Table of Contents
The AI opt-out rule could make it easier for website owners to control whether their content is used to train generative AI models. Google, OpenAI, Microsoft and Anthropic have agreed to support a shared signal that publishers can add through their existing robots.txt files.
Announced on October 30, 2026, the proposal aims to reduce the need for publishers to create separate AI training restrictions for each participating company. The shared policy token is being developed through the Internet Engineering Task Force (IETF).
How the AI Opt-Out Rule Works
The proposal introduces a shared identifier called __usage_ai-training. Website administrators can add it to their robots.txt configuration and use Disallow instructions to prevent specified content from being used for AI training.
For example, publishers could restrict their entire website or selected directories. They could also use Allow rules to make particular sections, such as a press resources folder, available for training.
The signal is specifically designed for AI training preferences. It is not a general crawling restriction and does not directly prevent pages from appearing in conventional search results.
The proposal is still part of the IETF standards process, rather than a completed universal standard. Its effectiveness will depend on adoption and implementation by individual providers.
Google, OpenAI, Microsoft and Anthropic Set Different Timelines
All four companies have committed to supporting the shared signal, but their implementation schedules differ.
| Company | Announced implementation plan |
| Anthropic | Has started supporting the directive |
| Microsoft | Plans to honour the preferences from February 1, 2027 |
| Targeting early 2027 | |
| OpenAI | Plans implementation in the coming months |
OpenAI has described the initiative as voluntary and clarified that the shared signal will not replace its existing robots.txt controls.
Google’s John Mueller has also explained that the shared token will allow websites to set broader preferences across participating AI providers. Google has indicated that it will interpret the training opt-out as a Google-Extended block when a website does not separately specify Google-Extended.
This is important because Google-Extended also relates to the use of content in certain AI experiences. Publishers should therefore review their existing instructions before making changes.
Meta has expressed support for a common industry approach, but has not confirmed participation in this specific signal.
Microsoft Introduces Separate AI Grounding Controls
Microsoft is also expanding the controls available to website publishers through Bing. Its approach separates AI training from AI grounding, which involves using website information to generate responses through AI-powered services.
One option, DisallowExtended, allows publishers to support Microsoft’s own AI experiences while restricting the use of their content by third-party grounding services. Separate preferences are available for managing AI training and grounding.
Microsoft has outlined the following transition dates for websites using older NoArchive or NoCache tags:
| Date | Planned change |
| December 15, 2026 | Temporary grounding settings become available in Bing Webmaster Tools for eligible sites |
| February 1, 2027 | Microsoft begins honouring the new AI training and grounding preferences |
| March 1, 2027 | Deadline for affected publishers to complete the required transition steps |
Microsoft has stated that these controls will not affect a website’s eligibility for regular Bing search results. Publishers should review Microsoft’s guidance to understand which transition steps apply to their websites.
What the AI Opt-Out Rule Means for Publishers
The AI opt-out rule could simplify how news organisations, independent blogs and other publishers communicate their content-use preferences.
Until now, website administrators have often needed to manage different crawler instructions for individual AI providers. A shared signal could reduce this work for participating companies.
However, the directive is voluntary. It does not guarantee that every AI company will respect the instruction, nor does it prevent all forms of access to website content. Publishers will still need to maintain relevant company-specific settings and check which providers have implemented the shared signal.
The distinction between AI training and grounding is also important. A website may want to prevent its content from being used to train models while still allowing it to appear in search results or support certain AI-generated answers.
Frequently Asked Questions
1. What is the AI opt-out rule?
It is a proposed shared robots.txt signal that allows websites to communicate whether their content can be used for AI model training.
2. Which companies support the shared AI training signal?
Google, OpenAI, Microsoft and Anthropic have announced plans to support it.
3. Does the AI opt-out rule block search engine crawling?
No. The shared token concerns AI training, not general crawling or conventional search indexing.
4. Can website owners choose which content AI companies can use?
Yes. The proposed Allow and Disallow instructions can be used to express different preferences for website sections.
5. Does the signal prevent AI grounding?
Not automatically. AI training and grounding are separate uses. Microsoft provides additional controls for managing grounding preferences.
Conclusion
The AI opt-out rule represents an effort to give website publishers a more consistent way to communicate their AI training preferences. Support from Google, OpenAI, Microsoft and Anthropic could simplify content management across participating providers.
However, different rollout schedules, voluntary adoption and separate grounding controls mean publishers will still need to monitor company-specific guidance as the proposal develops.