y and disallowance to n. A category without an explicit preference has the value unknown.

The attachment draft also shows Content-Usage: train-ai=n in an HTTP response. The proposed header associates preferences with the message's content. For a website operator, these are draft mechanisms for communicating a usage preference. The vocabulary draft leaves the recipient's choice about following preferences outside its scope.

FAQ

Does AIPREF provide a way to enforce my preferences?

The charter explicitly excludes technical enforcement. Its remit concerns expressing and associating preferences with content.

Where can I follow the technical discussion?

The vocabulary draft directs you to the working group's ai-control@ietf.org mailing list. It also identifies github.com/ietf-wg-aipref/drafts as the source repository and issue tracker.

Sources

Check how a page reads to answer engines. Open the GEO score analyzer