# OpenAI, Google and Microsoft Back One robots.txt Signal for AI Training Opt-Outs: What It Covers, What It Does Not, and the Dates That Matter - Published: 1 November 2026 - Updated: 1 November 2026 - Author: RenderBob team - Category: News - Canonical: https://renderbob.ai/blog/shared-robots-txt-ai-training-opt-out-signal On 30 October, OpenAI, Google and Microsoft each said they will honour a single robots.txt instruction for opting a site out of generative AI training. Most of it is a plan, not a switch you can flip today. On 30 October, OpenAI, Google and Microsoft each said they will honour a single robots.txt instruction for opting a site out of generative AI training. Reports say Anthropic has started implementing it too. Most of it is a plan, not a switch you can flip today. ## What was announced PPC Land and Xenospectrum, both on or about 30 October, report the following. We could not open the companies' own posts, so this is secondary. - OpenAI says it aims to support the signal "in the coming months" and calls its participation voluntary. It says the signal will not replace its existing robots.txt controls. - Google aims to support it in 2027. For sites that do not set Google-Extended, Google will treat the shared training opt-out as a Google-Extended opt-out. - Microsoft says Bing will honour it from 1 February 2027. - Anthropic. PPC Land says the author of the IETF draft behind the signal, whom it identifies as working at Anthropic, told the working group's mailing list that Anthropic had started implementing it. No date is given. ## How the signal works The draft, an individual submission to the IETF's AI Preferences work, uses a policy token in place of a crawler name. The example reported is: > User-Agent: __usage_ai-training. Disallow: /. Allow: /press-kit/. Matching is exact apart from case. A wildcard group (User-Agent: *) does not count as a preference, and a file that never names the token expresses none. Existing tokens such as Google-Extended keep their current meaning. The draft has no IETF endorsement and, per PPC Land, expires on 3 May 2027. ## What it covers and what it does not - Covers use of content to train generative models, at site or path level. - Does not block crawling. It is a use preference, not a crawl directive, so Googlebot and Bingbot are unaffected. - Does not reach content already collected. Preferences attach to content when it is retrieved. - Is not a licence or a payment term, and a private licensing agreement may override it. - Is not enforced. Nothing in the draft makes a provider comply. - Does not cover answer-time use in general. The reports describe training and "grounding", meaning AI referencing the web to write an answer, as separate things. Microsoft adds its own grounding controls. Google offers no separate grounding-only option, and PPC Land's reading is that opting out with the shared token alone may also remove a site from Gemini grounding. That reading is PPC Land's, not Google's. ## Microsoft's dates - 15 December 2026: temporary answer-use settings open in Bing Webmaster Tools for sites that use NoArchive or NoCache. - 1 February 2027: the new preferences start being honoured as pages are crawled. - 1 March 2027: Bing stops reading NoArchive and NoCache as training or answer-use signals. - 30 September 2027: the temporary settings expire. Microsoft also adds page-level settings through a meta tag or header that override robots.txt, with an AI Training setting and a three-value AI Grounding setting (Allow, Disallow and DisallowExtended). Microsoft warns that "neither choice guarantees inclusion, traffic, or payment." ## Who is missing The four commitments do not name Meta, Amazon or Apple. That matters because of the Meta terms we covered on 30 October: a platform contract and a robots.txt preference are different instruments, and one does not stand in for the other. ## What studios and creators can do - Decide your position first. Opting out of training and being quoted in AI answers are separate choices. - Check the site's robots.txt and see what it says now about crawlers you already name. - Add the token when you are ready, knowing OpenAI and Google have not yet said it is live. - Keep unreleased material behind authentication, since a preference is not access control. - Note the Bing dates if you rely on NoArchive or NoCache. - Keep a dated copy of your robots.txt, because preferences attach at retrieval time. ## What to watch Details from OpenAI on how its signal interacts with existing controls, Google's start date and how it treats paths, whether Meta, Amazon or Apple adopt the token, and what the IETF working group does with the draft. ## RenderBob 3 and this story RenderBob 3 keeps references and prompt logs with the studio and lets you choose local generation per clip, so what you decide about training and web visibility does not have to depend on where a shot was made.