Google documents how to control brand appearance in AI Overviews and AI Mode

Google Search Central published (and updated in December 2025) a page dedicated to explaining how AI Overviews and AI Mode handle third-party content within Search. The document doesn't introduce a new feature, but organizes something many SEO teams still confuse: appearing in AI-generated responses and allowing AI to use the content to train or ground responses are two different technical decisions, with different controls. Understanding this separation is the first step for any GEO (generative engine optimization) strategy.

How Search decides who becomes an AI source

According to the documentation, both AI Overviews and AI Mode use a technique called "query fanning": instead of answering only the user's literal question, the system fires off several related searches, across different subtopics and data sources, to assemble the final response. This means a page can become a source for a response even without ranking well for the original keyword, as long as it satisfies one of the derived queries the model generated internally.

AI Overviews appears only when Google's systems conclude it will add something to traditional Search, so not every query triggers the feature. AI Mode, in turn, is aimed at questions that would require multiple manual searches, such as comparisons or exploratory research. The documentation is direct about the technical requirement to be eligible for either one: the page needs to be indexed and meet the same technical requirements that already apply to appearing with a snippet in regular Search. There is no additional technical requirement specific to AI.

Google-Extended is not the same block as Googlebot

This is where the point that causes the most confusion in robots.txt audits lies. Googlebot is the crawler that decides whether a page is crawled and indexed for Search as a whole, AI included. Blocking Googlebot removes the site from everything, including traditional results. Google-Extended is a separate user-agent token, whose function is to specifically control whether the content can be used to train models and to ground responses from systems such as AI Overviews, AI Mode, and Gemini, without affecting normal indexing.

In practice, a site can allow Googlebot and block only Google-Extended:

User-agent: Google-Extended
Disallow: /

This block says: keep crawling me and showing me in normal results, but don't use my content as a source for training or grounding AI-generated responses. It's the opposite for those who want to appear more in AI responses: in that case, the path is simply not to block Google-Extended and to ensure Googlebot has unrestricted access, including on CDNs and hosting infrastructure, which the documentation explicitly cites as points of accidental blocking.

The controls, side by side

The documentation lists control mechanisms that operate at different layers. Mixing these controls without understanding the difference is the most common cause of a brand disappearing from the wrong place or remaining visible where it shouldn't.

The rest of the checklist is SEO that already existed

The documentation reinforces that, aside from the controls above, there is no optimization exclusive to AI. The recommended best practices are the same ones that already apply to traditional Search: ensuring that robots.txt and the entire CDN infrastructure allow crawling, keeping content navigable through internal links, offering a good page experience, presenting important content in text format (not just in image or video), and keeping structured data consistent with the text visible on the page.

This last point matters more than it seems: schema that describes something different from what the page shows the user is treated as a sign of low reliability, and this affects both the classic snippet and citation in AI responses.

The documentation also calls for attention to up-to-date Merchant Center and Business Profile information, when applicable, and recommends verifying the site in Search Console to quickly diagnose technical issues.

How to measure the effect in Search Console

Here's the most useful piece of data on the page for anyone doing technical SEO: sites that appear in AI Overviews and AI Mode are counted in the same overall traffic calculation in Search Console, reported under the "Web" search type in the Performance report. There's no filter dedicated solely to clicks coming from AI within the tool, which makes the analysis more indirect: the recommended approach is to track aggregated variations in impressions and clicks in the Performance report and cross-reference them with Google Analytics to measure time on site and conversion.

The documentation itself notes an observation about click quality: according to Google, clicks originating from results pages that contain AI Overviews tend to be of higher quality, in the sense that the user is more likely to spend more time on the destination site after clicking. This is a useful indication to justify investment in quotable content, but it doesn't replace direct measurement via Analytics: the claim is qualitative, with no published reference number.

When blocking comes at a cost

Blocking Google-Extended has a trade-off that the documentation doesn't hide, but also doesn't emphasize: the site stops being cited as a source in responses that, increasingly, replace the click on the traditional organic result. For brands that depend on discovery traffic, this can reduce overall visibility even while keeping classic ranking intact. It makes sense to block when the content is proprietary, subscription editorial, or when the brand has an explicit policy against use in AI training; it doesn't make sense to block by default on sites that depend on top-of-funnel SEO, where appearing cited in AI Overviews still generates the click from those who want to go deeper.

If the question is whether a control was applied correctly, the documentation recommends testing with Search Console's URL Inspection tool, looking at the HTML that Googlebot actually received, and giving Google time to recrawl the page: the timeframe can range from a few days to several months, depending on the crawl frequency Google has already calculated for that page. If the problem persists after that, the recommended channel is the Google Search Central Help Community.