What Google announced, and what it did not
On October 30, 2026, Google Search Central announced that it plans to support a shared AI training control in robots.txt. The control is based on work from the IETF AIPREF working group. According to the post by John Mueller, Google Search Advocate, site owners will be able to put one shared “policy token” in robots.txt to set broad AI training preferences across participating AI providers. Today they have to write separate instructions for each AI developer. Google says it aims to support the control, along with other AI providers, in 2027. Nothing is live yet, and Google calls the mechanism a “draft proposal.”
The detail merchants need is how Google will read the token. Google states that if a site uses the token to opt out of training and Google-Extended is not explicitly specified, Google plans to treat that as if the publisher had disallowed Google-Extended. The same post states that sites which specifically allow Google-Extended are eligible to appear in responses and receive clicks via grounding in Gemini Apps. Our read is that the shared token is a training switch. It does not let a brand say “don't train on my catalog” while keeping “do cite my products.” For Google, a training opt-out on its own may take Gemini Apps grounding eligibility with it.
How the shared token maps onto Google-Extended
Google describes Google-Extended as the AI training control it has offered since 2023. It only works for Google: it tells Google, and nobody else, what a site prefers. The new token is meant to be read by any participating provider. Google states that existing company-specific controls, including Google-Extended, will keep working alongside the joint signal, so site owners can still write different instructions for each AI developer.
| robots.txt state | Today | Once Google supports the token (2027 target) |
|---|---|---|
| Explicit Google-Extended allow | Eligible for Gemini Apps grounding, per Google | Google-Extended keeps working, per Google |
| Explicit Google-Extended disallow | Disallowed | Google-Extended keeps working, per Google |
| Shared token opt-out, no Google-Extended rule | Not applicable; token not supported | Google plans to read it as a Google-Extended disallow |
| Shared token opt-out plus an explicit Google-Extended rule | Not applicable | Not covered in the post |
The condition “and Google-Extended is not explicitly specified” suggests, on our reading, that an explicit Google-Extended rule wins over the shared token for Google. Google does not state a precedence rule in those terms. It also does not say what happens when the two signals disagree.
Where the visibility exposure is, and where it is not
It is easy to read this announcement as “one wrong line could cost you AI visibility everywhere.” The source supports a narrower claim.
The only visibility surface the post links to Google-Extended is grounding in Gemini Apps. Google states that sites which specifically allow Google-Extended are eligible to appear in responses and receive clicks there. The post does not state the reverse, that a disallow removes that eligibility. The wording points that way, but that is our inference, not something Google says.
The post says nothing about Googlebot, Google Search, AI Overviews or AI Mode. Nothing in this source backs the claim that the shared token affects product citations on those surfaces. If you track AI Overviews and AI Mode visibility per product, this post gives you no reason to expect those numbers to move.
For Google, a shared training opt-out with no explicit Google-Extended rule is read as a Google-Extended disallow. The token alone does not separate training from Gemini Apps grounding.
What it means for ecommerce brands
When merchants think about training opt-outs, they usually mean catalog content: product descriptions, buying guides, size charts, editorial collections. One cross-provider line will appeal to brand and legal teams who want to protect that copy. The practical problem is that the line now has a Google-specific cost that the token's name does not tell you about.
In our view, three groups are most exposed:
- Brands that use Gemini as a discovery channel. Google ties eligibility for grounding and clicks in Gemini Apps to allowing Google-Extended. A token opt-out on its own works against that. Know which of your products Gemini cites before you make the call.
- Agencies that manage robots.txt for many storefronts. Suppose one templated opt-out is pushed to every client. Under Google's stated rule, it becomes a Google-Extended disallow on every site that has no explicit Google-Extended rule.
- Stores whose robots.txt is set by an app or a platform setting. If a third-party tool adds the token for you, the Google-Extended result depends on whether that tool also writes an explicit Google-Extended rule.
The trade-off is real. On our reading of the post, a brand that wants to stay out of training and stay eligible for Gemini Apps grounding has one route with Google: pair the opt-out with an explicit Google-Extended allow. But Google itself calls Google-Extended its AI training control. Allowing it therefore means accepting Google's training use, which partly cancels the opt-out for Google. The token does not give a training-only exit where Google is concerned.
Open questions Google has not answered
The post leaves several practical questions open. Any guidance that claims to answer them today is speculation.
- Syntax. The exact name and syntax of the shared policy token are not published.
- Providers. The post refers only to “participating AI providers” and says other providers “can handle it similarly.” It names none.
- Date. Google gives 2027 as the target but no month or day.
- Scope. It is unclear whether the token covers only training or can express other AIPREF preferences. Google says the rest of the AIPREF vocabulary is still under multistakeholder discussion.
- Conflicts. The post does not say how Google resolves a shared token and a Google-Extended rule that disagree.
- Path scoping. It is unknown whether the token can be limited to specific paths, such as product pages versus blog content, the way Disallow rules can.
- Search surfaces. The post does not say whether a token opt-out affects Google Search, AI Overviews or AI Mode.
- Draft status. The post gives no IETF draft identifier and does not quote the IETF definition of “AI training” that Google says it will adopt.
What to do this week
- Don't edit robots.txt for this yet. The token syntax is not published and support is targeted for 2027. Any line you write for this mechanism today would be a guess.
- Check your current Google-Extended status. Open your robots.txt, look for a
User-agent: Google-Extendedgroup, and note whether it is allowed, disallowed or absent. Absent is “not explicitly specified,” which is exactly the case Google's new rule covers. Our guide to robots.txt for AI crawlers covers the wider set of user-agents. - Make the Gemini decision on purpose. Decide with your ecommerce, SEO and legal owners whether Gemini Apps grounding eligibility matters to your brand. Write the decision down so the 2027 token choice isn't made by default.
- Ask whoever controls robots.txt. That might be an agency, a developer or a platform app. Ask how they plan to handle the shared token and whether they will keep an explicit Google-Extended rule in place.
- Watch the two sources that matter. Follow the Google Search Central Blog and the IETF AIPREF working group for the syntax, the list of participating providers and the support date.
- Record a baseline now. If you change robots.txt in 2027, you will want before-and-after numbers. Product-level AI tracking shows which products Gemini and the other engines cite today.
The strategic read
This announcement is more about governance than technology. Google presents the proposal as a response to publishers, civil society groups, regulators and other stakeholders who have called for industry-wide controls on AI training. It also says the proposal is consistent with the EU Code of Practice for General Purpose AI Model Providers under the EU AI Act, which, according to Google, recognizes robots.txt as a primary way for publishers to express machine-readable preferences. A shared token gives AI providers one standard signal they can point to.
For merchants, our read is that the convenience of one line cuts both ways. Robots.txt files have been picking up AI crawler rules one vendor at a time, as the list of AI engines grew. A shared token replaces that sprawl, but one edit can now have different effects at each provider, and you have to look each one up. Google has published how it will read the token. No other provider has, and the post names none.
The posture that will hold up is to make the training decision and the visibility decision separately, then write the visibility decision as explicit per-provider rules for any provider whose surface drives revenue. Until the IETF syntax and the list of participating providers are public, treat this as planning for 2027, not a configuration change for this quarter.
Frequently asked questions
Is Google's shared AI training token in robots.txt live yet?
No. Google announced plans on October 30, 2026 and calls the mechanism a draft proposal. It aims to support the control, along with other AI providers, in 2027, and has not given a specific date.
How will Google read a shared token opt-out if I have no Google-Extended rule?
Google states that if a site uses the shared policy token to opt out of training and Google-Extended is not explicitly specified, it plans to treat that as if the publisher had disallowed Google-Extended.
Does the shared training token affect Google Search, AI Overviews or AI Mode?
Google's post does not say. The only visibility surface it links to Google-Extended is grounding in Gemini Apps. It does not mention Googlebot, Google Search, AI Overviews or AI Mode.
Will Google-Extended keep working after the shared token launches?
Yes, according to Google. The post states that company-specific controls, including Google-Extended, will keep working alongside the new joint signal, so site owners can still give individual AI developers different instructions.
What is the exact robots.txt syntax for the shared policy token?
Google has not published it. The token name, syntax and list of participating providers are still open, so merchants should wait for the IETF AIPREF syntax before editing robots.txt.
Ask AI about this
Ask an AI engine how Google's planned shared training token interacts with Google-Extended and Gemini Apps grounding.