Blog

  • Blog Home
  • Google’s Shared AI Training Control: What to Do Now
Google’s Shared AI Training Control: What to Do Now

November 12, 2026

10 min Read

Google’s Shared AI Training Control: What to Do Now

Google's Shared AI Training Control: What to Do Now

In This Article:

  • What Google announced on October 30, 2026
  • What stays the same in your robots.txt today
  • A short checklist to prepare your site

Quick answer: Google says it plans to support a shared AI training control in robots.txt, so one preference could cover several AI providers. It is not live yet. Google aims to support it in 2027, and the existing Google-Extended control keeps working in the meantime. Nothing in the announcement requires an immediate change, but it is a good moment to read your file and decide your position.

On October 30, 2026, the Google Search Central Blog announced that Google plans to support a shared AI training control in robots.txt. If you run a website, the practical question is whether this changes what you should do this week.

Mostly it does not, but it does change how you should think about the file. The announcement is a plan, not a launch, and several details are still open. Here is what Google said, what it did not say, and how to prepare without guessing at wording that is not final.

What Google Announced

The announcement, written by John Mueller, describes a way for a website to state its AI training preference once, instead of managing a separate instruction for each AI developer. It builds on the work of the IETF's AI Preferences (AIPREF) working group, which is developing standard ways to express preferences about how content is collected and processed for AI.

The key points, summarised:

  • Site owners would use a shared "policy token" in robots.txt to set a broad preference across participating AI providers.
  • If a site opts out of training with that token and does not explicitly specify Google-Extended, Google plans to treat it as disallowing Google-Extended.
  • Existing company-specific controls, including Google-Extended, continue to work alongside the new signal.
  • Google aims to support the new control in 2027, together with other AI providers.

Google's crawler documentation adds one detail: Google intends to support the AIPREF training preference signal, using the user-agent value "__usage_ai-training", in early 2027.

Not live yet. This is a planned control, and Google calls the proposal a draft. Wait for Google's official documentation before adding any new line to your robots.txt, and do not assume the wording above is final.

What Is Still Unclear

Google's post and other providers' announcements say different things, so it helps to keep them apart.

  • What Google's post says. Google aims to support the control in 2027, and its crawler documentation narrows that to early 2027. Neither gives a launch date, so Google's support is planned, not live. The post refers to "participating" AI providers but does not name them.
  • What Microsoft says. In its Bing announcement of October 30, 2026, Microsoft says it will begin honoring the new preferences as it crawls on February 1, 2027. Sites that use NoArchive or NoCache should publish the new preferences before March 1, 2027.
  • What OpenAI says. In its announcement of the same date, OpenAI says it will respect the shared robots.txt signal and aims to implement support in the coming months. It says it will share details at that point and that its existing robots.txt controls remain in force.
  • What is not final. The IETF text is an Internet-Draft, and Microsoft notes that the working group is still moving toward a final standard. Google's post adds that the rest of the AIPREF vocabulary is still under discussion, so the shared control covers AI training only for now.
The controls that matter today and the one that is coming
Control Status What it does, per Google
Google-Extended in robots.txt Available now Lets publishers manage whether content Google crawls may be used to train future Gemini models and for grounding in Gemini Apps. It does not affect inclusion in Google Search.
Shared AI training control Announced; Google aims for 2027, Microsoft says February 1, 2027, OpenAI says coming months Meant to express one training preference across participating AI providers. At Google, an opt-out would be read as disallowing Google-Extended if that token is not explicitly specified.
Company-specific rules Available now Each AI company's own controls, which keep working alongside the new signal.

What Stays the Same Today

Three basics do not change with the announcement, and they are worth checking now.

Location and scope. A robots.txt file must sit at the root of the site host it applies to, such as example.com/robots.txt. Google's guide to creating the file explains that it applies only to the protocol, host, and port where it is posted, so a subdomain needs its own file if you want different rules there.

It is a request, not a lock. Google's introduction to robots.txt says the file is not a mechanism for keeping a page out of Google and cannot enforce crawler behavior. Reputable crawlers follow it, while others might not. Material that must stay private belongs behind a login, not behind a robots.txt rule.

Google-Extended is separate from Search. Google states that it does not impact a site's inclusion in Google Search and is not used as a ranking signal.

A Decision Worth Making Before 2027

The new control will not make the decision for you. It will only make the decision easier to express, so it helps to know your position first.

Three separate things often get blurred together, so keep them apart when you decide.

  • AI training. Whether your content may be used to train future generative models. This is what the shared control is designed to express.
  • Google Search inclusion and ranking. Google says Google-Extended does not affect inclusion in Google Search and is not a ranking signal. Microsoft likewise says its new preferences do not change eligibility for traditional Bing results.
  • Gemini grounding. Whether content Google has crawled may inform answers at the moment someone asks, in Gemini Apps and in Grounding with Google Search on Vertex AI. As Google documents it, Google-Extended covers grounding as well as training. Google's post says sites that explicitly allow Google-Extended may appear in Gemini app answers and be clicked, which is a possibility, not a guarantee.

Google's planned fallback is conditional. It applies only when a site opts out of training with the shared token and Google-Extended is not explicitly specified. In that case Google plans to treat the site as disallowing Google-Extended. Because Google-Extended also covers grounding, check whether that outcome suits you. Google's post does not spell out the grounding effect. If your file already names Google-Extended, that explicit rule is the more precise tool, so decide on it deliberately rather than by default.

Three positions are common:

  • Stay open. Add no AI-related rules and accept each provider's defaults. This suits sites comfortable with both training and grounding use. It does not improve rankings or guarantee citations or traffic.
  • Opt out broadly. Refuse training across providers once the shared control is available, and accept that at Google the fallback may also cover Gemini grounding. This suits sites with original, licensed or paid content they want kept out of both uses. Microsoft warns that refusing grounding can reduce brand visibility in AI experiences.
  • Mix the two. Refuse training broadly, then make exceptions only where a provider's documented controls allow them. Microsoft documents a separate page-level grounding setting, with Allow, DisallowExtended and Disallow options. Google documents Google-Extended as one token covering both training and grounding, so it does not offer that split. Where a provider documents no separate control, no exception is available.

Provider-specific exceptions depend on what each provider actually offers, so read its documentation before relying on one. Microsoft also notes that it must crawl a page to discover a setting.

A Practical Checklist for This Month

  • Read the live file. Open your domain followed by /robots.txt in a browser, note what is there, and record the date.
  • Keep one record per host. The www version, the non-www version and any subdomain can each have a different file. For each host, log: the current Google-Extended rule (absent, Allow or Disallow, exactly as written; absent means no explicit rule today); your training preference (allow or refuse); your grounding preference (allow or refuse, noting that Google-Extended covers both); a one-line reason; and a next review date.
  • Find out how the file is produced. WordPress outputs a default robots.txt through its do_robots() function, which plugins can modify through the robots_txt filter. Always read the live file rather than assume.
  • Move private material behind a login. A robots.txt rule cannot protect it.
  • Test on a staging copy first. Where your hosting offers one, try changes there, then re-read the live file afterwards.

Early 2027 is a sensible default review date for each host, or sooner if Google publishes updated documentation. If your pages use NoArchive or NoCache, review them before Microsoft's March 1, 2027 deadline.

Frequently Asked Questions

Is the shared AI training control available now?

Not at Google. Google calls it a plan and aims to support it in 2027, and its crawler documentation says early 2027. Microsoft says it will begin honoring the new preferences on February 1, 2027, and OpenAI says it aims to implement support in the coming months. Until then, Google-Extended remains the control available for Google's AI training.

Will blocking Google-Extended hurt my Google Search ranking?

Google says Google-Extended does not impact a site's inclusion in Google Search and is not used as a ranking signal. The rule concerns AI use, not ordinary search results.

Should I add the __usage_ai-training line now?

Wait for Google. It says it intends to support that signal in early 2027 and describes the proposal as a draft, so a line added today would not be a supported Google control, and the final wording could change. Microsoft documents the same token for Bing from February 1, 2027, so check that documentation if Bing matters to you.

Will every AI company follow the shared signal?

Not automatically. Google's post names no other providers and says only that others can treat the token in a similar way. Microsoft and OpenAI have published their own announcements, with different timing, and the IETF text is still an Internet-Draft. Compliance with robots.txt is voluntary, so treat the signal as a way to state your preference, not a guarantee.

Key takeaway: Google's shared AI training control is a 2027 plan, not a setting you can use today. Use the time to read your robots.txt host by host and decide whether to stay open, opt out broadly or mix the two, remembering that training, Google Search inclusion and Gemini grounding are separate questions, and that Google-Extended covers both training and grounding.

Deciding who may reuse your content is one half of the job, and being able to change your site safely is the other. Z.com's WordPress Hosting lists a staging site and free automated backups among its features, which gives you a place to try changes before they reach visitors. Explore the plans if you are setting up or moving a WordPress site.

Explore WordPress Hosting