A Google AI Overview is an AI-generated summary box that appears at the top of Google Search results, where Google synthesizes information from multiple web pages to answer a searcher’s query directly. A Google AI Overview delivers **synthesized answers at the top of Google Search results**, ahead of the organic listings. Google AI Overview was initially called the Search Generative Experience and was developed after testing in Google Labs before its public launch. Google AI Overview works by retrieving passages from multiple pages and then generating a single answer from them.

This page covers what an AI Overview is made of, where it sits on the page, the six steps behind it, which Gemini generation runs it, when it launched and how far it has come, whether it carries ads, what determines a citation, and how a publisher leaves. Six first-party measurements taken in late August and early September 2026 sit alongside the published figures, and where the two disagree, both are shown.

An AI Overview appears on roughly one in five SERPs, according to the largest published count, though estimates range from 9% to 25% depending on what is counted. The trigger conditions page sets out why.

Model names, availability figures and opt-out mechanics on this page reflect Google Search as of 10/13/2026.

What is a Google AI Overview?

A Google AI Overview is an AI-generated summary box that appears at the top of Google Search results, where Google synthesizes information from multiple web pages to answer a searcher’s query directly. Google positions this generative AI box (an AI-generated summary) above the traditional listings. The Google AI Overview synthesizes web information to deliver an instant answer to the searcher within Search results. This AI Overview is a built-in part of normal Google Search, not a separate product. It requires no special activation from the user. The box sits above traditional organic results. It can appear alongside SERP features such as featured snippets, People Also Ask, and the local pack. Because a Google AI Overview answers on the page, it contributes to zero-click behavior across Google Search.

What is the purpose of a Google AI Overview?

The purpose of a Google AI Overview is to condense multiple web sources into a single, structured summary so users can get an answer to a query without clicking through to any pages. Google added the AI Overview instead of only ranking pages because ranking returns 10 links that the user still has to read and compare, while the AI Overview does the reading and synthesis for the user. Google AI Overview draws on multiple pages at once and merges them through synthesis (combining separate sources into one summary). Because the AI Overview delivers the answer directly on the search results page, the search can end without the user visiting any website. Shaping content so that it is selected as a cited source in the summary is a separate discipline, discussed later on this page.

What components make up an AI Overview response?

Google AI Overviews synthesize web information into an answer shown at the top of Google Search results. An AI Overview response is made up of six components:

• a structured summary (the AI-generated answer body)
• bullet points (short parallel takeaways)
• cited sources (the pages Google drew from)
• supporting links (AI Overviews usually draw from between four and eight pages at once, depending on topic and query)
• images or visual placement (inline media beside the text)
• follow-up prompts (interactive questions below the summary)

Each of the following sections examines one component of an AI Overview in turn. The structured summary carries the synthesis; the citations and links carry attribution.

Does an AI Overview cite multiple sources or one?

An AI Overview cites multiple sources, not one. Most AI Overviews draw from four to eight pages per answer, depending on topic and query. Google synthesizes information from several web pages into a single structured summary, then attaches cited sources as supporting links. A single AI Overview answer distributes visibility across those pages rather than one, so individual publisher visibility drops per search even when a page is cited.

Counting the citation set on 91 AI Overviews on 31 August 2026:

The four-to-eight figure holds at the median. The other 31 fell outside it, and 28 of those fell above: nine, ten, eleven pages, and in one case, eighteen.

The published range describes the middle of the distribution, not its edges. A page competing for a slot in an eighteen-source overview is in a different contest from one competing in a four-source overview, and the single figure hides which contest a query is.

What follow-up features appear inside an AI Overview?

An AI Overview shows four follow-up features inside the response:

• an expansion control (a ‘show more’ button that reveals the full AI-generated summary)
• follow-up questions (related queries listed below the answer)
• dive-deeper links (paths into Google AI Mode)
• inline images (visual media beside the text)

The visible paragraph is not where a Google AI Overview ends, because the expansion control and follow-up layer extend it below the initial summary.

Across 20 AI Overviews inspected on desktop on 6 September 2026:

Two of those four are not what the component list describes. The AI Mode entry point is no longer a labeled “dive deeper” link. It is an open input box sitting below the overview, and reaching AI Mode means typing into it. And where a follow-up appeared, it was always a single question, never three and never a set of clickable chips: one conversational sentence at the end of the answer, asking whether the reader wanted a narrower version.

The four components are a maximum set, not a fixed one. Only two appeared in every overview, and the shapes of both have changed since the feature was documented.

Where do AI Overviews show up on Google?

The AI Overview appears at the top of the Google search results page, above the organic listings. This placement holds on both desktop and mobile. On local-intent queries, the AI Overview layers above the local pack. The AI Overview can displace the featured snippet that would otherwise appear at the top of the results page.

What does the AI Overviews logo represent?

The AI Overview logo is a Gemini-style sparkle icon paired with the ‘AI Overview’ label. The icon is a star-shaped sparkle glyph. The label reads ‘AI Overview’ beside it. Readers identify an AI Overview at a glance by the sparkle icon and label positioned above the standard organic listings.

As rendered on 2 September 2026 on desktop in a Bangladesh locale, the header carried a four-point sparkle glyph, the mark Google uses across its AI surfaces, followed by the label “AI Overview” and a Bengali language toggle.

That fixed placement settles where the answer sits.

How do AI Overviews work?

Google AI Overview works in two stages, retrieval then synthesis, and six steps sit inside them:

1. The searcher enters a query.
2. Google classifies its intent and context.
3. Google fetches candidate passages from the index, weighing related searches and what searchers tend to click for similar queries.
4. Gemini compresses and synthesizes those passages into one answer.
5. Google attaches citations to the source pages — the step it calls grounding.
6. The system adjusts over time based on which sources searchers click.

Steps 3 and 6 explain the otherwise puzzling citations. Retrieval is not about reading the top ten and summarising them; it is about weighing a wider set of candidates against related queries, and it learns from this. Full mechanics are on retrieval-augmented generation.

That synthesis step runs on one model, which raises the next question.

Which AI model powers Google AI Overviews?

Google AI Overviews run on Gemini, Google’s family of large language models. Gemini is the full model family, not a separate lightweight or reduced sub-model. Google does publish the generation, and it moves: AI Overviews shifted to Gemini 3 globally on 27 January 2026, having previously run on 2.5-series models, with Gemini 2.0 before that from March 2025.

Naming the current generation dates a page within weeks. The useful thing to know is not which number is live today, but that the trigger and selection behavior is a model judgment, so it can move without any change in your content or the query. Every cross-niche figure quoted below predates the January 2026 switch.

What triggers Google to show an AI Overview?

Google shows an AI Overview when its query intent classification (how Google labels what a searcher wants) and query type signal that a synthesized answer fits the search. Query intent and query type name that trigger and set the conditions for retrieval are explained in Query Signals and Conditions.

Google AI Overview triggers across six intent types:

• Informational searches that ask a direct question.
• Commercial searches that compare products or services.
• Transactional searches that signal a purchase step.
• Navigational searches that target a known destination.
• Branded searches that name a specific company or product.
• Local-intent queries that combine a service with a location.

Local intent splits three ways in Whitespark’s local-search research: AI Overviews appear on around 15% of pure “service + location” searches, on 68% of local business-type queries, and on 92% of local queries carrying an educational or advisory layer.

An AI Overview fails to render for a given query or user due to four restriction types. Age restrictions block sensitive topics. Region limits coverage by country. Language narrows supported locales. Query type excludes searches that Google classifies as unfit for synthesis.

Google AI Overview launched inside Search on a specific date, which fixes when this triggering behavior first reached users.

When did Google AI Overviews start?

Google launched AI Overviews in the United States on 14 May 2024 at I/O, graduating the Search Generative Experience out of Labs under a new name (Google, “New generative AI experiences in Search”).

In October 2024, the share of keywords triggering an AI Overview began to increase across commercial, transactional, and navigational searches, according to Semrush.

No first-party tool reaches the beginning of that timeline. Search Console keeps 16 months of data, which, as of September 2026, reaches back to May 2025, a year after launch. The Generative AI report began collecting in mid-2026 with no backfill. The traffic effect of the May 2024 launch is no longer measurable from any site’s own data and will not become measurable again.

What was AI Overviews called before its rebrand?

AI Overview was previously called Search Generative Experience (SGE), the name Google used during its Google Labs testing phase. Search Generative Experience and AI Overview are the same feature in Google Search. The rebrand moved the feature out of an opt-in Labs experiment and into default Search results and shipped a custom Gemini model built for Search alongside the new name. SGE names the Labs phase; AI Overviews names the product that replaced it. These overviews initially carried the SGE label, then shifted to the AI Overview label across Google Search.

What countries have AI Overviews rolled out to?

Tested from a Bangladesh location on 6 September 2026, AI Overviews fired on 20 of 20 English-language queries and 20 of 20 Bengali-language equivalents on the same twenty topics, in a market that appears in no published rollout list. No topic returned an overview in one language and not the other.

Country availability and language availability are separate conditions, and in this set, both were met. The English arm was checked twice, once manually in the browser and once via a depersonalized SERP API, and both passes returned 20 of 20.

Google AI Overview has rolled out to more than 200 countries and territories, including the United States, India, and Germany. Market availability depends on the reader’s country and language, so a reader in a confirmed market such as the United States, India, or Germany can self-check. Google AI Overview availability is defined by both a country set and a supported-language set, so a reader in a supported country sees it, while an unsupported language excludes it, per the Google Search Central Blog.

How do AI Overviews change search traffic?

AI Overviews change search traffic by synthesizing answers at the top of the search results page and reducing clicks to organic listings for informational searches. An AI Overview appears above the organic results, so the same organic position can produce fewer clicks when an AI Overview satisfies the query, and it can produce higher-quality clicks for comparative or decision-driven queries. This shift increases zero-click searches (searches that end without a click to any website) and lowers click-through rates for affected pages. AI Overviews appear above the local pack for advisory local queries, which affects Google Business Profiles. The size of this effect varies by search intent, so click loss is concentrated where the summary fully answers the query, per the Google Search Central Blog (May 2025). Because informational queries lose clicks while comparative queries keep decision-driven visits, the click count drops, but referral value shifts by intent.

If you need the click numbers rather than the mechanism, that is documented in AI Overviews and Zero-Click Search, and the value shift is on AI referral traffic quality.

Are ads displayed within AI Overviews?

AI Overviews carry ads in two distinct placements: above or below the overview, which serve across markets where AI Overviews run, and inside the overview, which are narrower.

Eligibility considers the AI Overview’s own content alongside the query, so an ad competes against how Google framed the answer rather than against the search string alone. An advertiser can neither target nor opt out of the placement.

What decides whether a page gets cited?

Four levers decide it: intent match, factual consistency with other sources, one-question-per-passage structure, and presence in the set Google already retrieves. Each is a body of work rather than a setting, and each is covered in depth on AI Overview Optimization. This page stops at naming them.

One mechanism finding constrains all four, and it sits in the section below.

How does an AI Overview select its citations?

An AI Overview selects its citations from pages whose passages repeat the same verifiable facts that other pages confirm across the web. The mechanism favors passages that repeat the same verifiable facts confirmed across the web, so a weaker page gets cited when its passage matches the cross-web consensus more cleanly than a stronger page, not because of overall domain strength.

An AI Overview weighs three signals when choosing which page to cite:

• Shared facts, statements that other pages confirm.
• Consistent explanations: the same claim is explained the same way across sources.
• Top-10 presence, where the page already ranks in the retrieved set.

The scoring is based on passage ranking (scoring a single passage rather than the whole page). For the scoring mechanism itself, see how passages are scored and selected.

Google’s patent on generative search results, US11769017B1, describes a selection process that deliberately seeks diversity: when the top-ranked results for a query are homogeneous, the system reaches into closely related queries rather than citing more of the same. Rich Sanger’s reading of the patent gets straight to the point. The AI Overview draws from the top sources for a query.

That is consistent with what the SERPs show. Across 89 AI Overview SERPs checked on 31 August 2026, 40 included a cited page with fewer referring domains than an uncited page ranking above it. Regarding how two-factor authentication works, the cited page ranked 8th with 1 referring domain, while an uncited page at position 3 had 4,871. On four of the five widest gaps, the uncited page was Wikipedia.

Referring domains are one proxy for authority, not authority itself. What the count shows is that link profile and rank together do not determine the citation set: on nearly half of these SERPs, Google reached past a stronger, higher-ranking page for a weaker, lower one.

How do you track AI Overview visibility?

Three methods track AI Overview visibility, and none of them measures the same thing: rank-tracker flags, GA4 referral segmentation, and manual result checks. A SERP-tracking flag tells you an AI Overview exists for a query, not whether your page is in it. GA4 records the session but not the surface, so no AI Overview visit can be distinguished from ordinary organic traffic there. Only the manual check reaches citation level, and only one query at a time. The full method for each level is available at AI Overview visibility measurement and tracking. This page stops at naming them.

Is there an API to access AI Overview data?

No public API exists to access AI Overview data directly. Google provides no official standalone API to programmatically retrieve AI Overview data, so large-scale collection relies on third-party tracking tools rather than a Google-provided endpoint. Semrush offers such tracking at scale.

Search Console’s Generative AI report, added in June 2026, is the only first-party source that shows which of your pages earned impressions in AI Overviews and AI Mode, by page, country, device, and date. It carries no clicks or query dimension and blends the two surfaces. Availability is uneven at the property level rather than the account level.

For the full tracking methods, see AI Overview visibility measurement and tracking. Coverage aside, retrieval quality raises a separate question. How accurate are AI Overviews?

How accurate are AI Overviews?

AI Overview accuracy has no published rate, and the mechanism makes it hard to define. Decomposing 18 AI Overviews into 91 scored elements, 47 of them, 51.6%, could not be traced back to any cited page. Some of those were restructured figures rather than errors: percentage breakdowns and worked examples that no single source presents in that form. An untraceable claim is not automatically a wrong one, but it is a claim you cannot verify from the sources shown, and that is the practical accuracy problem for a reader. Method and thresholds are on AI Overview Optimization.

The mechanism behind that is retrieval plus synthesis. Google scans multiple pages for shared facts and consistent explanations, then synthesizes them into a structured overview. When those sources conflict, the synthesis carries a hallucination (a confidently stated but incorrect answer). Low content quality raises hallucination risk. This retrieval mechanism is detailed in Understanding RAG: How Retrieval-Augmented Generation Powers Google AI overviews.

These accuracy limits lead some searchers to want the feature disabled.

Can you turn off AI Overviews?

Searchers cannot fully turn off AI Overviews in standard Google Search, and since 3 June 2026, a publisher can remove its own domain from them. AI Overviews appear by default across most informational queries. Three user-side controls reduce exposure: switching to the Web filter or tab in Google Search, using a browser extension that hides the generative AI box, and choosing a search engine without an AI Overview layer. The browser extensions that hide the Google AI Overview panel exist and remain available.

Publishers can now exit, and the position changed recently enough that most published guides still say otherwise. Google added a domain-level control in Search Console on 3 June 2026 under a UK Competition and Markets Authority conduct requirement, began honoring it on 17 June, and has since extended the rollout beyond the UK. It sits under Settings, in a section called Search generative AI: include the site in AI Overviews, AI Mode, and Discover’s generative features, or exclude it. Google states that the setting is not used as a ranking or inclusion signal anywhere else in Search. Changes take one to two days to propagate, and page-level control is scheduled for March 2027.

Two things sit outside that toggle. It does not govern model training, which runs through Google-Extended. And Google-Extended does not govern AI Overviews, which is the reverse mistake and the more common one.

Before that control existed, the only options were nosnippet or max-snippet, which removed or limited the text Google could lift across Search, rather than from AI Overviews alone. Glenn Gabe tested this and found Google honored it, though slowly, and at the cost of the regular search snippet. Full mechanics are on zero-click search.

Where the numbers on this page come from

Published figures and first-party measurements are kept separate. Where a figure is someone else’s, the source is named. Where it is mine, the sample and date are stated with it.

What limits these numbers

The samples are small and single-snapshot. AI Overview citations change between generations, so a count taken once understates variation. Ahrefs found the surface details of an AI Overview, its wording, URLs and entities, changing roughly every two days while the underlying meaning stays stable.

Referring domains are a proxy for authority, not authority. The link-inversion finding shows that link profile and rank together do not determine the citation set. It does not show what does.

The language test ran on twenty matched topics, not a random sample. Both arms fired on every topic, which means the test found a ceiling rather than a rate.

The component audit is desktop only. Mobile layout differs and was not checked.

Everything above is what an AI Overview is and how it behaves. The questions below sit at the edge of that: each comes up whenever the feature is explained, and each belongs mainly to another page in this cluster.

Related questions

How is an AI Overview different from a featured snippet?

A featured snippet lifts one passage from one page; an AI Overview synthesises across several and cites each. The two can appear on the same SERP, and an AI Overview can displace a snippet that would otherwise have rendered. The mechanism difference, and what it changes about how a passage is written, is on featured snippets vs AI Overviews.

Does an AI Overview appear in AI Mode?

No. AI Mode is a separate tab that replaces the classic results rather than sitting above them. The two surfaces run their own retrieval and cite differently, so a page cited in one is not automatically cited in the other. The comparison is on AI Overview vs AI Mode.

Do AI Overviews cite AI-generated content?

More often than they cite human-written content, on Ahrefs’ reading of the data. Running one million SERPs and 1.9 million cited URLs, Ahrefs found AI Overviews citing AI-generated pages at a higher rate than human-written ones. That is a correlation across a large sample rather than a stated preference, and it says nothing about which kind of page earns a citation in any single niche.

What to take from this

An AI Overview is a synthesis, not a snippet, and that single distinction explains most of its behaviour. It draws from several pages rather than one, so being cited splits attention rather than winning it. It rewrites rather than extracts, so half of a given answer traces back to no cited page at all. It runs on a model that changes, so the trigger threshold moves without your content moving. And since June 2026 you can leave it, at a cost. What to do about all of that is on AI Overview Optimization; what it costs you in clicks is on zero-click search.