Google AI Overviews: how they work and how to get cited

AI Overviews are the summaries that appear above the usual results for some searches. Google launched them in the US in May 2024, growing out of its Search Generative Experience trial, and has since rolled them out to many more countries and languages. They appear for a share of queries, most often longer informational questions, and not at all for many navigational or commercial ones.

AI Mode is a separate, conversational way to search. It takes more complex questions, handles follow-ups, and reasons across more sources before answering. Google launched it in the US in 2025 and has expanded it since.

Both are built on Gemini models and both draw on Google's search index. Google describes the difference like this: AI Overviews help people get the gist of a topic quickly, while AI Mode suits questions that need more exploration, reasoning or comparison.

For a site owner, they matter for the same reason. The answer now sits above the blue links, and whether your page appears as a cited source can matter as much as where it ranks. Both features show links to their sources, as a list beside the answer and as inline links, and Google says they often show a wider range of sites than a traditional results page.

Google says both AI Overviews and AI Mode use a technique it calls query fan-out: issuing several related searches across subtopics and data sources, then building a response from what comes back.

Take a question like "best way to chase a late invoice as a freelancer in the UK". Behind the scenes that might become searches about statutory late payment interest, polite reminder wording, when to use a debt recovery service, and small claims court fees. Each sub-search has its own set of results, and a page can be cited for one of those sub-questions even if it ranks nowhere near the top for the original query.

That changes what to aim for:

• Cover the questions around your topic, not just the head term. A page that answers "how much interest can I charge on a late invoice" cleanly is a candidate for every broader question that fans out to it.
• Make passages self-contained. A paragraph that states the answer, with the specific figure, date or rule, is easier to lift than one that depends on the three before it.
• Be precise. Specific facts, named entities and clear definitions give the model something concrete to cite.

Citations still come from Google's index. There is no separate AI crawler for Search: Googlebot does the crawling, and ordinary indexing decides what is available to cite.

Google's guidance is plain: there are no extra technical requirements and no special optimisation for AI features. A page must be indexed and eligible to be shown with a snippet. Beyond that, the same things that make a page good for search make it good for AI Overviews.

The levers that matter most:

1. Get indexed. A page that isn't in the index can't be cited. Start with our Google indexing guide if Search Console shows gaps.
2. Answer first. Put the direct answer in the opening lines of a section, then the detail. Use headings that match real questions.
3. Use tables and lists for comparisons and steps. They are easy to extract accurately.
4. State facts precisely. Numbers, prices, dates, limits and names beat vague claims.
5. Keep pages current. For topics that change, date your content and update the facts, not just the date.
6. Show who is behind it. Clear authorship and a trustworthy site help; see E-E-A-T.
7. Add structured data that matches the page. It helps Google understand entities, though it isn't required for AI features. See schema markup.

What you can't do is request inclusion, pay for it, or guarantee it. Treat citations as an outcome of being the clearest answer to part of a question.

There is no AI-specific opt-out for AI Overviews or AI Mode. The controls are the same snippet and indexing controls that apply to all of Search.

• nosnippet stops Google showing or using text from a page. It works for AI features, but it also removes your normal search snippet, which usually costs clicks.
• data-nosnippet is an HTML attribute you can put on specific elements. Text inside them won't be used in snippets or AI features, while the rest of the page stays eligible.
• max-snippet:[number] limits how many characters Google may use.
• noindex, or blocking Googlebot in robots.txt, removes the page from Search altogether.

A common mistake is to assume Google-Extended handles this. It doesn't. Google-Extended is a robots.txt token that controls whether your content is used for training Gemini models and for grounding in some other Google products. Google has said it does not affect inclusion in Search, and AI Overviews are part of Search.

For most sites, opting out makes little sense. If a competitor is cited in the overview and you aren't, blocking Google from quoting you won't bring the click back. The better question is which passages you don't want reproduced, and data-nosnippet handles those precisely.

Traffic from AI Overviews and AI Mode is counted in Search Console's Performance report under the Web search type, mixed in with everything else. Clicks, impressions and position from AI features are included in those totals, so you can't separate them in the main report. Google counts all links inside an AI Overview at the position of the overview itself.

In 2026 Google added a Generative AI performance report. It shows impressions from AI Overviews and AI Mode over time, by page, country and device. It reports impressions only, not clicks, and it is a slice of data already counted in the Web totals rather than a new search type. Google is rolling it out gradually, and sites below a minimum impression threshold won't see it.

Practical things to watch:

• Pages whose impressions rise while click-through rate falls, often a sign an overview is answering the query on the results page.
• Which pages appear in the Generative AI report, and for which topics.
• Queries where position improves but clicks don't follow.

On clicks, be careful with numbers you see quoted. One of the better public studies, from Pew Research Center, tracked the browsing of about 900 US adults in March 2025. They clicked a traditional result on 8% of visits to a results page with an AI summary, against 15% without one, and clicked a link inside the summary on 1% of visits.

Search Console tells you about impressions. It won't tell you which competitor Google cited instead of you, or whether your product is named in the answer at all. For that you have to look at the answers themselves, regularly, for the questions that matter to your business.

A simple routine is to keep a list of target questions, run them on a schedule, and record whether your site is cited, which pages are, and who else appears. Answers vary from run to run and by location, so a single check proves little; trends over weeks are what count.

OpenHelm's AI visibility tracking does this as a scheduled job. It asks answer engines the questions from your keyword plan, through whichever answer-engine connections you have set up, and records whether your product is cited. If no engine is connected, it reports that it is blocked rather than guessing. A weekly SEO digest reads Search Console alongside it. More at AI visibility.