Google, Microsoft, OpenAI and Anthropic have committed to respecting a shared signal that lets website owners refuse the collection of content for AI training. Currently, owners must set separate rules for each company. The proposal would let them communicate their preferences to several companies through a single setting.
The signal goes in robots.txt, a file that gives instructions to automated programs visiting websites. Owners can express preferences for an entire site or selected sections, allowing some material while excluding the rest. The proposal remains a draft and has not received formal approval from the Internet Engineering Task Force, or IETF.
Anthropic has begun work on supporting the signal. Microsoft says it will honour the setting from 1 February 2027, while Google plans to introduce support in early 2027. OpenAI intends to implement it in the coming months.
The signal expresses a preference about training rather than a general instruction to stop all automated visits. It does not remove pages from ordinary Google or Bing search results, and it cannot reverse earlier collection of content. Its effect therefore concerns future collection for training, not material that companies already hold.
