IETF aipref sets AI content permissions
Content Signals is a reference site explaining the IETF's proposed "aipref" standard, which lets publishers declare structured preferences for AI training, search indexing, and real-time inference using HTTP headers and robots.txt. Cloudflare — the standard's co-author — has already deployed default Content-Signal headers across 3.8 million domains and ships a "Markdown for Agents" feature that cuts AI agent token consumption by ~80%.
This is one of the most important infrastructure proposals of the AI era — a machine-readable rights layer for the web that could finally give publishers a standardized way to separate "okay for RAG" from "okay for training."
- –The three-signal system (`search`, `ai-input`, `ai-train`) is more precise than robots.txt's binary allow/disallow — publishers can say yes to retrieval, no to model training
- –Cloudflare's 3.8M domain default rollout means aipref is already in the wild at scale before the IETF formally ratifies it
- –The Markdown-for-Agents feature (80% token reduction by serving clean Markdown instead of HTML) is the concrete cost win for developers building AI agents that browse the web
- –Claude Code and OpenCode already send `Accept: text/markdown` — adoption on the agent side is already happening
- –Critics (including Google's John Mueller) argue stripped Markdown loses context modern LLMs already handle — the debate is active and unresolved
DISCOVERED
74d ago
2026-03-15
PUBLISHED
74d ago
2026-03-15
RELEVANCE
AUTHOR
DIY Smart Code