Anthropic, Google, Microsoft and OpenAI say they will honour a shared robots.txt signal for AI training opt-outs; the IETF draft behind it has no formal standing

Kevin Kelley of Anthropic published an Internet-Draft, draft-kelley-aipref-training-preference-signal-00, on 30 October 2026.^[1] It describes how a website operator can bind a preference such as “no AI training” to paths by writing ordinary robots.txt groups under a policy token, __usage_ai-training, in the User-Agent line.^[1] The draft’s example disallows every path except /press-kit/.^[1]

The Verge reported that OpenAI, Google, Microsoft and Anthropic say they will adopt the signal, and that today sites must state their AI training preferences separately to each company.^[2] The Verge reports that Anthropic has already adopted the signal, that Google and Microsoft plan to support it next year and that OpenAI says it will implement it in the “coming months”, while Thurrott reports that Google, Microsoft and OpenAI will do so in 2027.^[2][4] TNW reports Kelley wrote that Anthropic has already started work on it and quotes him: “It’ll take us some time to add support across our services”.^[5] TNW also reports Microsoft will honour it from 1 February 2027, and Search Engine Roundtable quotes a Google crawler-documentation note saying Google intends to support it in early 2027.^[3][5]

Krishna Madhavan, a principal product manager for Microsoft Bing, wrote: “Publishers should not have to choose between being discoverable in Bing and deciding how generative models use their work.”^[4][5] OpenAI’s announcement said: “This voluntary effort makes it easier for publishers and website operators to communicate the same preference to multiple participating providers at once.”^[4] Meta communications head Andy Stone said: “We believe creating a common signal that conveys such preferences to multiple providers at once is an important step.”^[2] The Verge reports he did not say whether Meta will honour it.^[2]

Google Search Advocate John Mueller wrote that when a site uses the token and does not explicitly specify Google-Extended, Google plans to treat it as if the publisher had disallowed Google-Extended, and that existing company-specific controls will keep working.^[3] TNW reports Microsoft said blocking AI grounding “will significantly reduce your content’s visibility in AI-generated experiences”, and The Verge reports Microsoft said sites will still appear in Bing search results whatever their preferences.^[2][5] TNW reports Microsoft also said the new preferences “do not establish licensing or payment arrangements”.^[5]

The draft says the token states a preference and is not a directive for crawling, that a site with no such group has expressed no preference, and that private agreements may override it.^[1] The IETF datatracker page says the document “is not endorsed by the IETF and has no formal standing in the IETF standards process”.^[6] Nate Hake, who says he has taken part in the working group, wrote on X that it is not finished at the IETF level, that significant areas of dispute remain, and that he is unhappy the current draft exempts the preferences from applying to anything a user does.^[3] Paul Thurrott wrote that the change is “largely symbolic”.^[4]