AI Overviews

How Google AI Overviews work — the Gemini-generated box in Google's own SERP, query fan-out, how it sources and cites pages from the core index, and how it differs from featured snippets and AI Mode.

First published: Jul 3, 2026 · Last updated: Jul 21, 2026 · Advanced
AIAuthored byAI v2

AI Overviews are Gemini-generated summaries Google shows above or within regular results. They are a Search feature rooted in Google's core Search ranking and quality systems, and supporting pages must be eligible for Search snippets. Google says no additional technical requirements apply. AI Overviews may use query fan-out—several related searches across subtopics and data sources—to assemble an answer with supporting links. They differ from featured snippets and from AI Mode, and independent studies of their triggers, placement, citations, and clicks should be read as dated observations rather than permanent platform rules.

Eligibility still starts with an indexedStoring a crawled page in the search index so it can appear in results. Crawled is not the same as indexed — Google selects what to keep, and indexing isn't guaranteed., snippet-eligible page rather than a separate AI-specific technical standard. Evidence for this claim Google says pages shown as supporting links in AI Overviews must be indexed and eligible to appear in Search with a snippet, with no additional technical requirements. Scope: Google Search AI features; eligibility does not guarantee selection. Confidence: high · Verified: Google: AI features and your website Google also documents query fan-outQuery fan-out is the technique where an AI search system breaks a single user question into multiple related sub-queries, runs those searches concurrently, and synthesizes the retrieved results into one answer. Google confirms AI Overviews and AI Mode 'may use a query fan-out technique' issuing multiple related searches across subtopics. as multiple related searches across subtopics and sources—not as a guaranteed step in every response. Evidence for this claim Google documents query fan-out as issuing multiple related searches across subtopics and data sources for AI features. Scope: Google AI Overviews and AI Mode; Google says the technique may be used, not that every response uses it. Confidence: high · Verified: Google: AI features and your website

TL;DR — AI OverviewsAI Overviews are the AI-generated summary box Google shows above or within its regular search results, written by Gemini models from pages retrieved out of Google's normal Search index. It's a Search feature, not a separate platform or index. are a Gemini-generated summary rendered inside Google’s own SERP. Google says the feature is rooted in its core Search ranking and quality systems, then generates prose with supporting links. It uses retrieval and groundingGrounding is anchoring an AI model's answer to source documents it retrieves at the moment you ask — not to the patterns frozen into its weights during training. Retrieval-Augmented Generation (RAG) is the most common way to do it., groundingGrounding is anchoring an AI model's answer to source documents it retrieves at the moment you ask — not to the patterns frozen into its weights during training. Retrieval-Augmented Generation (RAG) is the most common way to do it.), frequently powered by query fan-outQuery fan-out is the technique where an AI search system breaks a single user question into multiple related sub-queries, runs those searches concurrently, and synthesizes the retrieved results into one answer. Google confirms AI Overviews and AI Mode 'may use a query fan-out technique' issuing multiple related searches across subtopics.: multiple sub-searches whose retrieved pages Gemini extracts facts from and stitches together. Eligibility is ordinary: a page must be indexed and snippet-eligible, with “no additional technical requirements.” AI Overviews are distinct from featured snippets (one lifted passage from one page) and from AI Mode (a full conversational surface). They don’t always sit at position 1 (~8.64% appear elsewhere), they skew heavily toward informational, long-tail, non-commercial queries, and the correlation between ranking #1 and being cited has weakened materially as fan-out has matured.

Same index, same ranking — a feature, not a platform

The single most important fact about AI Overviews is that they aren’t a new search engine bolted onto Google. They’re a presentation layer over the results Google already produces. Google says it plainly in its May 2026 optimization guide: “our generative AI features on Google Search are rooted in our core Search ranking and quality systems,” and therefore “from Google Search’s perspective, optimizing for generative AI searchAI search uses large language models and retrieval-augmented generation (RAG) to synthesize an answer from multiple sources rather than returning a ranked list of links. Examples include Google AI Overviews, ChatGPT Search, and Perplexity. is optimizing for the search experience, and thus still SEO.”

That’s why eligibility is unremarkable. To be shown as a supporting link, “a page must be indexed and eligible to be shown in Google Search with a snippet, fulfilling the Search technical requirements. There are no additional technical requirements.” The crawl → index → rank chain is the whole prerequisite — the same chain that governs crawlingCrawling is how search engines use automated bots (like Googlebot and Bingbot) to discover URLs and download pages. A page has to be crawlable to be indexed, but crawling on its own isn't a ranking factor. and ordinary ranking. Whatever keeps a page out of the normal results keeps it out of AI Overviews too.

This contrasts sharply with engines like ChatGPT or Perplexity, which build their own retrieval pools. Those are the multi-platform concern that generative engine optimizationGenerative Engine Optimization (GEO) is the practice of optimizing content and brand presence so AI-powered search engines and assistants — Google AI Overviews, ChatGPT, Perplexity — cite, recommend, or mention you when generating answers. Google's position is that it's still SEO. and answer engine optimizationAnswer Engine Optimization (AEO) is the practice of structuring content so engines deliver it as a direct answer — featured snippets, voice assistants, and AI search — rather than just a ranked link. Coined for voice search in 2018 and revived for the LLM era. Google's position is that it's still SEO. own. AI Overviews are the Google-specific case: same index, same rules.

The mechanism: grounding + query fan-out

Under the hood, AI Overviews are a RAGRAG is the retrieve-then-generate pattern behind AI search: the system retrieves relevant passages from an external index at query time, injects them into the model's context, and generates an answer grounded in those sources — without changing the model's weights. system — retrieval-augmented generation, which Google itself calls grounding. The general mechanics of retrieve-then-generate belong to the RAGRAG is the retrieve-then-generate pattern behind AI search: the system retrieves relevant passages from an external index at query time, injects them into the model's context, and generates an answer grounded in those sources — without changing the model's weights. and groundingGrounding is anchoring an AI model's answer to source documents it retrieves at the moment you ask — not to the patterns frozen into its weights during training. Retrieval-Augmented Generation (RAG) is the most common way to do it. deep dives, so I won’t re-derive them here. Google’s own definition is worth keeping in front of you, though: RAG is “a technique (also known as grounding) used to improve the quality, accuracy, and freshness of AI responses by relying on our core Search ranking systems to retrieve relevant, up-to-date web pages from our Search index.”

The AI-Overview-specific twist is query fan-out. Rather than answering from a single result set, Google can decompose your question: “Both 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.” Gemini then “analyzes the content from these various pages, extracts the most relevant facts, and stitches them together into a single, comprehensive answer.” So one query can pull from the SERPs of several sub-queries you never typed — which is why depth across the sub-questions of a topic matters more here than ranking for one head term. (Fan-out is also the mechanism behind AI Mode and the reason it shares so much with Overviews.)

While the answer is being assembled, Google says it casts a wider net for links than a classic result page: “our advanced models identify more supporting web pages, allowing us to display a wider and more diverse set of helpful links associated with the response than with a classic web search.”

TIP Turn one topic into the sub-questions a fan-out system may retrieve

This is a planning simulation, not Google's hidden query log. It generates five related sub-queries and compares them with a supplied page chunk using lexical coverage; embeddings are explicitly not evaluated.

Generate a fan-out and review which answer shapes your page actually covers with my free Query Fan-Out Simulator Free

  1. Enter the user-level topic and the page passage you expect to support it.
  2. Review the generated definitional, comparative, entity-expansion, and implicit sub-questions as planning hypotheses.
  3. Confirm answer shape and evidence manually; lexical coverage is not proof that Google issued or selected a query.
Fan-out broadens the retrieval surface. This local simulation helps inspect topic coverage without claiming to reveal Google's actual queries.

The simulator lists five generated sub-queries for technical SEO audit: the head topic, a definition, benefits and limitations, alternatives, and how to choose. It reports lexical term coverage from 60 to 100 percent and explicitly says embeddings are not evaluated in this run.

What triggers an AI Overview

AI Overviews don’t fire on every search — by design. Google: “AI Overviews are only shown when our systems determine that it is additive to classic Search, and as such, often don’t trigger.” The pattern in the aggregate data is strongly skewed:

  • Informational intent dominates. In an Ahrefs analysis of 300K keywords, 99.2% of AI-Overview-triggering keywords were informational, and AIOs had only about a 10% chance of showing for commercial and transactional queries.
  • Longer, harder-to-monetize queries. AIO keywords averaged a ~4-word phrase length vs. ~2 words for non-AIO SERPs, with a much lower median Keyword Difficulty (~12 vs. ~33). In my study of 55.8M AI Overviews across 590M searches, AIOs showed more for informational, longer, higher-volume queries — and less for branded, local, and short queries. About 71.67% of AIO searches carried no CPC data at all, i.e. they lean non-commercial.
  • They co-occur with other SERP featuresSERP features are any element on a search results page beyond the classic ten blue links — featured snippets, People Also Ask, knowledge panels, sitelinks, image and video packs, AI Overviews, and structured-data-driven rich results. Google documents that losing rich-result eligibility doesn't affect ranking; their SEO relevance is CTR (they redistribute clicks). Some are unlocked by markup; most are purely algorithmic.. In the 300K-keyword study, 99.9% of AIO SERPs returned at least one other feature (three on average), with featured snippets in over half of queries.

So an AI Overview is most likely on an open-ended informational question and least likely on a branded or buy-now query. That’s a behavioral pattern, not a lever — but it’s useful for predicting where you’ll see one.

Sourcing: it’s not “rank #1 and you’re cited”

Here’s where AI Overviews get genuinely interesting, and where my own data has moved the story. The intuition that being cited just means ranking #1 for the exact query is wrong, and it’s getting more wrong over time.

In an original 2025 study we ran on 1.9M citations from 1M AI Overviews, about 76% of cited URLs also appeared in the top 10 organic results. When Xibeijia Guan re-ran it in March 2026 on a 2x sample (863K keyword SERPs, 4M AI Overview URLs), that number had dropped to 37.9%, with the rest split almost evenly between positions 11–100 (31.2%) and beyond the top 100 (31.0%). As the write-up put it: “This indicates that Google is selecting far fewer pages straight from the original SERP (~76% in July 2025 vs. ~38% today)… AI Overviews are relying less on the direct search results and more on the sources showing up in fan out query SERPs.”

That’s the fan-out effect showing up in the data: as Google runs more sub-queries, it pulls citations from those sub-query SERPs rather than your head query’s top 10.

One striking consequence — YouTube gets pulled in independent of ranking. Among AI Overview cited pages that didn’t rank in Google’s top 100 for the query, 18.2% were YouTube URLs; YouTube is now the most-cited domain in AI Overviews and grew 34% over six months. Video content can be surfaced without ranking in the classic blue links at all.

Position and format of the citations

Two more AI-Overview-specific facts about where an Overview sits and how it links.

First, the Overview itself isn’t always at the top. In my July 2025 study of 10M SERPs with AI Overviews (1M per country across 10 countries), “AI Overviews show outside of position 1 for 8.64% of queries. The lowest position we saw them was 6th.” There were country differences too — Japan and Spain had more AI Overviews outside position 1 — which made me wonder “if they have some kind of scoring for how good the generated AI Overview result is.”

Second, the link slot is shared. An Overview typically cites several sources, so any one page’s share of the attention is diluted across the linked set — unlike a single blue-link result, where one URL owns the position. That’s part of why the click math (below) is messy.

What it does to clicks — the honest, contested version

This is where you should be skeptical of anyone selling certainty. The direction is real; the magnitude is contested and moving.

  • The decline. In Ahrefs’ April 2025 study of 300,000 keywords, the presence of an AI Overview correlated with a 34.5% lower average CTR for the top-ranking page, versus similar informational keywords without one.
  • The partial recovery. A Seer Interactive study reported by Search Engine Land in April 2026 (53 brands, 5.47M queries, 2.43B impressions) foundA 302 (\"Found\") is a temporary redirect: it forwards users to a new URL while telling search engines the original URL should stay in the index. It's a weak canonicalization signal, not the zero-equity dead end of SEO folklore. AIO CTR climbed to ~2.4% by February 2026 from a ~1.3% low in December 2025. Being cited within the Overview roughly doubled CTR (~2.1% vs. ~0.9% without a citation) — though both still trailed the ~3.3% CTR when no AI Overview showed at all. (I’d treat SEL as a secondary citation here and check Seer’s original before quoting the exact figures as settled.)
  • Google’s claim, in tension. Sundar Pichai has said “if you put content and links within AI Overviews, they get higher clickthrough rates than if you put it outside of AI Overviews.” That’s directionally consistent with the cited-vs-uncited gap at the per-click level, but it sits in tension with the aggregate CTR decline — so treat it as a claim, not a settled fact.

My own read, from the 55.8M-Overviews study: “Companies investing in their content are also the most likely to show for AIOs. Some may even benefit from them, but the vast majority, I think, will lose traffic.”

Measuring it: the two-report reality in Search Console

A persistent myth: that you can filter Search ConsoleA free Google service that reports how a site performs in Google Search and surfaces problems with how Google crawls, indexes, and serves it. It's first-party data straight from Google — but you don't need it to appear in results. by “Search type = AI Overviews.” You can’t. As of mid-2026 there are two separate places AI feature data lives:

  1. The regular Performance reportThe Google Search Console report that shows how your site actually performed in Google Search, built from real impressions and clicks. It reports four metrics — clicks, impressions, average CTR, and average position — and keeps the most recent 16 months of data., where AI Overview and AI Mode impressions are folded into the “Web” search type — there’s no filter value that isolates them. Google: sites appearing in AI features “are reported on in the Performance report, within the ‘Web’ search type.”
  2. A separate Generative AI performance report, launched June 2026 and rolling out to a subset of properties. It shows impressions only — broken out by Pages, Countries, Dates, and Devices — with no clicks, no CTR, and no query dimension. Google’s help docs: “The generative AI performance report includes data from the Web search type in the Performance report.”

So GSC can now show you visibility inside AI features, but it still can’t isolate AI Overview clicks from ordinary blue-link clicks. As Ahrefs’ Ryan Law noted when the June 2026 report was still a wish, “there is still no way to disambiguate AI Overview clicks and impressions from the rest of your Search ConsoleGoogle's free tool for monitoring crawling, indexing, and search performance. data” — and the new report, being impressions-only, doesn’t fully solve that. The practical workflow is two-layered: use the Generative AI report for impressions/visibility, and estimate click impact indirectly (before/after Performance-report comparisons, or SERP-feature filtering in tools like Ahrefs).

Bing, for its part, shipped a comparable but arguably richer tool — AI Performance in Bing Webmaster Tools (Feb 2026) — which uniquely surfaces the actual grounding queries (the fan-out-style phrases that triggered a citation), something Google doesn’t expose anywhere.

Where this fits

AI Overviews are Google’s specific implementation of the same machinery covered elsewhere in this cluster: groundingGrounding is anchoring an AI model's answer to source documents it retrieves at the moment you ask — not to the patterns frozen into its weights during training. Retrieval-Augmented Generation (RAG) is the most common way to do it. and RAGRAG is the retrieve-then-generate pattern behind AI search: the system retrieves relevant passages from an external index at query time, injects them into the model's context, and generates an answer grounded in those sources — without changing the model's weights. for the retrieve-then-generate pattern, embeddingsEmbeddings are dense numerical vectors — lists of floating-point numbers — that represent the meaning of text in a high-dimensional space. Semantically similar content lands close together, so search and AI systems can match by meaning, not just keywords., semantic searchSemantic search is meaning-based retrieval — matching what a user means, not just the words they typed. Search engines detect entities, expand synonyms, infer intent, and rank by conceptual relevance, which is why keyword stuffing lost its power and topical depth gained it., vector searchVector search finds content by comparing the meaning of a query against stored content as numerical vectors, retrieving the closest ones in a high-dimensional embedding space. At scale it uses approximate nearest neighbor (ANN) algorithms — not exact comparison — to search billions of vectors in milliseconds., and chunkingChunking is splitting a document into smaller passages so AI systems can embed, index, and retrieve the single most relevant piece — not the whole page — in response to a query. It's a foundational step in RAG pipelines and the conceptual cousin of Google's passage ranking. for the retrieval layer, passage rankingPassage ranking is a Google AI system that scores individual sections ('passages') of a page so a single page can earn multiple relevance scores for different queries. Google still indexes whole pages — only the ranking changed. for how a single passage of a page gets surfaced, and AI hallucinationsAn AI hallucination is when a large language model generates output that is confidently stated but factually wrong, made up, or unsupported by its source. It's a side effect of next-token prediction — not a bug that can be fully eliminated. for the failure mode grounding is meant to reduce. For the practical “how do I optimize for this” version — and the null results on schema and llms.txtllms.txt is a proposed (not adopted) Markdown file at /llms.txt that gives AI systems a curated map of a site's most important pages. Proposed by Jeremy Howard in 2024, it's read mostly by coding agents like Claude Code — not search crawlers — and Google ignores it. — see the AI Overviews optimizationAI search optimization is the practice of making your brand and content visible, citable, and accurately represented across AI-powered search — Google AI Overviews, ChatGPT, Perplexity, Copilot. It's built on traditional SEO plus a heavier emphasis on off-site brand mentions and content AI systems can cite. work in the optimization cluster.

The one-line version: AI Overviews didn’t create a new game. They put a Gemini-written front end on the game you were already playing — the one that starts with being crawlable and indexed.

Add an expert note

Pin an expert quote

New person? Create their unclaimed profile at /admin/experts/ → Pin a quote first.