How Google AI Overviews Work: Where the Answers and Links Come From

The paragraph at the top of your Google results — the one that answers the question before the links begin — is called an AI Overview, and most people have no idea how it got there or why it appears for some searches and not others. It is not written by a person, not copied from the top result, and not shown on a whim. This guide explains the machinery from Google’s own documentation and the best independent measurement of how often it appears and what people do with it. It is part of our pillar series explaining AI search; the pillar covers all three big systems, while this article stays inside Google’s.

Step 1: Google decides whether your search gets an Overview at all

Most searches still return the classic page. Google says AI Overviews “are only shown when our systems determine that it is additive to classic Search, and as such, often don’t trigger.” What kinds of searches trigger them is visible in Pew Research’s measurement: in March 2025, about 18% of all tracked US Google searches produced an AI summary, but the share rose steeply with how the query was phrased — just 8% of one- or two-word searches, 53% of searches with ten or more words, 36% of searches written as full sentences, and 60% of searches beginning with question words like who, what, when or why. In other words, the more your search looks like a question a person would ask aloud, the more likely Google is to answer it like one. The median summary in Pew’s sample was 67 words — a paragraph, not an essay.

Step 2: The query fan-out — your one question becomes many searches

Here is the part Google named itself. Its Search Central documentation explains that AI Overviews and AI Mode “may use a ‘query fan-out’ technique — issuing multiple related searches across subtopics and data sources — to develop a response,” and that “while responses are being generated, our advanced models identify more supporting web pages,” producing “a wider and more diverse set of helpful links” than a classic search shows. Translated: when you ask “what is the best way to treat a sprained ankle,” Google doesn’t just run that string. It quietly runs a family of related searches — symptoms, recovery time, when to see a doctor, ice versus heat — gathers pages for each, and hands the bundle to the model. That is why an Overview can feel broader than any single result on the page: it was built from several searches you never saw. It is also the first place errors can enter — if the fan-out chases the wrong subtopics, the answer inherits the detour.

Step 3: Pages are pulled from the same index as classic results

There is no separate “AI index” your favourite website must join. Google’s documentation is explicit: “To be eligible to be shown as a supporting link in AI Overviews or AI Mode, a page must be indexed and eligible to be shown in Google Search with a snippet… There are no additional technical requirements.” The same crawling, the same index, the same snippet rules — a page blocked by a noindex tag or stripped by snippet controls is out of the AI answers too, and site owners control their content through those same standard levers (robots.txt for crawling, nosnippet and max-snippet for previews). This is why the sources inside summaries look familiar. Pew found the most frequently cited sources in AI summaries were Wikipedia, YouTube and Reddit — together 15% of all sources, nearly the same share (17%) they hold in standard results — with government sites notably more common in summaries (6% of sources) than in standard results (2%). The summary is drinking from the same well as the list beneath it.

Step 4: The model writes the paragraph and hangs links on it

With pages gathered, a generative model composes the answer text and attaches links to the pages presented as its support — 88% of the summaries Pew captured cited three or more sources, and only 1% cited a single source. Two things follow. First, the links are genuinely worth more than decoration: they are the audit trail, the only way to check which page stands behind which sentence. Second, the attachment is done by the same kind of model that writes the text, so a citation can point to a page that merely sounds relevant, or summarise a page loosely — the failure family our guide to why AI search answers go wrong examines with the Tow Center’s data. Google’s own framing of the feature’s purpose is worth quoting because it states the intended division of labour: Overviews “help people get to the gist of a complicated topic or question more quickly, and provide a jumping off point to explore links to learn more.” A jumping-off point. The design intent is a start, not a substitute.

Step 5: What happens next is the measurable part

Google says that “when people click from search results pages with AI Overviews, these clicks are higher quality” — visitors spend more time on the site they land on. Pew’s numbers describe the other side of that coin: far fewer clicks happen at all. Users clicked a traditional result link in 8% of visits with a summary versus 15% without; clicked a link inside the summary itself in just 1% of visits; and ended their browsing session on the spot in 26% of summary visits versus 16% of no-summary visits — the signature of a question answered well enough to stop searching, for better or worse. Around two-thirds of all searches in the study ended without any result click either way, so Overviews intensify an existing habit rather than inventing one. But the direction for anyone who publishes, sells or informs through the web is unmistakable, and it is why this feature — a paragraph — has become one of the most argued-about objects in media.

AI Overviews versus AI Mode, in one paragraph

Overviews are the summary inside normal Search; AI Mode is the separate, fully conversational experience Google built for “queries where further exploration, reasoning, or complex comparisons are needed,” where you can ask follow-ups and get a comprehensive response with links. Google notes the two “may use different models and techniques, so the set of responses and links they show will vary” — the same question can produce a different answer, and different sources, in each. If you are checking a fact, that variance is itself information: agreement between the Overview, AI Mode and a classic result is a stronger signal than any one of them alone.

What this means

Understanding the fan-out changes how a careful reader uses the feature. Because the Overview is a synthesis of several hidden searches, its blind spots are systematic: subtopics the fan-out didn’t chase are simply absent, and no amount of staring at the paragraph reveals them — the fix is to ask the follow-up question the fan-out skipped, or run the specific search yourself. Because its sources come from the ordinary index, the summary inherits the ordinary web’s quality distribution — strong on documented facts, thin on anything the web itself argues about, and skewed toward the sources Pew measured: encyclopaedias, video, forums and government pages. And because the whole design trades your clicks for Google’s synthesis, the reader’s counter-habit has to be deliberate: treat the paragraph as the map, not the territory, and spend your saved time on the one or two links that carry the claim you actually need. For US readers especially — Pew’s panel is American, and Overviews rolled out in the US first — this is now the default reading posture for everyday questions. The feature is genuinely useful, genuinely fallible, and completely transparent about its ingredients if you know where to look: the links are the ingredients. Read them when it matters.

The reader-side consequence is subtler and worth naming. When the first screen of results was links, choosing was the reader’s job and everyone knew it; when the first screen is a paragraph, choosing has quietly been done for you — which sources, which framing, which caveats survived. That is a transfer of editorial judgement from reader to system, and it is mostly invisible because the paragraph looks like a neutral summary rather than an edited selection. The healthiest habit the Overview era offers is therefore not suspicion but sequence: read the summary for orientation, then make one deliberate click — to the primary source when the matter has one — before the summary hardens into what you believe. Orientation is a genuine gift; the mistake is only in letting it be the whole journey.

Bottom line

The paragraph at the top of Google is a synthesis of hidden searches over the ordinary index — convenient, and blind exactly where its fan-out did not look. Ask the follow-up it skipped, and click through when the answer matters.

Read Next

• AI Search Explained: How AI Overviews, ChatGPT Search and Perplexity Build Their Answers
• Perplexity vs ChatGPT Search vs Google AI Overviews: Which Should You Use?

Sources

• AI Features and Your Website — Google Search Central (accessed 7 November 2026)
• Do People Click on Links in Google AI Summaries? — Pew Research Center (accessed 7 November 2026)