What is generative engine optimization (GEO)?
How AI assistants choose which sources to cite, and what you can change.
On this page
Short answer
Query fan-out is the way Google’s AI search treats one question as several. The system breaks the question into smaller parts, runs a separate search for each part at the same time, and writes one answer from everything those searches bring back. Google uses the term for AI Mode and AI Overviews. For a website owner the key effect is this: your page can be picked for a part of the question the person never typed.
What Google has said about it
Google named the technique when it presented AI Mode at Google I/O on May 20, 2025. In the announcement, Elizabeth Reid, head of Google Search, wrote that AI Mode uses query fan-out to split a question into subtopics and send many queries at once on the user’s behalf. Google said this lets Search go deeper into the web than a classic search. The same post described Deep Search as the same technique at a larger scale, able to run hundreds of searches for one request.
Google’s Search Central page on AI features adds the details that matter for site owners:
- Both AI Overviews and AI Mode may use the technique, running related queries across subtopics and data sources.
- Because of this, the answer can show a wider and more varied set of links than a classic result.
- No special optimization is required. To be shown as a supporting link, a page must be indexed and eligible to appear in Search with a snippet.
What Google does not publish: which sub-queries it runs for a given question. Any tool that claims to show you the exact list is making an estimate.
An example of the idea
Here is how one question might fan out. This is our illustration of the principle, not data from Google.
The question: “Are dental implants worth it at 60, and what will they cost?”
| Possible sub-question | Page type that can answer it |
|---|---|
| typical price range for one implant | price page with a table |
| what changes the price | page section on bone grafting, materials, number of visits |
| age and health limits | treatment page with a section on who it suits |
| implants compared with bridges or dentures | comparison guide |
| how long treatment takes | process page with steps and timing |
| clinics nearby with good reviews | Business Profile, review sites, local directories |
A clinic with a single “Implants” page that says “contact us for a quote” covers none of these well. A clinic with a clear price range, a section on what affects cost and a short comparison has a chance in four or five places.
What it means for a service business
Your service page no longer competes only for the main phrase. It competes for each supporting question a buyer would ask a friend before choosing. Pages that answer those questions plainly get extra chances to be linked.
Our own site shows a small version of this. The pages that earned AI Overview links on aoe.kz in our fixed Russian-language Almaty query set (details in the AI citations guide) open with a direct answer and continue with a table and a block of questions. We cannot see which sub-query triggered each link. Google does not report that, and we would rather say so than guess.
Google named the technique when it presented AI Mode at Google I/O on May 20, 2025.
How we map fan-out for a client
- Start from the buyer’s
main question, worded the way a customer would ask it, not as a keyword.
- Collect the likely sub-questions
from “People also ask”, related searches, the links that the current AI Overview shows, and keyword ideas from Google’s Keyword Planner.
- Check each sub-question against the site
Is there a sentence that answers it directly? Where is it?
- Close the gaps
A small sub-question gets a section with its own heading on an existing page. A sub-question with its own search demand gets its own page.
- Re-run the main question
every month and record which URLs the AI answer links, yours and your competitors’.
The output is a simple table: sub-question, where the answer lives, who is linked today. It turns a vague goal like “rank in AI” into a list of pages and sections to write.
Common mistakes
- One giant page. Trying to answer every sub-question on one URL tends to give vague answers to all of them. Split by topic.
- Keyword variants. Repeating the main phrase in ten forms does not cover sub-questions. Different questions need different facts.
- Presenting guesses as data. Sub-query lists from tools are estimates. Label them that way in reports.
- Forgetting the basics. Google states that a page must be indexed and snippet-eligible. A page blocked by noindex or nosnippet cannot be one of the links, however good it is.
- Facts in pictures. A price table saved as an image gives the system nothing to quote. Keep facts as text.
FAQ
Is query fan-out the same as grounding?
They are two steps of one process. Fan-out is the search step: deciding what to look up. Grounding is the writing step: building the answer from what was found and linking to it. See our grounding guide.
Do ChatGPT and Perplexity do the same?
Google is the one using the name. Its Gemini API documentation says the model can write one or more search queries for a single prompt. OpenAI’s documentation says its model decides when to search. Other vendors do not describe sub-queries in the same terms, so we treat a similar approach as likely but not confirmed.
Can I see the fan-out queries for my site?
Not in Search Console. Google counts AI Overviews and AI Mode traffic inside the “Web” search type with no breakdown by sub-query. Developers building on the Gemini API can see the queries their own app runs, but that is a separate product, not Google Search.
Where do I start?
Pick your three most valuable services and map their sub-questions with the steps above. For page layout, use our AEO checklist. For the terms, see the AI search glossary. For why fewer people click, read about zero-click search.
