Should You Block AI Training on Your Store Site?

Block it if you do not want your content training Google models, and know that you also block your pages from appearing in Gemini answers and earning clicks there. Google announced a shared control in robots.txt on October 30, 2026 that states this preference to several providers in one step. One decision, and it has a price.

Should you block AI training on your store site?

Blocking has a gain and a loss, and the loss is the less obvious one. Google says the new shared control states a training opt out to participating providers once, instead of configuring each provider separately. And if you use it to opt out without naming Google-Extended explicitly, Google treats that as disallowing Google-Extended itself.

This is where the loss sits. Google-Extended does not govern training alone. It also governs grounding, which means handing your pages to the model at prompt time so it builds its answer from them. Google says in the announcement that sites which explicitly allow Google-Extended are eligible to appear in those answers and receive clicks through them in Gemini Apps. So the opt out closes training and visibility together, not training alone.

The control is based on a draft from the AIPREF working group at the IETF, and John Mueller of Google announced it on October 30, 2026. Google crawler documentation gives its actual name, __usage_ai-training, and says Google intends to support it in early 2027. Company specific controls keep working alongside the shared signal, so you can opt out broadly and make one exception.

What is different about a Saudi account?

The difference is that this change reaches your Saudi store on the same day it reaches an American one. It is not a surface Google launches country by country. It is a line in a file on your own server. The pattern we have grown used to, where the new surface opens in English first and Arabic arrives later, does not apply here. I wrote about that pattern when Google added seven languages to AI Brief and Arabic was not among them.

A second difference is that whoever writes this file in a Gulf store is usually not the person answerable for the number. The file sits with a developer or with the hosting platform, and the choice inside it is a marketing choice, not a technical one. Someone who does not know the file holds a line closing a visibility channel will never ask anyone about it.

A third difference is that the regulatory pressure behind this control is European, not Gulf. Google says the proposal is consistent with the EU Code of Practice for General Purpose AI model providers under the EU AI Act. So the ready made template you copy from a foreign blog may be written for an obligation that does not bind you, and it costs you visibility that does.

When does this not work?

There is nothing you add to your file today. Google says it aims to support the control alongside other providers in 2027, and its documentation says early 2027. The shared control is not live now. Reading this news is a decision you write down before anyone asks you for it, not work you execute this week.

A second limit is that this opt out does not shield you from generated answers in Search. Google says in its crawler documentation that Google-Extended does not impact a site's inclusion in Google Search, and is not used as a ranking signal there. So anyone who blocked it believing it removes them from AI Overviews or AI Mode blocked one channel and stayed in the other.

A third limit is that the control states a preference and does not enforce one. It works with the providers participating in it, and Google says support expands as more players adopt it. A provider that has not adopted it is not bound by your line.

A fourth limit is that I hold no figure for the size of what you lose. Google publishes no estimate in this announcement for the number of clicks coming from grounding in Gemini Apps. Anyone telling you the opt out costs a specific share of your traffic is saying what they cannot source.

The test I run

Facing a Saudi store whose robots.txt has not been opened since launch, I add nothing and delete nothing. I open the file and read it. Then I ask who wrote it and why.

I read before writing because a training opt out also blocks grounding. Grounding means handing your pages to the model so it answers from them. Whoever blocked the first blocked the second and its clicks. They only meant the first.

Then I write the decision as one explicit line in the file. I record its date and who asked for it. Someone who leaves the file undecided finds an opt out there in 2027. Another person added it for a reason they do not know. They lose a visibility channel they never knew was open.

Frequently asked questions

Does blocking Google-Extended remove my site from Google results?

No. Google says in its crawler documentation that Google-Extended does not impact a site's inclusion in Google Search, and is not used as a ranking signal there. The opt out covers training and grounding in Gemini Apps, not indexing and not ranking.

When does the shared control take effect?

Not yet. Google says it aims to support it alongside other providers in 2027, and its crawler documentation says early 2027 and names it __usage_ai-training. So there is no live shared control today that you can add to your site file.

What do I gain by allowing Google-Extended?

Google says sites that explicitly allow it are eligible to appear in answers and receive clicks through grounding in Gemini Apps. That is the only gain named in the announcement, and Google publishes no figure for its size. The accounts I run are detailed on the work page.