How Google AI Overviews choose sources

AI Overviews are the AI-written summaries that appear at the top of some Google results, with links to the pages behind them. AI Mode is a separate, conversational search experience where you can ask longer questions and follow-ups. Both are built on Google Search and its index (Google Search Central).

For brands, the important part is the links. When an AI Overview answers "which project tool suits a small agency?", the pages it links to get attention, and the brands it names get considered. Google has described both features as a way to surface a wider range of sources than a classic results page. How much traffic those links send is a separate question, and one to track in your own analytics rather than assume.

Google selects supporting links from its search index, using its usual systems to judge which pages are relevant and helpful for the question (Google Search Central). There is no separate AI index to get into.

Google also explains that AI Overviews and AI Mode may use a "query fan-out" technique: the system runs several related searches across subtopics and data sources to build the response. In practice, that means a page can be linked because it answers one part of the question well, even if it would not rank first for the whole query.

That has a practical consequence. Pages that answer one specific sub-question clearly are easier to use than long pages that cover everything loosely. A buyer's question about "project tools for small agencies with client billing" may draw on a page about agency billing, a pricing page and a comparison article at once.

Two things: it must be indexed, and it must be eligible to be shown in Google Search with a snippet. Google states there are no additional technical requirements for AI Overviews or AI Mode (Google Search Central).

So the checklist is the classic one:

• Crawlable. Googlebot can reach the page, and robots.txt doesn't block it.
• Indexable. No noindex, a working canonical, and a 200 response.
• Snippet allowed. No nosnippet on the page or the important section.
• Text in the HTML. The key content is in text Google can read, not only in images.

Use Search Console's URL Inspection tool to confirm a page is indexed. If it isn't, nothing else in this guide applies.

No. Google says you don't need to create new machine-readable files, AI text files or special markup to appear in AI Overviews or AI Mode (Google Search Central).

Structured data still has a job in Search: it makes pages eligible for rich results, and Google asks that it match the visible content. But it isn't a ticket into AI features, and adding more of it won't make a page more likely to be linked. The same goes for files such as llms.txt: Google doesn't use them for these features. Spend that time on the content instead.

The same thing that makes it perform in Search: unique, helpful content written for people. Google's own guidance for AI experiences says to focus on visitors and give them satisfying content, noting that people now ask longer, more specific questions and follow-ups (Google Search Central Blog).

What that looks like on a page:

1. Answer the real question. Write for the full question buyers ask, including the context: team size, budget, market, use case.
2. Be specific. Prices, limits, integrations, dates and named examples give the system something concrete to use.
3. Show experience. Original data, screenshots of real use, honest limits and named authors are things a summary can't get from a generic page.
4. Make the page easy to use. Google's guidance mentions page experience, including how the page works on different devices and whether visitors can find the main content easily.
5. Keep facts consistent. If your pricing page, docs and Google Business Profile disagree, you give every system a reason to hesitate.

Usually because their page answers a specific part of the question more directly, or because a third-party page that names them is the one being linked. Check the links before you rewrite anything.

Search the question yourself, logged out, and open every page the AI Overview links to. Sort them:

• Competitor pages. Compare the passage that seems to be used with your equivalent page. Is their answer in the first sentence? Do they state a price, a limit or a use case you leave vague?
• Third-party pages. Reviews, comparison articles, forums and directories. Are you listed? Is the information about you current?
• Your own pages that should qualify. Are they indexed? Do they answer the question, or a slightly different one?

Then fix the gap that shows up most often. Repeat the search over several days, because the summary and its links change between sessions and over time. One check is a hint, not a result. Prompt tracking runs the same questions on a schedule so you see the pattern.

With the same snippet controls you use for regular Search. Google applies them to AI Overviews and AI Mode as well (Google Search Central).

• nosnippet stops Google from showing any snippet of the page, which also keeps it out of AI features as a direct source.
• data-nosnippet on an element excludes just that part of the page.
• max-snippet limits how much text Google can show.
• noindex removes the page from Search altogether.

These controls are blunt. Limiting snippets to keep content out of AI Overviews also limits how the page appears in classic results. Use them for content you truly want to protect, not as a general setting.

A related point that often confuses teams: the Google-Extended token in robots.txt controls whether your content is used for Gemini models and grounding, but it doesn't affect inclusion in Google Search or AI Overviews (Google Search Central). Our AI crawler list explains each token.

Google includes traffic from AI features in Search Console's Performance report, within the regular web search data (Google Search Central). Since June 2026 Google has also been rolling out a separate generative AI report in Search Console that counts impressions in AI Overviews and AI Mode by page, country and device. It has no clicks, and those impressions are already inside the Web totals, so don't add the two together.

Visibility is easier to measure from the answer side. Track a fixed set of buyer prompts, record which brands and pages appear, and compare over time. For traffic from AI assistants outside Google, see how to track ChatGPT and Perplexity traffic in GA4. For how Google's assistant describes your brand, the Gemini tracker shows your mentions, position and sources every day.

One more practical point: check your pages on mobile. Most AI Overviews are seen on phones, and a page that's slow or cluttered on a small screen makes a poor destination for the click you earn.

Start with eligibility, then work on the pages that should answer your buyers' questions. In order:

1. Confirm your key pages are indexed and snippet-eligible.
2. List the questions buyers ask before they buy, in their full form.
3. For each one, check which page on your site answers it in the first two sentences.
4. Fix or write the missing answers, with specifics and evidence.
5. Track the prompts across Google's AI features and other engines, and review monthly.

Most of this work also helps classic rankings and the other AI engines, so none of it is wasted if Google changes how AI Overviews look next quarter. Eligibility and clear answers are the parts that last.

AI Overviews and AI Mode link to indexed, snippet-eligible pages chosen by Google's usual systems, sometimes through several related searches per question. No special markup or AI file is required. Write specific, helpful answers, keep your facts consistent, use snippet controls only where you must, and measure the result. For the wider strategy, read our GEO guide.