AI Overviews

Google says supporting pages must be indexed and snippet-eligible, with no extra technical requirements. 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: 2026-07-14 Google: AI features and your website Its AI features may also run several related searches through query fan-out. 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: 2026-07-14 Google: AI features and your website

TL;DR — An AI Overview is the AI-written summary box Google shows at the top of some search results. Google’s Gemini model reads a handful of web pages, pulls out the relevant facts, and writes a short answer with links you can click. It’s a feature inside Google Search — not a separate product — and the pages it summarizes come from Google’s normal search index. So the same things that have always mattered (being crawlable, indexed, and clear) are what decide whether your page can be used.

What an AI Overview is

Run a search on Google and you’ll sometimes see a summary at the top, written in full sentences, with a few links off to the side or below. That’s an AI Overview. Instead of only listing blue links, Google uses Gemini to generate an answer from retrieved information and show supporting links. Those sources can come from related fan-out searches, so they are not necessarily the pages that rank for the exact query.

Google’s own description: AI 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.”

The key thing to understand is what it isn’t:

• It’s not a separate search engine or a separate index. The pages in an AI Overview come from Google’s regular search index — the same one that powers the normal results.
• It’s not shown for every search. Google only shows an AI Overview when it thinks the AI answer adds something over the normal results, so plenty of searches don’t get one at all.
• It’s not the same as a featured snippet (the older box that quotes one paragraph from one page). An AI Overview writes a new answer by combining several pages.

How it puts an answer together

Roughly:

1. You search for something.
2. Google may quietly run several related searches behind the scenes (this is called query fan-out) to gather more angles on your question.
3. Gemini reads the top pages from those searches, pulls the relevant facts, and stitches them into one short answer.
4. It shows that answer with links to the sources.

Because it’s reading real pages from the index, a page has to be found and indexed by Google before it can ever be used. There’s no special “AI Overview signup.”

What this means for you

Google is clear that there’s no secret trick here: “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 and eligible to show with a snippet. In other words, do the normal SEO work well and you’re eligible.

Want the real mechanics — query fan-out, how sourcing has changed over time, why AI Overviews aren’t always at the top, and the honest data on what they do to your clicks? Switch to the Advanced tab.

Eligibility still starts with an indexed, 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: 2026-07-14 Google: AI features and your website Google also documents query fan-out 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: 2026-07-14 Google: AI features and your website

TL;DR — AI Overviews 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 grounding, grounding), frequently powered by query fan-out: 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 search 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 crawling 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 optimization and answer engine optimization 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 RAG system — retrieval-augmented generation, which Google itself calls grounding. The general mechanics of retrieve-then-generate belong to the RAG and grounding 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.”

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 features. 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) found 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 Console 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 report, 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 Console 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: grounding and RAG for the retrieve-then-generate pattern, embeddings, semantic search, vector search, and chunking for the retrieval layer, passage ranking for how a single passage of a page gets surfaced, and AI hallucinations 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.txt — see the AI Overviews optimization 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.

AI summary

A condensed take on the Advanced version:

• AI Overviews = Gemini-generated summaries inside Google’s own SERP. A feature of Search, not a separate platform. No separate AI index, no separate AI ranking system — same core index and ranking/quality systems as normal results.
• Eligibility is ordinary: a page must be indexed and snippet-eligible; Google says there are “no additional technical requirements” and “no additional requirements… nor other special optimizations necessary.”
• Mechanism = RAG (grounding) + query fan-out. Google may issue multiple related sub-searches, then Gemini extracts facts from the retrieved pages and stitches them into one cited answer. Depth across sub-questions beats ranking for one head term.
• Triggering is skewed, not random: ~99% informational intent; longer (~4-word) queries; lower Keyword Difficulty (~12 vs ~33); ~10% chance for commercial/transactional; ~72% carry no CPC data. Overviews only show when “additive to classic Search.”
• Not always position 1 — ~8.64% appear elsewhere (as low as 6th).
• Sourcing is loosening from rank: top-10 citation share fell from ~76% (mid-2025) to ~38% (early 2026) as fan-out matured; YouTube is cited heavily and independent of ranking (18.2% of non-top-100 citations).
• Clicks: ~34.5% CTR reduction at position 1 (2025) → partial recovery (~2.4% by Feb 2026); citation within the Overview roughly doubles CTR vs. uncited. Google (Pichai) claims cited links click better — a claim in tension with the aggregate decline.
• Measurement: GSC has no “AI Overviews” filter. AI impressions fold into the “Web” search type; the June 2026 Generative AI performance report shows impressions only (no clicks/CTR/queries). Click impact must be estimated indirectly. Bing’s AI Performance report uniquely shows grounding queries.
• Not a featured snippet (generated prose from many pages vs. one lifted passage) and not AI Mode (a full conversational surface), though both share fan-out.

Official documentation

Primary-source documentation from Google and Microsoft on AI Overviews and the measurement tools around them.

Google

• AI features and your website — how AI Overviews and AI Mode work, eligibility, query fan-out, and where the data shows up in Search Console.
• Google’s Guide to Optimizing for Generative AI Features on Google Search — the May 2026 guide: “still SEO,” the RAG/grounding definition, and the mythbusting section.
• Introducing Search Generative AI performance reports in Search Console — the June 2026 launch of the separate, impressions-only Generative AI performance report.
• Generative AI performance report (Search) — Search Console Help — dimensions (Pages, Countries, Dates, Devices), and the note that it shows impressions, not clicks/CTR/queries.

Bing / Microsoft

• Introducing AI Performance in Bing Webmaster Tools (Public Preview) — Bing’s comparable measurement tool, including the “grounding queries” view Google doesn’t expose.

Quotes from the source

On-the-record statements from Google and Microsoft. Each link is a deep link that jumps to the quoted passage on the source page where the anchor is supported.

Google — what AI Overviews are and how they source

• “AI 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.” — Google Search Central, AI features and your website. Jump to quote
• “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.” Jump to quote
• “While responses are being generated, 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…” Jump to quote

Google — eligibility and triggering

• “There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary.” Jump to quote
• “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, fulfilling the Search technical requirements. There are no additional technical requirements.” Jump to quote
• “AI Overviews are only shown when our systems determine that it is additive to classic Search, and as such, often don’t trigger.” Jump to quote

Google — it’s still the core index (May 2026 guide)

• “In short, yes! The best practices for SEO continue to be relevant because our generative AI features on Google Search are rooted in our core Search ranking and quality systems.” Jump to quote
• “From Google Search’s perspective, optimizing for generative AI search is optimizing for the search experience, and thus still SEO.” Jump to quote
• “Retrieval-augmented generation (RAG): 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.” Jump to quote

Google — measurement in Search Console

• “…they’re reported on in the Performance report, within the ‘Web’ search type.” — Google Search Central, AI features and your website. Jump to quote
• “The generative AI performance report includes data from the Web search type in the Performance report (Search results).” — Search Console Help. Jump to quote

Microsoft Bing — the comparable tool

• “We are happy to introduce AI Performance in Bing Webmaster Tools, a new set of insights that shows how publisher content appears across Microsoft Copilot, AI-generated summaries in Bing, and select partner integrations.” — Bing Webmaster Blog, February 2026. Read the announcement

The mental models

1. Feature, not platform. An AI Overview is a Gemini-written front end over Google’s normal results. There’s no separate AI index and no separate AI ranking system. If a page can’t be crawled, indexed, and shown with a snippet, it can’t appear — full stop. Fix the crawl → index chain before you think about anything “AI-specific.”

2. Grounding + fan-out. The answer is generated by RAG (grounding) over retrieved pages, and Google can fan out into several sub-queries first. So the unit of competition isn’t “rank #1 for the head term” — it’s “be a good answer to the sub-questions the model decomposes your topic into.”

3. Being cited ≠ ranking #1. The correlation between a top-10 organic ranking and being cited fell from ~76% (2025) to ~38% (2026). Ranking well helps, but fan-out increasingly pulls citations from sub-query SERPs — and some formats (notably YouTube) get pulled in with no organic ranking at all.

4. The click picture is contested — hold it loosely. Aggregate CTR dropped (~34.5%), then partially recovered; being cited roughly doubles CTR vs. being an uncited source; Google claims cited links click better than blue links. All of these can be partly true at once. Don’t anchor a strategy to any single number.

5. Measurement is two-layered and incomplete. GSC has no “AI Overviews” filter. Impressions fold into the “Web” search type; the separate Generative AI report adds impressions-by-dimension but no clicks/CTR/queries. Treat visibility (measurable) and click impact (estimated) as two different questions.

AI Overviews — cheat sheet

What it is in one line A Gemini-generated summary box inside Google’s SERP, written from pages retrieved out of Google’s normal index — a feature of Search, not a separate platform.

AI Overviews vs. featured snippets vs. AI Mode

How it works (pipeline) query → (optional) query fan-out into sub-searches → retrieve pages from the core index → Gemini extracts + stitches facts → generated answer with citations

Triggering — where you’ll see one

• Skews informational (~99% of triggering keywords), longer-tail (~4 words), lower difficulty (~KD 12).
• ~10% chance on commercial/transactional; ~72% of AIO searches have no CPC data.
• Only shown when “additive to classic Search,” so many queries get none.
• Not always position 1 — ~8.64% appear elsewhere (as low as 6th).

Sourcing facts

• Top-10 citation share: ~76% (mid-2025) → ~38% (early 2026) as fan-out matured.
• YouTube is the most-cited domain; 18.2% of non-top-100 citations are YouTube.

Click facts (contested)

• ~34.5% CTR drop at position 1 (2025) → partial recovery (~2.4% by Feb 2026).
• Cited within the Overview ≈ 2x CTR vs. uncited (~2.1% vs ~0.9%); both below ~3.3% for no-AIO SERPs.

Measurement

• No “AI Overviews” filter in GSC. Impressions fold into the “Web” search type.
• Generative AI performance report (June 2026): impressions only; Pages/Countries/Dates/Devices; no clicks/CTR/queries.
• Bing AI Performance report uniquely shows grounding queries.

Which thing am I actually looking at — and how do I track it?

A quick way to tell AI Overviews apart from the lookalikes, and to route your measurement.

“Is this box an AI Overview, a featured snippet, or AI Mode?”

1. Is it a full conversational search experience with a chat-style back-and-forth on its own surface? → That’s AI Mode, not an Overview on the regular SERP.
2. Is the text a single quoted paragraph attributed to one page (with that page’s normal blue-link title)? → That’s a featured snippet — one lifted passage from one ranking page.
3. Is the text newly written, in full sentences, drawing on several linked sources at once? → That’s an AI Overview.

“I want to know how my site is doing in AI Overviews — where do I look?”

1. Do you want visibility (were my links shown)? → Use the Generative AI performance report in Search Console (if your property has it yet). Impressions by Pages / Countries / Dates / Devices. No clicks, no CTR, no queries.
2. Do you want click impact (did the Overview cost or add traffic)? → GSC can’t isolate this. Estimate it: compare Performance-report clicks/CTR before-and-after for AIO-triggering queries, or filter by the “AI Overview” SERP feature in a tool like Ahrefs Site Explorer, then watch position-1 CTR on those keywords.
3. Do you want the query phrases that triggered a citation? → Google doesn’t expose them. Bing’s AI Performance report shows “grounding queries.”

“My page ranks #1 but isn’t cited — is something broken?”

1. Is the page indexed and snippet-eligible? (URL Inspection.) → If no, that’s the whole problem; fix eligibility first.
2. If yes: nothing is necessarily broken. Only ~38% of citations now come from a top-10 ranking; fan-out pulls from sub-query SERPs. Cover the topic’s sub-questions rather than chasing the single head term.

Measure AI Overviews as a SERP feature

Google does not provide a clean AI Overview traffic segment in standard reporting. Treat feature observations and Search Console outcomes as related evidence, not perfect request-level attribution.

Resources worth your time

My related writing & research

• Insights From 55.8M AI Overviews Across 590M Searches — my largest-scale AIO study: prevalence, triggering patterns, and top-cited domains.
• 8.64% of AI Overviews Appear Outside Position #1 (And as Low as Position #6) — my position study across 10M SERPs and 10 countries.
• 76% of AI Overview Citations Pull From the Top 10 — the original citation-vs-ranking correlation study (1.9M citations).

My speaking

• GEO? AEO? LLMO? What’s With All This AI SEO Stuff? — my Ahrefs Evolve 2025 talk on the AI search landscape and why the indexing prerequisite hasn’t changed. (My standing disclaimer applies: this is my understanding of these systems, not a guaranteed-complete account.)

From around the industry

• Update: 38% of AI Overview Citations Pull From The Top 10 — Xibeijia Guan (Ahrefs); the March 2026 re-run on a 2x sample showing the correlation loosening and YouTube’s outsized share.
• I Analyzed 300K Keywords. Here’s What I Learned About AI Overviews — Louise Linehan (Ahrefs) on the foundational triggering-conditions data (intent, length, KD, SERP-feature co-occurrence) that Patrick’s later studies build on.
• AI Overviews Reduce Clicks by 34.5% — Ryan Law & Xibeijia Guan (Ahrefs); the flagship CTR-impact study and the “no way to disambiguate AI Overview clicks in Search Console” point.
• The Great Decoupling (or Why Your Clicks Are Down and Impressions Up) — Ryan Law (Ahrefs) on the clicks/impressions divergence AI Overviews drive.
• Google AI Overviews CTR shows early signs of recovery: Study — Search Engine Land on Seer Interactive’s 2026 recovery data (cited-vs-uncited CTR gap).
• Google’s New AI Search Guide Calls AEO And GEO “Still SEO” — Search Engine Journal on the May 2026 optimization guide and its mythbusting section.
• AI Overviews optimization guide: How to rank in generated results — Search Engine Land’s practical guide (compare its per-rank citation-probability figures against Ahrefs’ aggregate top-10 share — they measure different things).

Stats worth citing

• ~99.2% of AI-Overview-triggering keywords are informational, and AIOs have only a ~10% chance of showing for commercial/transactional queries. Source
• AIOs show for ~9.46% of keywords on desktop (16% in the US), and lean toward informational, longer, higher-volume, non-commercial queries — ~72% of AIO searches carry no CPC data. Source
• ~8.64% of AI Overviews appear outside position 1 (as low as position 6), from 10M SERPs across 10 countries. Source
• Top-10 citation share fell ~76% → ~37.9% between mid-2025 and early 2026 as query fan-out matured; 18.2% of non-top-100 citations were YouTube URLs. Source
• ~34.5% lower position-1 CTR when an AI Overview is present, across 300,000 keywords. Source
• AIO CTR recovered to ~2.4% by Feb 2026 from a ~1.3% low; being cited within the Overview roughly doubled CTR vs. uncited (Seer Interactive, via Search Engine Land). Source

Test yourself: AI Overviews

Five quick questions on how Google AI Overviews work. Pick an answer for each, then check.

AI Overviews

AI 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.

Related: Retrieval-Augmented Generation (RAG), Grounding

AI Overviews

AI Overviews are the AI-generated summary Google shows at the top of (or within) its regular search results, generated by Gemini models built into Search. The important framing: it’s a feature of Google Search, not a separate product or a separate index. The pages it summarizes and links to come from Google’s normal Search index, chosen by the same core ranking and quality systems that decide who ranks — Google’s own guidance is blunt that “our generative AI features on Google Search are rooted in our core Search ranking and quality systems.”

Mechanically, AI Overviews are a RAG (retrieval-augmented generation, which Google calls grounding) feature. To answer, Google can run query fan-out — issuing several related sub-searches across subtopics, then having Gemini extract facts from the retrieved pages and stitch them into a single answer with links. To be eligible to be cited, a page only has to be indexed and eligible to appear with a snippet; Google says there are “no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary.”

Three things trip people up. AI Overviews are not featured snippets — a snippet lifts one verbatim passage from one page, while an Overview generates new prose from many pages. They are not AI Mode — a separate, full conversational search experience — though both share the fan-out mechanism. And they don’t always sit in position 1: Ahrefs found they appear outside the top spot for about 8.64% of queries. Overviews trigger disproportionately on informational, longer-tail, non-commercial queries, and “are only shown when our systems determine that it is additive to classic Search, and as such, often don’t trigger.”

Related: Retrieval-Augmented Generation (RAG), Grounding

Changelog