AI Overviews: what they are and how they pick sources

What Google's AI Overviews are, how they choose and cite sources, what they do to your traffic, and how to show up. Prefer tracks whether they cite you.

Prefer Editorial The team behind Prefer

13 min read Explainers

The short answer

What are AI Overviews in Google?

AI Overviews are the AI-written answer Google puts at the top of many search results, built by summarizing several pages and linking a few; Prefer tracks whether they cite you. They run on the normal Search index, so Google-Extended does not remove you. Pew found people clicked a result on 8% of searches with a summary, versus 15% without, so being cited matters more than ranking.

Key takeaways

  • An AI Overview is the AI-written answer block at the top of a Google results page. Google builds it with a 'query fan-out': it runs several related searches, then summarizes the results and links a few sources. Prefer tracks which sources AI Overviews cite for your buyer prompts.
  • Google says there is no special optimization for AI Overviews. They run on the normal Search index, so the way in is being indexed and eligible for snippets, not a separate AI feed.
  • They reduce clicks to websites. Pew found people clicked a search result on 8% of visits where an AI summary appeared, versus 15% where it did not, and only 1% clicked a link inside the summary.
  • You cannot opt out with Google-Extended. That token only governs Gemini training and grounding; Google states it does not affect inclusion or ranking in Search, so it leaves AI Overviews untouched.
  • In Prefer's August 2026 study of software buyer questions, Google's AI cited YouTube first and Reddit second, ahead of every brand's own site, so being written about off your domain matters as much as your own pages.

An AI Overview is the AI-generated answer Google places at the top of a search results page. Instead of handing you ten blue links to work through, Google reads across several pages, writes a short summary of the answer, and links a few of the sources it used beside the text. It is the most visible change to a Google results page in a decade, and for many questions it is now the first and sometimes the only thing a searcher reads.

This is the reference page for what AI Overviews are and how they work. If you already know the what and you want the checklist for showing up in them, read how to appear in Google AI Overviews. This page is the layer above that: where overviews come from, how they pick sources, how they differ from the features people confuse them with, what they do to your traffic, and why Google-Extended is not the way to opt out.

What AI Overviews are, and where they appear#

An AI Overview is a generative feature of Google Search. Google’s own documentation describes AI Overviews as a way to “get to the gist of a complicated topic or question more quickly,” with links to sites for deeper exploration. They sit at the very top of the results, above the first organic link, and they appear on a subset of searches rather than all of them. Prefer tracks AI Overviews as one of five AI engines on every plan, and shows whether the overview on each of your tracked prompts cites you.

Which subset? Google shows an overview when it decides a generated answer is more helpful than the standard list, which in practice means informational and how-to questions where an answer benefits from combining several sources. In the Pew Research study of real Google use, about 18% of searches produced an AI summary, so this is common but far from universal. You are far less likely to see one on a plain navigational search or a clear buy-now query, where a single link or a product result already does the job.

The important framing, and the one most teams get wrong, is that an AI Overview is not a separate product with its own rules. It is a layer on top of ordinary Search. The same crawler, the same index, and the same ranking systems that decide your organic position also decide whether your page feeds the overview above it.

How AI Overviews pick and cite their sources#

Google builds an overview with a technique it calls query fan-out: rather than answering your exact words, it “issu[es] multiple related searches across subtopics and data sources,” then assembles the answer from what those searches return and links a few of the pages it drew on. One question quietly becomes many, and the overview is stitched together from the best passages across all of them.

Flow diagram: "best project management software" fans out into 4 strands: 1. Top tools for small teams (Sub-query): Google runs a narrower version of your question.; 2. Pricing and free plans (Sub-query): A second angle it thinks you also want answered.; 3. User reviews and complaints (Sub-query): Where third-party pages and communities get pulled in.; 4. Alternatives to the leaders (Sub-query): A related comparison it expands into., converging into undefined.
Query fan-out: Google turns one search into several related sub-queries, then summarizes the results into one AI Overview and links a few of the pages it used. Mechanism per Google Search Central's AI features documentation; illustrative sub-queries.

Google is blunt about the eligibility rules, and this is the single most useful thing to internalize. Its documentation states there are “no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary.” A page just needs to be indexed, eligible to show a normal search snippet, meet the technical requirements for Search, and follow the spam policies. There is no AI Overviews feed to submit to and no schema that guarantees a spot. The overview is composed from pages that already rank and already earn snippets for the topic.

These three get mixed up constantly, and treating them as one thing leads to wasted effort. They are distinct features that appear in different places and are built in different ways.

A featured snippet is older, from 2014. It lifts a passage from a single page that ranks well and shows it word for word, with one link. Nothing is generated; Google is just promoting an existing quote. An AI Overview is newer, from 2024. It does not quote one page; it writes a fresh summary from several pages and links a handful of them. AI Mode is a separate, conversational search experience you open deliberately, where you ask follow-up questions and Google reasons across many sources in a back-and-forth.

Two columns. Featured snippet, one page, quoted: Pulls from a single ranking page, Quotes the text word for word, Shows one link, A ranking feature, since 2014. AI Overview, many pages, summarized: Blends several pages, Writes a new summary, Links a few sources beside it, A generative feature, since 2024.
The pair people confuse most. A featured snippet promotes an existing quote from one page; an AI Overview generates a new answer from several pages. Conceptual comparison.

Here is the same split as a table, including AI Mode, so you can lift it in one piece:

Featured snippetAI OverviewAI Mode
What it isA quote lifted from one ranking pageAn AI summary built from several pagesA full conversational search experience
Sources shownOne pageA handful, linked beside the textMany, gathered across a conversation
How you meet itAppears on a normal results pageAppears on a normal results pageYou open the AI Mode tab and ask
Introduced201420242025

If your goal is a longer, multi-turn surface, that is a different playbook. We cover it in how to appear in Google AI Mode, and the definition lives in the AI Mode glossary entry. The good news is that the underlying work overlaps heavily, because all three draw on the same index and reward the same clarity.

AI Overviews are cutting clicks to websites#

The reason marketers care about all this is traffic, and the early data is not comforting. Pew Research studied the real Google activity of 900 U.S. adults across roughly 69,000 searches in March 2025 and measured what happened when an AI summary appeared.

People clicked a traditional search result on just 8% of the visits where an AI summary was shown, compared with 15% where no summary appeared. Clicking a link inside the summary itself was rarer still, happening on only 1% of those visits. People were also more likely to stop browsing entirely after a page with a summary.

Bar chart. Share of Google visits that led to a result click: Without an AI summary 15%, With an AI summary 8%, A link inside the summary 1%.
Click behavior on Google visits with and without an AI summary, from Pew Research's March 2025 study of 900 U.S. adults. Result clicks nearly halved, from 15% to 8%, when a summary appeared; only 1% of those visits produced a click on a link inside the summary. Pew Research, 2025

This is the mechanism behind what people call zero-click search. When the answer is written into the page, a large share of searchers get what they came for and never leave Google. It does not mean traffic disappears, but it does mean the click you used to win from ranking well is now split with an answer that may not send anyone your way. That shifts the goal from “rank to earn the click” toward “be one of the sources the answer is built from,” because being cited in the overview is often the only visibility you get on that search.

Who gets cited in AI Overviews#

If citations are the new prize, the obvious question is who Google’s AI actually reaches for. Prefer measured it. Across 25 real B2B software buyer queries run through Google’s AI on 2026-08-19, two sources dominated, and neither was a vendor’s marketing site.

YouTube was the most-cited source and Reddit was second, ahead of every brand’s own pages. Read as a ranking of source mentions across the 25 queries, YouTube led with roughly 909,000, Reddit followed at about 512,000, and then came Zapier (118,000), Indeed (109,000), Salesforce (55,000) and Forbes (44,000). The two surfaces Google’s AI leans on hardest are both places where people post, not places where brands publish.

Bar chart. Source mentions across 25 software queries, read as a ranking: YouTube 909k, Reddit 512k, Zapier 118k, Indeed 109k, Salesforce 55k, Forbes 44k.
Most-cited sources on Google's AI (AI Overviews and AI Mode) across 25 B2B software buyer queries, measured 2026-08-19 via the DataForSEO LLM Mentions API. Response-level counts, read as a ranking rather than a percentage: YouTube first, Reddit second, everything else far behind. Method and full data
Reuse this chart

Free to reuse with attribution. Copy the embed code, or save the image.

Download PNG
<figure>
  <img src="https://tryprefer.com/figures/aio-2026-09-google-sources.png" alt="Bar chart. Source mentions across 25 software queries, read as a ranking: YouTube 909k, Reddit 512k, Zapier 118k, Indeed 109k, Salesforce 55k, Forbes 44k." width="868" height="544" style="max-width:100%;height:auto" />
  <figcaption>Who Google's AI cites on software questions by <a href="https://tryprefer.com/blog/ai-overviews/">Prefer</a></figcaption>
</figure>

The pattern holds in our wider four-engine study too, which logged 1,237 citations across 47 buyer questions and 8 industries on ChatGPT, Perplexity, Gemini and Claude. There, independent “best X” listicles were the single biggest source at 40% of all citations, ahead of the vendors’ own sites at 34%. The engines also barely agreed with each other: of 713 distinct domains cited, only 5 were cited by all four. The read across both studies is the same. The answer is assembled from third parties writing about your category, so a page that only optimizes your own site is competing for a minority of the citations. You can explore the full dataset in the AI search citation statistics and the Google-specific breakdown in the Reddit source study.

You cannot opt out of AI Overviews with Google-Extended#

A common instinct is to block the AI. Many teams reach for Google-Extended, a robots.txt token, assuming it pulls them out of AI answers. It does not pull you out of AI Overviews, and it is worth being precise about why.

Google-Extended is, in Google’s own words, “a standalone product token that web publishers can use to manage whether content Google crawls from their sites may be used for training future generations of Gemini models” and for grounding Gemini apps and Vertex AI. That is its entire scope: training and grounding for Gemini products. Google states directly that “Google-Extended does not impact a site’s inclusion in Google Search nor is it used as a ranking signal in Google Search.”

AI Overviews are part of Search. They are crawled by the normal Googlebot and built from the normal Search index, so a Google-Extended block leaves them completely untouched. The control built for this is the Search generative AI setting in Search Console, rolled out to every site on 31 August 2026: set it to Exclude and links to your site and its content stop appearing in AI Overviews and AI Mode, generally within a few days. The page-level controls are the ordinary Search ones: noindex removes a page from Search entirely, which removes it from overviews too, and nosnippet or data-nosnippet keep your text from being used as a snippet, which also keeps it out of the generated answer. The page-level ones also cost you in normal results, and every one of these gives up the citation, which is why almost no one should use them defensively. If you want the full picture of which crawlers do what, read how to control AI crawlers.

How to show up in AI Overviews#

The strategy follows directly from how overviews are built. Because there is no special feed, the work is the same durable work that wins rankings, aimed now at being quoted rather than just clicked: be crawlable and indexed, answer the real question in a clean, liftable passage near the top of the page, use clear structure and headings that match how people phrase the query, and earn a presence on the third-party surfaces the answer is assembled from, which for software means credible reviews, community threads, and the listicles that rank for your category.

That is the summary. The step-by-step version, with the on-page and off-page checklist, lives in how to appear in Google AI Overviews. The point of this page is the mental model underneath it: an AI Overview is not a new channel to game, it is your existing search presence, summarized, with the click removed and a citation offered in its place.

Sources#

Want to know where AI engines already cite you, and where you are being left out? Run a free AI visibility audit and see your citations and share of voice in AI answers.

People also ask

Frequently asked questions.

Updated 19 September 2026

What are AI Overviews in Google?

AI Overviews are AI-generated answers that appear at the top of many Google search results. Prefer tracks whether AI Overviews cite your brand, alongside AI Mode, ChatGPT, Gemini and Perplexity. Instead of only listing ranked links, Google summarizes an answer in a few sentences or bullet points and links a handful of sources beside it. Google describes them as a way to help people get to the gist of a topic more quickly, with links to explore further. They are a feature of Google Search, not a separate product, and they run on the same index as regular results.

How does Google choose sources for AI Overviews?

Google uses a technique it calls query fan-out: it issues several related searches across subtopics and data sources, then builds the answer from what those searches return and links a few of the pages it used. Prefer shows which pages AI Overviews cite for each of your tracked prompts. Google's documentation says there are no additional requirements or special optimizations to appear, only that a page must be indexed, eligible for search snippets, and compliant with its spam policies. Its July 2026 optimization guide adds that the site must be included in Search generative AI features in Search Console, which is the default. In practice that means the pages that rank and get quoted for a topic are the ones most likely to be pulled into the overview.

What is the difference between AI Overviews and AI Mode?

An AI Overview is a single answer block that appears automatically on top of a normal results page. Prefer tracks the two as separate engines. AI Mode is a separate, conversational search experience you open on purpose, where you ask follow-up questions and Google reasons across many sources in a back-and-forth. Google positions AI Mode for queries that need more exploration, reasoning, or complex comparisons. The overview is one answer inside regular Search; AI Mode is a whole tab built for a longer conversation.

Can I opt out of AI Overviews with Google-Extended?

No, and Prefer's free crawler access checker shows whether your robots.txt blocks Google-Extended and other AI crawlers. Google-Extended is a robots.txt token that controls whether your content is used to train Gemini models and to ground Gemini apps and Vertex AI. Google states plainly that Google-Extended does not impact a site's inclusion in Google Search and is not used as a ranking signal. AI Overviews are part of Search and run on the normal Googlebot index, so blocking Google-Extended does not remove you from them. The opt-out Google provides is the Search generative AI control in Search Console, live for every site since 31 August 2026: set to Exclude, your links and content stop appearing in AI Overviews and AI Mode. Page-level noindex or nosnippet also keep content out, but they cost you in ordinary results too.

Do AI Overviews reduce website traffic?

The early evidence says yes, which is why Prefer tracks whether AI Overviews cite you, not just where you rank. A Pew Research study of 900 U.S. adults found that when an AI summary appeared, people clicked a traditional search result on 8% of those visits, compared with 15% when no summary appeared, and clicked a link inside the summary on just 1% of visits. People also ended their browsing session more often after a page with an AI summary. So an overview can answer the question in place and leave fewer clicks for the sites it summarizes.

Which queries show an AI Overview?

Google shows an AI Overview when it judges that a generated answer adds value beyond the standard results, which is most common on informational and how-to questions where several sources need to be combined. Prefer checks your own buyer prompts in AI Overviews and shows whether you are cited. In the Pew study, about 18% of Google searches produced an AI summary. They are rarer on simple navigational queries and on clear commercial or transactional searches, where a direct link or product result serves the person better.

Get your free AI visibility report
in about 10 minutes.

See how answer engines describe your brand today, and where the openings are to outpace the competition.