The companies backed a shared website signal for publishers to tell AI crawlers not to use their content for model training; Anthropic said implementation had begun.
On October 30, 2026, OpenAI, Google, Microsoft and Anthropic backed a shared website signal for publishers who want to keep their content out of AI training. The Internet-Draft, proposed by Anthropic standards lead Kevin Kelley, adds the token `__usage_ai-training` to a site’s existing `robots.txt`; a `Disallow: /` rule can cover the whole site, with narrower rules for selected paths. Anthropic has begun implementation. OpenAI says it plans to support the signal in the coming months; Microsoft’s start date is February 1, 2027, while Google targets 2027. The proposal could spare publishers from maintaining separate training opt-outs for each company’s crawler, a practical shift as AI developers compete for web data and publishers seek control over reuse. The token concerns training, not ordinary search indexing, and cannot reverse earlier collection; compliance is voluntary. It builds on crawler-specific controls already used by publishers, but adoption will be uneven until each company rolls out support.