AIPREF: The IETF Standard for Declaring AI Usage Preferences

AIPREF is an IETF working group standardising how a publisher declares the terms on which content may be used for AI. It exists because the current mechanism — a growing set of vendor-specific robots.txt tokens like `GPTBot` and `Google-Extended` — requires every publisher to track every vendor's product names, and gives no way to state an intent that applies to companies that do not exist yet. The work splits into two drafts: - **`draft-ietf-aipref-vocab`** defines the vocabulary. Two categories: `train-ai`, covering use of an asset in producing or refining a generative model, and `search`, covering use in an application whose primary purpose is to locate assets and direct users to them. Each takes a single-character value, `y` to allow or `n` to disallow. A category that is absent is simply unspecified rather than denied. - **`draft-ietf-aipref-attach`** defines how the expression is attached: a `Content-Usage` HTTP response header, and a `Content-Usage` rule inside a robots.txt group. In both forms the expression looks the same, for example `train-ai=y, search=n` — permit training, refuse search indexing. The separation matters. Splitting training from search lets a publisher take the position that most news organisations actually hold — remain findable, do not become training data — which vendor tokens only recently began to support and only inconsistently. Status, stated precisely: the attachment draft carries a 2026 milestone to send a standards-track specification to the IESG, and the vocabulary draft still contains sections explicitly marked as not reflecting working-group consensus. It is a live standards effort rather than a deployed one. Publishing a `Content-Usage` line today is inert for crawlers that do not implement it, and costs one line if it lands. Compare llms.txt: The Proposed AI-Readable Site Summary, and Its Adoption Gap for a convention that did not go through a standards body, and Really Simple Licensing (RSL): Machine-Readable AI Licensing Terms for the licensing-and-payment counterpart. Overview: The Four Layers of Machine-Readable AI Site Policy.