Prism splitting a single search query into multiple sub-queries that converge into one AI Overview answer with cited sources

Blog

What Is a Google AI Overview, and How Does It Choose What Goes In?

BalochDev 8 min read

A Google AI Overview is the AI-generated summary that appears at the top of many Google search results, powered by Google's Gemini model. It decides what goes in using a technique called query fan-out: it breaks your question into many sub-questions, searches each one at the same time, and stitches the best passages into a single answer with links to the sources. Crucially, it selects passages, not whole pages — which is why a page ranked lower on Google can be cited over the #1 result.

If you want your business to appear in these answers, you first have to understand how the machine builds them. Here's the mechanism, the selection criteria, and what it means for whether you get cited.

What an AI Overview actually is

An AI Overview is a short, AI-written summary that sits above the traditional blue links for many queries, pulling together information from multiple sources and linking to them. It launched broadly in the United States in May 2024 and now spans more than 200 countries and 40+ languages. At Google's May 2026 I/O keynote, Sundar Pichai said AI Overviews had crossed 2.5 billion monthly active users — a scale most publishers cannot afford to ignore.

Under the hood, it runs on Gemini, Google's family of AI models built by Google DeepMind. As of January 27, 2026, Google made Gemini 3 the default model powering AI Overviews globally, replacing the earlier setup where only the hardest queries were routed to it — the same model that runs its conversational search.

AI Overview vs AI Mode vs Gemini (they're not the same thing)

People confuse three Google products constantly, so let's separate them cleanly, because they share one engine but behave differently:

  • AI Overview — the AI summary above your search results. Non-interactive, one answer per query.

  • AI Mode — the dedicated conversational search experience, where you ask follow-ups.

  • Gemini — the standalone AI assistant app.

All three pull from the same Google index, use the same retrieval, and rely on the same Gemini reasoning. That's the good news for you: optimizing for one largely optimizes for all three. (Google announced in May 2026 that AI Overviews and AI Mode are converging into one search experience — the boundary is blurring, but the optimization work stays the same.)

How it works: query fan-out

Here's the core mechanism. When you type one question, the system doesn't run one search. It fans the query out into many related sub-questions and searches them concurrently, then synthesizes the results into one answer.

Ask "which smart ring is best for sleep tracking," and instead of a single lookup, the system quietly researches battery life, sensor accuracy, comfort, price, app quality, and comparisons — each as its own sub-query — then assembles one summary from across those results. Google's own documentation describes the pattern plainly: the model analyzes the prompt, generates one or multiple search queries, processes the results, and returns a grounded answer with inline citations. Analyses of the system suggest a typical query fans out into roughly 8–12 sub-queries, with complex "deep" questions issuing far more.

Two consequences follow, and they rewrite the rules of visibility:

  • Breadth beats narrow targeting. Because the system searches many sub-questions, a page that covers a topic thoroughly can be pulled in for sub-queries you never set out to rank for.

  • The top spot is no longer everything. A lower-ranked page can be cited if it answers a specific sub-question better than the page above it.

The key insight: it picks passages, not pages

This is the single most important thing to understand. Query fan-out doesn't pull pages — it pulls passages. The system extracts the specific block of text that best answers each sub-question and quotes that.

That's why the sources cited in AI Overviews often aren't the top organic results. BrightEdge's February 2026 tracking found that only 17% of AI Overview citations come from pages ranking in the organic top 10 — consistent with the broader pattern we cover in Why Ranking #1 on Google No Longer Gets You Cited by ChatGPT, where independent studies put the overlap between top rankings and AI citations at roughly 12–20%. A tightly written 40-to-60-word passage that directly answers a specific sub-question is more citation-likely than a rambling 3,000-word article with no clear answer boundaries. Some studies of AI Overview sources have found self-contained passages in the ~130–170-word range earn the highest citation rates.

The practical takeaway: structure your content into clean, standalone, answer-first blocks. You're not writing one page to rank for one keyword — you're writing many quotable passages, each ready to be lifted for a different sub-question.

What the system weighs when choosing sources

Beyond passage-level relevance, source selection leans on a familiar cluster of signals, now applied to AI answers:

  • Semantic relevance — how closely a passage matches the meaning of the sub-question, not just its keywords.

  • Freshness — AI Overviews favor recent content. One analysis found cited pages tend to be meaningfully fresher than typical organic results, so a visible, recent update date matters.

  • Structure and extractability — clear headings, direct answers, lists, tables, and Q&A blocks make a passage easy to lift.

  • Authority and entity recognition — the system prefers sources it recognizes and trusts, and content that names entities clearly.

  • Structured data — schema markup helps the machine confirm what a page is and who published it.

None of these are exotic. They're the same qualities that make content genuinely useful — the AI is just rewarding them at the passage level now.

Why this matters for your traffic

AI Overviews are consuming clicks. Multiple independent studies reached the same conclusion in 2026: when an AI Overview appears, click-through to the top organic result drops sharply. Ahrefs' December 2025 analysis of 300,000 keywords found a 58% reduction in click-through rate for the top-ranking page — up from a 34.5% drop measured just eight months earlier — and Seer Interactive separately found organic CTR falling around 61% on informational queries. The direction is unanimous even where the magnitude varies.

So the goal shifts. Ranking #3 on a query that triggers an AI Overview is worth less than being the cited source inside it. Being in the answer is the new front-page position.

How to earn a spot in an AI Overview

Pulling the mechanism together into action:

  1. Make sure Google can fully render and read your page. If content is hidden behind JavaScript, it may not be there when the system builds the answer. Run our 5-minute reachability check if you're unsure.

  2. Cover the whole topic, not one keyword. Anticipate the sub-questions a fan-out would generate and answer each in its own clean section.

  3. Lead every section with a direct, quotable answer in the first 40–60 words.

  4. Keep it fresh. Update important pages and show the date.

  5. Add structured data and name your entities so the machine can verify and trust your content.

Google's official position is that there's no special markup just for AI Overviews — you earn inclusion by following strong Search best practices and being genuinely helpful and reliable. The full playbook is in our pillar guide, How to Get Your Business Found by AI Search in 2026.

What if you don't want to appear?

Worth knowing: you have controls. Google introduced a Search Console opt-out toggle in June 2026 — under Settings → Search generative AI — that lets you remove your content from AI Overviews, AI Mode, and AI Overviews in Discover without affecting your normal rankings; Google states this setting is not used as a ranking signal. It rolled out UK-first, under a binding order from the UK's Competition and Markets Authority, with global expansion expected to follow. The data-nosnippet attribute separately lets you exclude specific passages from being used. Most businesses want in, but the opt-out exists.

Frequently asked questions

Do I need special markup to appear in AI Overviews? No. Google states there's no dedicated markup for AI Overviews. You earn inclusion through normal Search best practices — helpful, reliable, well-structured, crawlable content — plus structured data that helps machines understand your page.

Why is a lower-ranked page cited instead of the #1 result? Because AI Overviews select passages, not pages. If a lower-ranked page contains a passage that answers a specific sub-question more cleanly, it can be cited over the top-ranking page.

What model powers Google AI Overviews? Gemini. As of January 2026, Gemini 3 is the default model for AI Overviews globally — the same model behind Google's AI Mode.

Is optimizing for AI Overviews different from optimizing for AI Mode or Gemini? Largely no. All three share Google's index, retrieval, and Gemini reasoning, so improving one tends to improve visibility across all three.

This guide is maintained by BalochDev, an AI-first software development studio. We build products — and the sites that sell them — to be found by both people and machines. Last updated 9 July 2026.

Sources & further reading