Query fan-out means an AI search engine doesn't answer a question with a single search. It splits the question into subtopics, sends several related searches at the same time, and builds one answer from the results. Google uses it in AI Mode and AI Overviews, and similar patterns show up in ChatGPT and Gemini. For you, that changes the game: you can be found not only for the question someone asks, but for every sub-question the AI derives from it.
This guide is for marketing, SEO and content teams. It separates what Google has officially confirmed, what independent measurements show and what is only assumed. And it shows how to respond without running into Google's spam policies.
Key takeaways
- Google officially confirms query fan-out for AI Mode and AI Overviews. According to Google, Deep Search can issue "hundreds of searches".
- Google publishes no fixed number of sub-queries. Measurements typically land at around 5 to 11 sub-queries per prompt.
- Sub-queries are long: in an analysis of roughly 366,000 fan-out queries, 71.1% had seven words or more.
- In an AirOps study, 95% of fan-out queries had no measurable search volume, so classic keyword tools barely see them.
- Google explicitly warns that building pages mainly for fan-out queries to manipulate AI answers violates its scaled content abuse spam policy.
What is query fan-out?
Query fan-out is a technique used by AI search systems that expands one user question into several related search queries. The system runs these queries in parallel, collects the results and summarizes them into an answer with sources. That way, one answer covers more angles than a single search could.
Google defines the term in its guide to generative AI features as "a set of concurrent, related queries generated by the model to request more information and fetch additional relevant search results to address the user's query" (Google Search Central, updated July 10, 2026).
The difference from classic search is fundamental. Until now it was one query, one results page. With fan-out, the AI decides which sub-questions to ask. Your page can end up in an answer even if it doesn't rank for the original question at all.
What Google has officially confirmed
Many articles on query fan-out mix facts, measurements and guesses. So first, only what Google itself says:
| Statement | Source | Date |
|---|---|---|
| AI Mode uses the query fan-out technique, breaks questions into subtopics and issues "a multitude of queries simultaneously". | Google Blog, I/O 2025 | May 20, 2025 |
| Deep Search uses the same technique "taken to the next level" and "can issue hundreds of searches". | Google Blog, I/O 2025 | May 20, 2025 |
| AI Overviews and AI Mode may issue multiple related searches across subtopics and data sources. | Google: AI features and your website | Dec 10, 2025 |
| To be shown as a supporting link, a page must be indexed and eligible to appear with a snippet. There are no additional requirements or special optimizations. | Google: AI features and your website | Dec 10, 2025 |
| Creating pages primarily for fan-out queries to manipulate rankings or generative AI responses violates the scaled content abuse spam policy. | Google: generative AI guide | Jul 10, 2026 |
| There's no requirement to break content into tiny pieces. One page can cover multiple topics. | Google: generative AI guide | Jul 10, 2026 |
| Google Search ignores llms.txt and similar files. | Google: generative AI guide | Jul 10, 2026 |
| Use the Generative AI performance report in Search Console to measure performance in generative AI features. | Google: generative AI guide | Jul 10, 2026 |
What Google does not publish: how many sub-queries a question triggers, how they are phrased, and how much weight a page gets when it is found only through a sub-query.
What the measurements show
Because Google gives no numbers, every count comes from outside measurements. Most capture sub-queries through APIs (for example, the grounding data of the Gemini API), not directly from the AI Mode interface. Treat the numbers as approximations.
Number of sub-queries. According to Similarweb, an analysis by Nectiv puts it at around 5 to 11 sub-queries per prompt, sometimes far more (Similarweb, December 8, 2025).
Length of sub-queries. Dan Petrovic (DEJAN) analyzed roughly 365,920 fan-out queries from the grounding data of Google, OpenAI and Amazon. 71.1% had seven words or more, and only 3.9% had one to three. On average, Google queries were 52 characters long and OpenAI queries 60. Proper nouns made up 18.9% of OpenAI queries, more than double Google's 8.6% (DEJAN, March 20, 2026).
Search volume and citations. AirOps analyzed 548,534 retrieved pages across 15,000 prompts. 95% of the fan-out queries had zero monthly search volume by traditional metrics. 32.9% of cited pages appeared only in the results for a fan-out query, not for the original prompt (Growth Memo, March 31, 2026).
Three conclusions follow:
- Sub-queries are more specific than the original question. They include features, audiences, years and brand names.
- Classic keyword research captures them poorly, because almost none of them has measurable search volume.
- A third of citations can come through paths you never see when you only look at the original question.
Fan-out in Google, Gemini and ChatGPT
Not every AI search works the same way. Here's what is publicly known:
| Platform | What's known | Can you see the sub-queries? |
|---|---|---|
| Google AI Mode | Fan-out officially confirmed; Deep Search with hundreds of searches | Not directly. The Search Console Generative AI performance report shows the impact. |
| Google AI Overviews | According to Google, may also issue multiple related searches | Not directly. Clicks and impressions count toward the "Web" search type in Search Console. |
| Gemini API (grounded with Google Search) | The response contains a webSearchQueries field, documented as "an array of the search queries sent to Google Search" (Firebase docs) |
Yes, for developers via the API |
| ChatGPT search | According to OpenAI, it uses third-party search providers and content from partners (OpenAI) | Not officially documented. Specialized tools record the sub-searches per answer. |
What this means in practice: ChatGPT's and Google's sub-queries differ. According to the DEJAN analysis, OpenAI queries are longer on average and contain proper nouns and numbers more often. If you only watch Google, you only see part of the picture.
Example: one question, eight sub-searches
An example makes this concrete. Say someone asks AI Mode:
"Which CRM is best for a plumbing company with 15 employees?"
The sub-searches below are constructed for illustration. They follow the patterns from the measurements above but are not a real recording.
| Type of sub-search | Possible sub-query | Content that fits |
|---|---|---|
| Category | crm software for plumbing companies | Overview page or guide |
| Comparison | crm for trades businesses comparison 2026 | Comparison table with criteria |
| Price | crm for contractors cost per user per month | Transparent pricing page |
| Alternatives | [known vendor] alternative for small businesses | Alternatives or vs page |
| Fit | crm with quoting and job scheduling for plumbers | Feature page with concrete use cases |
| Freshness | best crm for plumbers 2026 | Clearly dated content |
| Proof | [vendor] reviews plumbing contractors | Reviews on third-party sites, case studies |
| Concerns | crm data security hosting location small business | Page on security and hosting |
Hardly any of these queries would show volume in a keyword tool. Yet each one helps decide which vendors make it into the answer.
What this means for your content
Google's statements and the measurements point to six principles. None of them is a trick, and all of them help your human readers too.
1. One strong page instead of many thin ones. Google says explicitly that content doesn't need to be split into tiny pieces and one page can cover several topics. Answer the main question and the obvious follow-ups on the same page: price, fit, comparison, concerns.
2. Lead with the answer. Each section should open with the result, followed by the reasoning. Fan-out queries are specific, and a page that answers a specific question directly is easier for the system to use.
3. Concrete details instead of marketing speak. Sub-queries contain audiences, features, numbers and years. Write "for teams of 10 or more, $29 per user, data hosted in the EU" rather than "for every company size, fair prices, secure".
4. First-hand knowledge. Google names original content based on what you actually know as the factor most likely to shape your presence in generative AI search over the long run. Your own data, tests and real examples are something no other site can supply.
5. Presence beyond your own website. In the AirOps study, a third of cited pages entered the selection only through fan-out queries. Those include review sites, trade articles and directories. Check which third-party sites list your competitors but not you.
6. Technical basics. Your page must be indexed and allow a snippet. Don't block the search engine or AI crawlers you want to reach in robots.txt. According to Google, an llms.txt file doesn't help in Google Search.
The mistake Google warns about
The obvious reaction to query fan-out is a separate page for every possible sub-query. That is exactly what Google warns against. Creating pages primarily for fan-out queries to manipulate rankings or AI responses violates Google's scaled content abuse spam policy (Google Search Central).
The difference lies in intent. A pricing page that explains real prices is helpful. Fifty near-identical pages along the lines of "CRM for plumbers", "CRM for roofers", "CRM for electricians" with no substance of their own are not. For every new page, ask yourself: would I write this even without AI search, because customers need exactly this information?
Fan-out analysis in 5 steps (with template)
Here's how to build fan-out thinking into your content planning. Time required: about half a day per topic area.
Step 1: Collect core questions. Write down 10 to 20 questions your customers would ask an AI, as full sentences. Sources: sales, support tickets, Search Console, prompt research.
Step 2: Identify sub-queries. For each core question, derive the sub-searches, either with a tool that records sub-queries, through the Gemini API's webSearchQueries, or manually using the eight types from the example above.
Step 3: Check coverage. Is there a place on your website that answers the sub-query directly and concretely? If not, does the answer belong on an existing page, or do you really need a new one?
Step 4: Check sources. Which third-party pages does the AI cite for these sub-queries? Are you listed there? If not, that's often a faster lever than new content.
Step 5: Measure and repeat. Track mentions in AI answers and the Search Console report over several weeks. AI answers fluctuate from day to day, so single measurements tell you little.
Template to copy:
| Core question | Sub-query | Type | Answered on (URL) | Concrete answer? (yes/no) | Third-party sources cited | Are we listed? | Action |
|---|---|---|---|---|---|---|---|
| Which CRM is best for a plumbing company with 15 employees? | crm for contractors cost per user per month | Price | /pricing | no | Comparison site X | no | Add per-user prices to /pricing, check listing on X |
How to measure fan-out
There are three ways to make query fan-out visible, each with limits:
- Google Search Console: The Generative AI performance report shows how your content performs in generative AI features. It doesn't show individual sub-queries.
- Gemini API: The
webSearchQueriesfield contains the searches the model sent to Google. It's an approximation, because the API and AI Mode are not the same product. - AI visibility tools: Some tools record which sub-searches a model runs for a question. In growResonance, for example, you see the sub-queries for each ChatGPT answer in the prompt view, together with the cited sources and the competitors mentioned.
Disclosure: This article is published on the website of growResonance, an AI visibility tool. All figures in this article come from the linked external sources, not from our own data.
What nobody knows for sure
Being credible means naming the gaps too:
- The exact number of sub-queries in AI Mode is unknown. Measurements mostly rely on APIs, not on the interface users see.
- The weight given to a page found only through a sub-query is unclear.
- Whether targeted fan-out optimization has a measurable effect isn't cleanly proven. Published tests so far are small and short.
- The differences between platforms change with every model update. What holds for ChatGPT today may be different in three months.
Our recommendation: use fan-out as a mental model to plan better, more complete content. Don't build a strategy on individual numbers.
FAQ
What is query fan-out?
Query fan-out is a technique used by AI search engines such as Google AI Mode. The system splits a question into subtopics, sends several related searches at the same time and combines the results into one answer. That means your page can be cited even if it doesn't rank for the original question.
How many sub-queries does Google AI Mode generate?
Google doesn't publish a fixed number. Outside analyses typically find around 5 to 11 sub-queries per prompt, sometimes far more. For Deep Search, Google itself confirms that hundreds of searches are possible. The count varies with the question and its context.
Do fan-out queries have search volume?
Mostly not. In an AirOps study covering 15,000 prompts, 95% of fan-out queries had no measurable monthly search volume. That's why classic keyword tools barely show them. It makes more sense to identify sub-queries through AI visibility tools, the Gemini API or systematic question research.
Should I create a separate page for every sub-query?
No. Google explicitly warns that pages created primarily for fan-out queries can violate its scaled content abuse spam policy. A better approach is one strong page that concretely answers the main question and the obvious follow-ups. A new page is only worth it if customers would need it without AI search too.
Can I see the fan-out queries for my website?
Not directly in Google. Search Console's Generative AI performance report shows how you perform in AI features, but not individual sub-queries. Developers can use the Gemini API's webSearchQueries field to see which searches the model runs. Specialized tools record sub-queries per AI answer, for example for ChatGPT.
Does query fan-out work the same way in ChatGPT and Google?
Not quite. According to OpenAI, ChatGPT uses third-party search providers. An analysis of roughly 366,000 fan-out queries shows that OpenAI's queries are longer on average than Google's and contain proper nouns and numbers more often. If you want visibility in both, track them separately.
Conclusion
Query fan-out changes how you think about visibility. It's no longer just the one question that counts, but the bundle of sub-searches an AI turns it into. Google officially confirms the technique, measurements show long, specific sub-queries with no search volume, and a third of citations can come through exactly these paths. The right response isn't mass-producing new pages. It's complete, concrete content and a presence on the sources AI systems draw from.
Next step: Find out whether ChatGPT, Gemini and Google AI Mode mention your brand for your customers' most important questions. Run the free visibility check
Read more: How we measure AI visibility · growResonance vs. Finseo · Track AI visibility
Sources
- Google Blog: AI Mode in Google Search, updates from Google I/O 2025, May 20, 2025. https://blog.google/products-and-platforms/products/search/google-search-ai-mode-update/ (accessed October 30, 2026)
- Google Search Central: AI features and your website, updated December 10, 2025. https://developers.google.com/search/docs/appearance/ai-features (accessed October 30, 2026)
- Google Search Central: Guide to optimizing for generative AI features on Google Search, updated July 10, 2026. https://developers.google.com/search/docs/fundamentals/ai-optimization-guide (accessed October 30, 2026)
- Firebase / Google: Grounding with Google Search (
webSearchQueriesfield). https://firebase.google.com/docs/ai-logic/grounding-google-search (accessed October 30, 2026) - OpenAI: Introducing ChatGPT search, update of February 5, 2025. https://openai.com/index/introducing-chatgpt-search/ (accessed October 30, 2026)
- Similarweb: Query fan-out, December 8, 2025 (citing Nectiv). https://aisearch.similarweb.com/blog/query-fan-out/ (accessed October 30, 2026)
- Dan Petrovic, DEJAN: Fanout Query Analysis, March 20, 2026. https://dejan.ai/blog/fanout-query-analysis/ (accessed October 30, 2026)
- Growth Memo: Growth Intelligence Brief #16 (AirOps study), March 31, 2026. https://www.growth-memo.com/p/growth-intelligence-brief-16 (accessed October 30, 2026)
