Google, Microsoft and OpenAI Will Honor One AI-Training Opt-Out. Microsoft Also Splits Training From Answers

On October 30, 2026, Google and Microsoft said they will honor one shared robots.txt line for opting out of AI training, and OpenAI said it will too. The line does not itself say anything about being cited in answers. Microsoft is the one that wrote a separate setting for that, with dates attached.

What is established

The shared line. Google's Search Central post, by John Mueller, announces plans to support a shared AI training control in robots.txt, based on work at the IETF's AI Preferences working group. A site writes a policy token as if it were a crawler name:

User-Agent: __usage_ai-training
Disallow: /

The syntax is the one in the Internet-Draft by Kelley, whose listed affiliation is Anthropic. It is a working draft that expires 3 May 2027, and the IETF's own note says drafts should be cited only as work in progress.

Google. It aims to support the control in 2027. If a site opts out with the token and does not mention Google-Extended, Google plans to treat that as if Google-Extended were disallowed. Existing controls keep working alongside it.

OpenAI. Its post says it will honor the signal, aims to implement it in the coming months, and that its existing crawler controls will continue to apply.

Microsoft. Bing's post, by Krishna Madhavan, says Microsoft joins the shared training preference and adds two things: the same training preference at page level, and a separate AI Grounding preference, set page by page only. There is no site-wide robots.txt setting for grounding. Grounding is when content is used as input to help inform a generated answer; Microsoft says indexing and ranking in ordinary search is not grounding.

Microsoft grounding setting	What it does
Allow (default)	Content can help inform generated answers
DisallowExtended	Limits grounding to Microsoft's own experiences such as Copilot and Microsoft 365, and excludes grounding services Microsoft provides to third parties
Disallow	Content is not used to help inform generated answers

Microsoft's own caution: choosing DisallowExtended or Disallow will significantly reduce your content's visibility in AI-generated experiences, without affecting eligibility for Bing search results. It also says neither choice guarantees inclusion, traffic or payment.

The part that is easy to get wrong

Training and answers are two decisions at Microsoft, and the post says so: your training setting does not change your grounding setting, and vice versa. A site can disallow training and still be eligible to inform answers.

Google's post does not give an equivalent. It says only that sites that specifically allow Google-Extended are eligible to appear in responses and receive clicks via grounding in Gemini Apps. It does not say what happens to a site that disallows Google-Extended, and it says the shared token would be read as that disallow. Until Google says more, adding the token is a training decision with an unstated answer-eligibility edge at Google.

Both settings are about use, not crawl access. Bing says an Allow does not override a robots.txt rule blocking Bingbot, and Microsoft must crawl a page to see the setting on it.

Dates

Date	What happens
December 15, 2026	Temporary grounding preferences open in Bing Webmaster Tools, only for sites using NoArchive or NoCache as of that date
February 1, 2027	Microsoft starts honoring the new training and grounding preferences as it crawls
March 1, 2027	After this date NoArchive and NoCache are no longer read as training or grounding preferences
September 30, 2027	Temporary Bing Webmaster Tools grounding preferences expire

Google says 2027 and OpenAI says the coming months; neither gave a date.

What to do

• List every AI-related rule in robots.txt (Google-Extended, GPTBot, ClaudeBot and others) and who added it. Keep them: the companies say their own controls continue to apply.
• Decide training and answer eligibility as two questions. For Bing, the answer question is a page-level tag or header, not a robots.txt line.
• If any page uses NoArchive or NoCache, put December 15 and March 1 in the diary. Pages that use neither need no transition step, according to the post.
• Do not block a search crawler to express a training preference. Blocking OAI-SearchBot costs ChatGPT search eligibility, which is a separate question from training.

What is not known

Whether adding the shared token, or setting Bing's grounding preference, changes how often a site is cited is not measured here, and Microsoft's visibility warning is its statement, not a Paralax finding. Google has not said whether opting out of Google-Extended affects Gemini Apps answers. Anthropic's own implementation plan was not found in the sources read. The draft can still change before it becomes a standard. The Machine Relations Index does not record any site's robots.txt, so it cannot show an effect.

Evidence scope. Google's, Microsoft's and OpenAI's posts and the IETF draft, read November 1, 2026. The AI crawlers and robots change log and the Copilot and Bing change log keep the dated record.