The four companies backed a shared website signal to let publishers opt out of having their pages used for AI training.

On Oct. 30, 2026, OpenAI, Google, Microsoft and Anthropic backed an Internet Engineering Task Force draft for a shared AI-training preference in websites’ robots.txt files. The token, `__usage_ai-training`, lets publishers block training site-wide with `Disallow: /` or use `Allow` rules to exempt selected paths. Anthropic engineer Kevin Kelley authored the draft, “Machine-Readable Training Preference Signal,” which expires May 3, 2027. Adoption will be staggered: Anthropic says implementation has begun; OpenAI plans support “in the coming months”; Google targets 2027; Microsoft says Bing will honor the signal from Feb. 1, 2027. The rule concerns training, not crawling or search indexing, and remains voluntary—so it gives publishers a common preference, not a technical barrier or guarantee that every crawler complies. It could replace a patchwork of company-specific instructions, but publishers should retain existing controls until each provider implements it.