How does Google pick the sources it cites in an AI Overview?

It breaks your question into several sub-queries (query fan-out), retrieves pages for each, runs them through an E-E-A-T quality filter, reranks them with Gemini at the passage level, and writes the summary citing the ones that recur most and answer best. On average it cites about 8 sources. It rewards self-contained passages that answer fast, with high fact coverage and schema.

Is showing up in an AI Overview the same as ranking in the top 10?

Not anymore. Per an analysis covered by Search Engine Journal, only around 38% of pages cited in AI Overviews are in the organic top 10, versus 76% less than a year earlier. The top 10 still helps, but passage-level citability (answering the exact question in an extractable block) now weighs as much as position.

How do I get Google to cite me in an AI Overview?

Five levers: answer the question in a self-contained passage of ~130-170 words, raise your fact coverage (cited pages cover more facts: 0.31 vs 0.24 per ZipTie's reverse-engineering), densify Knowledge Graph entities, clear the E-E-A-T threshold with authorship and verifiable data, and mark it up with schema. And measure: watch which queries surface an AI Overview and whether you break in.

How to show up in Google AI Overviews (and not vanish from the click)

Q: How much do clicks drop when there's an AI Overview?

A lot. An Ahrefs study put the click drop at around 34.5% on the query that triggers the AI Overview. SE Ranking found 43% of searches with an AI Overview end in no click at all, versus 34% without one. That's why being inside the summary stopped being an extra: it's defending the traffic the summary keeps.

Google AI Overviews is the costliest change to ignore of everything AI has brought to search. When a generated summary shows up above the results, the first blue link loses roughly a third of its clicks, and nearly half of those searches end in no click at all. Showing up inside that summary stopped being a bonus: it's defending the traffic the summary keeps for itself. The good news: the engine has rules, and you can work them.

Why Google AI Overviews isn't ChatGPT or Perplexity

All three generate answers, but AI Overviews plays a different game. ChatGPT, unless it's browsing, answers from memory; Perplexity searches live and shows its sources. AI Overviews lives inside the world's most-used search engine and leans on everything Google already had: its index, its Knowledge Graph and E-E-A-T. It's not a chat you go to — it's a layer wedged between the user and the ten links they used to see.

That difference has a direct consequence for you. On ChatGPT you fight to get into the model's weights; on Perplexity, to be one of the pages it retrieves. On AI Overviews you fight to be one of the passages Google extracts from its own index to build the summary — and it does that on the same old SEO signals, filtered by an AI layer deciding who it shows and who it hides.

How Google decides who to cite

Under the hood, AI Overviews works like a retrieval-and-generation system, but with its own trick: query fan-out. Instead of answering your question as-is, Google breaks it into several related sub-queries, retrieves pages for each, reranks them with Gemini at the passage level (not the whole page), and writes the summary citing the sources that recur most and answer best. On average, an AI Overview cites around eight sources.

The uncomfortable fact: ranking isn't enough anymore

For years, getting to the top was the goal. With AI Overviews, getting to the top no longer guarantees a citation. Per an analysis covered by Search Engine Journal, only around 38% of pages cited in AI Overviews sit in the organic top 10 — when less than a year earlier that figure was 76%. The top 10 still helps, but it stopped being the entry ticket.

Passage-level citability weighs as much as position. Google extracts specific blocks, not whole pages. If your answer is buried in paragraph nine, ranking won't save you: it can't find it for the summary.
Fact coverage decides. ZipTie's reverse-engineering found cited pages cover more facts than non-cited ones (a Fact Coverage of 0.31 versus 0.24, 29% more). Half-answering no longer cuts it.
Entities matter. Cited pages tend toward a high density of Knowledge Graph entities (on the order of 15+ per 1,000 words). Speaking Google's graph language makes you legible to its AI.

The levers that actually get you into the AI Overview

There aren't ten tricks: there are five things that move the needle and a lot of repeated filler. Here are the five, in order of impact.

Answer in a self-contained passage. The pages Google extracts for its summaries answer the question in short, complete blocks — on the order of 130 to 170 words — that make sense without surrounding context. Front-load the answer and make each section resolve a clear intent.
Raise your fact coverage. If your page covers more verifiable facts on the topic than the next one, you start ahead: cited pages cover more facts. Don't inflate words; add real information.
Densify Knowledge Graph entities. Name the relevant people, products, places and concepts in your category explicitly. The more recognizable entities, the easier you make it for Gemini to connect you to the query.
Clear the E-E-A-T threshold. Clear authorship, verifiable data, cited sources and signals that you know your stuff. It's a binary filter: clear it or you're not even considered for the citation.
Mark it up with schema and keep it fresh. Structured data (FAQPage, HowTo, Article) gives Google the map of your content, and freshness is a signal: a visible date and periodic revisions help it look at you again.

The cost of not being there (and why to measure it)

It's worth being clear on the numbers, because they explain the urgency. An Ahrefs study put the click drop at around 34.5% on the query that triggers the AI Overview. SE Ranking found that 43% of searches with an AI Overview end in no click at all, versus 34% when there's no summary. Translation: if your category triggers AI Overviews and you're not inside, you don't just lose some traffic — the summary keeps it before the user ever reaches you.

So the first thing is to measure, not guess. Look at which queries in your space surface an AI Overview, which pages it cites and whether you're among them. That's your real competition. When you ship or update something, watch over the following weeks whether you break into those citations. In GEO there's no official dashboard: you build your own — and whoever builds it decides on data while the rest opine.