What it is
The AIPref effort investigates a standardized way to express preferences to AI systems (including crawlers and agents). It is in early discussion and not ready for production use.
Evidence trail
- Nov 30, 2025 Vocabulary draft updated
- Jul 31, 2025 Milestone for protocol specifications
Examples
HTTP response header draft example Source
Active AIPref attachment work shows AI preferences delivered directly in the HTTP response.
HTTP/1.1 200 OK
Content-Type: text/plain
Content-Usage: train-ai=n robots.txt draft example Source
The same attachment draft also sketches a domain-level robots.txt expression.
User-agent: *
Allow: /
Content-Usage: train-ai=n HTML meta-tag draft example Source
Earlier vocabulary work also explored per-page HTML metadata for finer-grained preferences.
<meta name="AI-PREF" content="allow_training=false; retention_period=0">