What are Google AI Overviews?
AI Overviews are the AI-written answers Google places at the very top of many search results, generated by a custom version of its Gemini model. Instead of returning ten blue links, Google composes a short synthesized answer and attaches a set of source links it leaned on. As of Google's July 2025 earnings call, AI Overviews reached 2 billion monthly users across 200 countries — this is now the default top of the page, not an experiment.
Wasn't this called Google SGE?
Yes. Google SGE — the Search Generative Experience — was the 2023 Labs opt-in test. In May 2024 it graduated into search proper as AI Overviews, and it has expanded across queries, countries, and industries since, followed in 2025 by the fully conversational AI Mode. If you're reading advice written for "SGE," the underlying mechanics still apply. What changed is the stakes: an opt-in experiment became the surface hundreds of millions of people see first. Which raises the mechanical question — how do AI Overviews work once you hit enter?
Query fan-out: the engine behind AI Overviews
The single most important thing to understand about how AI Overviews work is that Google does not answer your query with one search. It runs many. Google documents this plainly in Search Central. In its own words, "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."
Here's what that means in practice. A search like "best CRM for small business" fans out into separate searches for pricing, feature comparisons, integrations, alternatives, and "is it worth it" — each run against Google's index, each returning its own results. The overview is then assembled from passages pulled across all of those sub-queries, not just the one you typed.
Google adds that during generation, "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." Translation: the pool of pages eligible to be cited is larger than the page-one results for your visible query. A page that answers only the head term competes in one retrieval; a page — or cluster — that also answers the fan-out sub-questions competes in five. That is the strategic core of AI Overview optimization.
How do Google AI Overviews choose sources?
Not by copying the top ten. This is the finding that reorders everything. Ahrefs analyzed 863,000 keyword SERPs and 4 million AI Overview URLs and found that only 38% of pages cited in AI Overviews also rank in the top 10 for the same query — down from roughly 76% a year earlier. The remaining citations were split almost evenly between positions 11–100 and pages beyond the top 100.
Read that again: the majority of AI Overview citations now come from pages that are not on page one of the visible search. Ahrefs attributes the drop directly to fan-out — Google is pulling less from the original SERP and more from the sub-query SERPs. So the old goal ("rank #1 and you'll be featured") is no longer sufficient, or even necessary. Being retrieved by a fan-out query and offering the most groundable passage now matters at least as much as your ranking position.
This is why smaller, sharper sites regularly out-cite bigger competitors that outrank them — they answered a specific sub-question cleanly, and the model could lift a clean sentence.
How do AI Overviews work step by step: query to citation
Here is how one search becomes an AI answer, and where your content competes at each stage.
| Step | What happens | Your leverage point |
|---|---|---|
| 1. Trigger | Google decides whether the query benefits from an AI answer | Informational and comparative queries trigger most |
| 2. Query fan-out | The query expands into related sub-queries, run in parallel | Cover the sub-questions, not just the head term |
| 3. Retrieval | Candidate passages are pulled from pages ranking across all those queries | Being indexed and rankable is table stakes |
| 4. Gemini grounding | The model drafts an answer constrained to the retrieved passages | Clear, self-contained, factual passages get used |
| 5. Citation assembly | Sources whose passages supported the claims fill the citation slots | Quotable, specific claims win the slots |
| 6. Display | The answer plus link cards render above the organic results | Titles and metas still earn the click |
"Grounding" in step 4 is the key term. Gemini does not free-associate; it is constrained to write answers it can support from the retrieved passages, and Google then displays the pages that did the supporting. That is the whole mechanic of how AI Overviews choose sources: the citation goes to the page that supplied the sentence, not the page with the most backlinks.
"The overview isn't quoting the most popular page. It's quoting the most usable passage. Write the sentence the machine can lift — specific, self-contained, and true — and you become the source it can't route around." — Thomas, Founder of AISEO USA, 16 years in digital marketing
What AI Overviews do to your traffic: the zero-click effect
Because the answer sits above position #1, the citation is now the prize — and the click is scarcer than it was. Pew Research studied real browsing behavior and found that Google users who encountered an AI summary clicked a traditional search result in just 8% of visits, versus 15% when no summary appeared — roughly half as many clicks. Only 1% of visits resulted in a click on a link inside the AI summary itself.
So AI Overviews compress downstream clicks across the board. The businesses that hold visibility are the ones cited in the overview, where the reader actually looks. Pew also found that 88% of AI summaries cited three or more sources — meaning there are usually several citation slots per answer, and they're winnable. The question is whether your page is structured to fill one.
How often do AI Overviews appear?
Prevalence is volatile, so treat any single number with care. Pew found that around one-in-five Google searches in March 2025 produced an AI summary. Semrush's study of 10 million keywords tracked the trigger rate rising from 6.49% of queries in January 2025 to 24.61% in July, then settling to 15.69% in November. Coverage also shifted by intent: informational queries made up 91.3% of AI Overview triggers in January but only 57.1% by October, as commercial and transactional queries pulled in overviews too. The takeaway isn't a precise percentage — it's that AI Overviews now reach into the money queries, not just the "what is" ones.
What makes a page citable in an AI Overview?
Across the AI-visibility audits AISEO USA runs, the pages that earn citation slots share the same handful of traits. Treat these as probability-raisers, not switches — no one can guarantee an AI citation, and anyone who promises one is guessing.
Notice that these overlap almost entirely with the fundamentals of answer engine optimization and generative engine optimization. The same extractable, entity-clear structure that wins an AI Overview citation also wins featured snippets and citations in ChatGPT and Perplexity — one investment, compounding across surfaces.
How to optimize for AI Overviews
So how do AI Overviews work in your favor? You work the pipeline in order:
- Confirm eligibility. Fix technical health and indexation first — a page Google can't crawl can't be retrieved or cited.
- Restructure high-intent pages into extractable answers. Question-form headings, a direct 40–80 word answer under each, tables, and visible update dates.
- Build out fan-out coverage. Create the pricing, comparison, and "is it worth it" content around each money page so you're eligible in more sub-queries.
- Add entity signals. Schema, consistent NAP and business facts, and a credible author so attribution is unambiguous.
- Maintain freshness. Update and re-date pages; overviews re-retrieve, and stale pages quietly drop out.
This is the day-to-day work of modern AI SEO services, and it's measurable — you can see which of your queries trigger an overview and whether you're cited. The honest framing is that it's probabilistic: you raise the odds and track the results, you don't buy a slot.
FAQ
How do AI Overviews work?
Google runs query fan-out — it expands your search into several related sub-searches, runs them all, and retrieves passages from the results. A custom Gemini model then writes one answer grounded in those passages, and the pages whose passages supported the claims are shown as citation links above the organic results.
Do you have to rank #1 to be cited in an AI Overview?
No. Ahrefs found that only 38% of pages cited in AI Overviews also rank in the top 10 for the same query, down from about 76% a year earlier. Citations frequently go to pages that win a fan-out sub-query or supply the clearest groundable passage, so well-structured smaller sites regularly out-cite bigger competitors that outrank them.
What is query fan-out in AI Overviews?
Query fan-out is Google's technique of issuing multiple related searches across subtopics and data sources to build one AI answer, described in Google's own Search Central documentation. In practice it means covering the questions around your topic — pricing, comparisons, alternatives — multiplies the number of retrievals your content can win, and therefore your chances of being cited.
How often do AI Overviews appear on Google?
It varies by dataset and query mix. Pew found around one-in-five searches produced an AI summary in March 2025, while Semrush tracked the trigger rate moving from 6.49% of keywords in January 2025 up to 24.61% in July and back to 15.69% in November. The share is volatile, but overviews now reach commercial queries, not just informational ones.
Do AI Overviews hurt website traffic?
They reduce downstream clicks. Pew measured that users clicked a search result in just 8% of visits when an AI summary appeared, versus 15% without one, and only 1% clicked a link inside the summary. The counter-move is to be cited in the overview itself — where 88% of summaries list three or more sources, meaning multiple slots are usually available.
What makes a page more likely to be cited in an AI Overview?
A direct, self-contained answer of about 40–80 words under a question-form heading; verifiable specifics like numbers, dates, and named entities the model can ground claims against; coverage of the fan-out sub-questions; clean structure with descriptive headings and tables; clear entity and schema signals; and freshness. These raise the probability of a citation — none of them guarantees one.
Are AI Overviews the same as Google SGE?
They're the same lineage. SGE (Search Generative Experience) was the 2023 opt-in Labs experiment; it launched into search as AI Overviews in May 2024 and later gained the conversational AI Mode. Advice written for SGE still broadly applies, but the mechanics have matured and the feature is now the default top of the page rather than an experiment.
How do I check if AI Overviews cite my business?
Search your most important money queries and inspect the citation cards to see who Google's AI is pulling from — or run a free AI visibility audit, which systematically checks what Google's AI Overviews, ChatGPT, and Perplexity say about your business across your key queries and returns the gap list in one report.