AI Overviews: what Google actually recommends, and what it ignores

Google answers more and more searches directly, with an AI summary above the classic results. The question clients ask us about it is always the same: do I now have to do something different?

Google's own short answer is no. The longer answer is more interesting, because in 2026 Google wrote down explicitly which of the currently popular measures do nothing for Google Search. Several of them have been recommended in trade articles for months.

Is SEO changing fundamentally?

No, and that is not a guess. Google's documentation on AI features states that the familiar SEO fundamentals still apply, because the generative features run on the same ranking and quality systems as ordinary Search.

To be eligible to appear in an AI answer, a page has to be indexed and allowed to show with a snippet in Search. Google names no additional technical requirements.

How to see whether you appear at all

This is the part most people skip, and it is measurable.

Appearances in AI features are counted in Search Console within the normal figures, under the Web search type. They are not broken out there.

Since June 2026 there is also a dedicated view for generative AI features. One thing matters about it: the metric it names is impressions. Clicks are not mentioned in Google's announcement. Anyone promising you an exact click count from AI answers did not get it from Search Console.

In practice: compare your totals over time and treat the AI view as an additional visibility indicator. A clean split of clicks by source does not exist.

What Google explicitly does not need

This is the interesting part for anyone currently receiving quotes for "GEO" or "AEO". Google's optimisation guide lists concretely what you can skip for Google Search.

llms.txt. This file has been widely recommended for a while, and in Germany alone it is searched for roughly 1,900 times a month. Google writes that Google Search does not use such files and ignores them. It does no harm, but for Google it does no good either. If you want to maintain one for other systems, fine. As an SEO measure it is not one.

Special markup for AI. There is no dedicated schema.org markup that gets you into AI answers. Structured data remains useful for rich results, but Google explicitly warns against overfocusing on it.

Chunking your content. Cutting text into pieces for AI systems is on Google's list of things to ignore for Google Search.

Rewriting text for AI. Also on that list.

Buying mentions. Google names inauthentic mentions explicitly as something you can save yourself.

None of this is a reason to ignore AI visibility. It is a reason not to spend money on measures the provider itself says do nothing at its end.

What you can actually control

If you want to influence how much of your content appears in previews and AI answers, you use the same controls as for ordinary snippets: nosnippet, data-nosnippet, max-snippet and noindex.

For nosnippet Google states explicitly that it applies to all forms of results, including AI Overviews and AI Mode, and additionally prevents the content being used directly there. max-snippet limits it in the same way.

Two pitfalls:

• These controls are blunt. Block use in AI answers and you usually lose the ordinary snippet too, and with it most of the clicks.
• A page blocked in robots.txt is never read, so your snippet rules there are never found.

Do not confuse this with Google-Extended. That is a separate token for controlling training and grounding of Gemini models. Google states that it does not affect inclusion in Google Search and is not a ranking signal. For AI Overviews in Search it is therefore not the lever it is often sold as.

What Google actually recommends

Google's own recommendations for AI search are unspectacular, and that is precisely the news:

1. Content with its own value, that does not already exist ten times over.
2. A good page experience, regardless of where the visitor arrives from.
3. Technical reachability: do not block Googlebot, the page works, the content is indexable.
4. Set preview controls deliberately rather than by accident.
5. Structured data that matches the visible content.
6. Think beyond text, so good images and video, and maintained entries in Merchant Center and your Business Profile.
7. Look at the full value of a visit instead of counting clicks alone.

Point seven changes the most in practice. If part of the answering now happens without a click, click count on its own is a worse measure than it used to be. Enquiries, calls and revenue from organic traffic say more.

One figure to treat carefully

Google writes that clicks from result pages with AI Overviews are of higher quality, because users are more likely to spend longer on the site afterwards.

That may well be true. Google simply publishes no numbers behind it, no sample and no period. It is a statement by the provider about its own product. If somebody uses that claim in a proposal as evidence, it is not evidence.

Conclusion

For Google Search, less changes about the work than the market currently suggests. What does change is the measurement: more visibility without a click, a separate impressions view since June 2026, and no clean click attribution per AI feature.

The money you would spend on llms.txt files and AI markup is demonstrably better spent on better content and a faster website. That is not just an agency writing it, that is Google.

If you want to know where your site stands technically before thinking about AI visibility, the website relaunch checklist is a good walk-through. And for local businesses, the Business Profile remains a more direct lever than any AI optimisation.

Sources

Read on 22 October 2026, all from Google's own documentation:

• AI features and your website
• AI optimization guide
• Succeeding in AI search
• Generative AI performance reports
• Robots meta tag specifications
• Google's common crawlers