# Google, OpenAI and Microsoft back a shared way for publishers to opt out of AI training

Published 2026-11-05T09:02:18.000Z · AI Pulse · https://ai-pulse.news/policy-law/publishers-ai-training-opt-out-robots-txt

*This story was written with the help of artificial intelligence, checked automatically, then reviewed by an editor before publication, under our AI policy : https://ai-pulse.news/ai-policy*

Google, OpenAI and Microsoft have agreed on a shared process that lets publishers state whether their content may be used for AI training, using robots.txt, with Google announcing its support on 30 October 2026 and Microsoft committing to honour preferences from 1 February 2027.

## Key points

- Google, OpenAI and Microsoft agreed to make it easier for publishers to refuse the use of their content for AI training, according to Siècle Digital (2 November 2026).
- The mechanism relies on robots.txt and a shared identifier, __usage_ai-training, within IETF work.
- It does not technically block AI crawlers: effectiveness depends on companies respecting the stated preferences.
- Reported start dates differ: Microsoft from 1 February 2027, Google in early 2027, OpenAI in the coming months; Meta has not confirmed adoption.

## What the companies have agreed

Google, OpenAI and Microsoft have agreed to let website publishers refuse more easily the use of their content to train AI systems, according to [Siècle Digital](https://siecledigital.fr/2026/11/02/google-openai-et-microsoft-promettent-de-ne-plus-utiliser-certains-contenus-pour-entrainer-leurs-ia/). Google announced on 30 October 2026 that it intends to support a common process through which publishers can express their preferences on AI training.

The arrangement builds on the robots.txt file and incorporates a shared identifier, __usage_ai-training, within work carried out at the IETF.

## How the signal works

An Internet-Draft, draft-kelley-aipref-training-preference-signal-00, dated 30 October 2026 and signed by Kevin Kelley of Anthropic, describes how an operator can link usage preferences (for example "no AI training") to paths. It does so through robots.txt groups indexed by a policy token. The draft says it *"introduces no new directive names and no new grammar"*.

The draft also states that it is not a crawl directive, that it is not endorsed by the IETF and that it has no formal status in the standardisation process. The shared preference is therefore a proposal under discussion, not a finished standard.

## Microsoft's version: AI Training and AI Grounding

In an October 2026 announcement on its [Bing blog](https://blogs.bing.com/search/2026/10/your-content-your-choice-clearer-preferences-for-training-and-grounding/), Microsoft presented new preferences covering training and "grounding" for generative AI. AI Training can be set in robots.txt or per page. AI Grounding is set at page level. Microsoft lists the options Allow, DisallowExtended and Disallow.

According to Microsoft, it will honour the new preferences as it crawls from 1 February 2027, and the NoArchive and NoCache signals will no longer be read as a preference after 1 March 2027. Microsoft says the AI Training preference is shared with other industry players and draws on an effort by the IETF AI Preferences working group, which has not yet produced a final standard.

## Limits and open questions

According to Siècle Digital, the system does not technically block AI robots. Its effectiveness depends on companies respecting the preferences publishers express.

Timelines differ by company, per the same report. Microsoft has committed to apply the preferences from 1 February 2027. OpenAI plans to do so in the coming months, and Google in early 2027. Anthropic has begun implementing the mechanism, while Meta has not confirmed whether it will adopt it.

The available material does not say how compliance would be verified, nor whether the IETF draft will become a formal standard. Because the draft carries no formal status, the final technical form of the shared signal remains open.

**Reported timing for applying publishers' AI training preferences** (data as of 2026-11-02)

| Company | Reported position |
| --- | --- |
| Microsoft | Committed to apply from 1 February 2027 |
| Google | Early 2027 |
| OpenAI | Planned in the coming months |
| Anthropic | Has begun implementing the mechanism |
| Meta | Adoption not confirmed |

## FAQ

### Does the new mechanism block AI crawlers?

No. According to Siècle Digital, it does not technically block AI robots; it depends on companies respecting publishers' preferences.

### When will Microsoft honour the preferences?

Microsoft says it will honour the new preferences as it crawls from 1 February 2027.

### Is the approach an official IETF standard?

No. The Internet-Draft is not endorsed by the IETF and has no formal status in the standardisation process.

### Has Meta adopted it?

Meta has not confirmed its adoption, according to Siècle Digital.

## Sources

1. [Google, OpenAI et Microsoft s’accordent pour permettre aux sites web de refuser l’utilisation de leurs contenus par les IA](https://siecledigital.fr/2026/11/02/google-openai-et-microsoft-promettent-de-ne-plus-utiliser-certains-contenus-pour-entrainer-leurs-ia/), Siècle Digital, 2026-11-02
2. [https://datatracker.ietf.org/doc/draft-kelley-aipref-training-preference-signal/](https://datatracker.ietf.org/doc/draft-kelley-aipref-training-preference-signal/), datatracker.ietf.org, 2026-10-30 (primary source)
3. [https://blogs.bing.com/search/2026/10/your-content-your-choice-clearer-preferences-for-training-and-grounding/](https://blogs.bing.com/search/2026/10/your-content-your-choice-clearer-preferences-for-training-and-grounding/), blogs.bing.com, 2026-10-01 (primary source)
