When someone types one question into Google AI Mode or ChatGPT, the AI often does not search for that exact question. It splits the question into several smaller searches, runs them in the background, and then builds one answer from everything it found. This is called query fan-out.
Fan-out explains something that confuses a lot of marketers. Your page can be cited in an AI answer for a question you never wrote about. And a page that ranks first for the main keyword can be left out.
What query fan-out means, in plain words
A "query" is just a search. "Fan-out" describes one thing spreading out into many, like the ribs of a hand fan. So the AI takes one question and turns it into many related searches. Each search covers one part of the question. The AI reads the results of all of them, then writes a single answer.
What Google and OpenAI have said about it
Google AI Mode and AI Overviews
AI Mode is a separate tab in Google Search where you can have a longer, chat-style conversation instead of getting a list of links. At Google I/O on May 20, 2025, Google's Head of Search, Elizabeth Reid, wrote that "AI Mode uses our query fan-out technique, breaking down your question into subtopics" and "issuing a multitude of queries simultaneously on your behalf."1 She added that this "enables Search to dive deeper into the web than a traditional search on Google."1
The same post said Deep Search, a more thorough mode, "can issue hundreds of searches" to put together a fully cited report.1 Google had already described the technique in March 2025, when it first announced AI Mode. It said AI Mode runs several related searches at once, each covering a different subtopic and data source, then combines the results.2
Fan-out also applies to AI Overviews, the AI-written summaries at the top of normal Google results. Google's Search Central page for site owners says: "Both AI Overviews and AI Mode may use a 'query fan-out' technique." While the answer is being written, the systems find more supporting pages. That lets Google show a wider set of links than a classic search would.3
The same page sets the rules for being included. A page must be indexed and eligible to show in Google Search with a snippet. Google also says: "There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary."3 In other words, there is no special fan-out setting. What matters is normal Search eligibility plus being a good result for one of the background searches.
ChatGPT search
OpenAI describes a similar process. Its help page on searching the web with ChatGPT says ChatGPT search "typically rewrites your query into one or more targeted queries" and sends them to its search partners. It may also send additional, more specific queries to those providers.4
The same page says that to appear in ChatGPT search results, a site must allow OpenAI's search crawler, called OAI-SearchBot, to crawl it. Even then, OpenAI says placement "is not guaranteed."4
How fan-out works, step by step
Neither company publishes its exact system. Treat this as a simple model based on their descriptions and the research below.
- The AI reads the question and works out which details matter (budget, team size, location).
- It writes its own searches. These often add words the person never typed, like the current year, "review," "price," or "vs."
- It runs those searches. Google says these can cover different data sources too, such as its shopping data or Knowledge Graph (Google's database of facts about people, places and things).2
- It pools the results. Pages from all the searches go into one shared pile.
- It writes the answer and picks citations. A "citation" is a link the AI shows as the source for part of its answer.
Step 2 matters most for marketers. The AI's searches are not the question the user typed. That gap is why keyword rankings and AI citations often don't line up.
What the data shows about fan-out
How often ChatGPT actually fans out
A large study from Ahrefs, published on October 12, 2026, looked at 66.6 million free, logged-out ChatGPT responses (from the GPT-5.6 Luna model) and their 15 million fan-out queries from August and September 2026.5 Logging out removes chat history and memory, which could change results.
| Finding | Number |
|---|---|
| Responses that needed a web search at all | 14.4%5 |
| Average fan-out searches when ChatGPT did search | About 1.65 |
| Fan-out searches using a "site:" operator (searching inside one website) | 6.8%5 |
| Most common "site:" searches that pointed at a brand's own site | 75.9%5 |
| English fan-out searches containing "2026" (September) | 22.9%5 |
| English fan-out searches containing "official" (September, up from 10.1% in August) | 13.1%5 |
Two lessons stand out.
First, the number of background searches depends heavily on the setup. Ahrefs notes that its broad mix of prompts differs from the commercial "best X" prompts that other studies used, and that free, logged-out sessions search less than paid or reasoning models.5 Semrush, by contrast, estimates that a single complex AI search can trigger 8 to 20 or more background searches, while noting that Google has not published exact counts.6 Both can be true. A simple fact question may need no search, while a detailed buying question may trigger many.
Second, the extra words the AI adds tell you what kind of page it wants. Ahrefs puts it this way: "The terms ChatGPT adds to its fan-out queries hint at the page types it wants to retrieve."5 In a sample of 1,474 ChatGPT responses about running shoes, "best" appeared 206 times, "2026" 160 times, "review" 117 times, "vs" 105 times and "official" 65 times.5 So the AI is looking for current reviews, comparisons and official brand pages.
Why pages get cited without ranking for the main keyword
Several studies have found that many cited pages do not rank well for the question the user actually asked. Fan-out is the most common explanation.
An Ahrefs study from August 2025 ran 15,000 long-tail prompts through AI assistants and compared the cited links with Google and Bing results for the same prompts. It found that "only 12% of links cited by ChatGPT, Gemini, and Copilot appear in Google's top 10 results."7 The authors suggested the assistants may never search the exact prompt the user typed, which would explain the gap. They also pointed to a merging method called Reciprocal Rank Fusion. In simple terms, it combines the result lists from several searches and gives extra credit to pages that appear consistently across them.7
For Google AI Overviews, Ahrefs updated its own earlier research in March 2026, using about 4 million AI Overview URLs across 863,000 keyword results pages.8
| Where the cited page ranked for the original search | Share of AI Overview citations |
|---|---|
| Top 10 results | 37.9%8 |
| Positions 11 to 100 | 31.2%8 |
| Not in the top 100 at all | 31.0%8 |
That top-10 share had been about 76% in Ahrefs' July 2025 study.8 Ahrefs offers fan-out as the most likely reason for the drop: pages that show up often in the results for the background searches get cited, even if they rank poorly for the main one.8 The authors hedge this with words like "may" and "suggests," because Google does not show its fan-out queries.8
A page that answers one narrow part of a question well can win a citation for the whole question, even if it never ranked for the question itself.
That is good news for smaller sites with specific, useful pages. It is bad news for anyone who expects a top ranking to carry over into AI answers.
Worked example: one buyer prompt, fanned out
Take this buyer prompt:
"Best project management software for a 20-person agency"
Important: the sub-queries below are illustrative. They are not real logs from Google or ChatGPT. They follow the patterns the studies above describe (adding the year, "vs," "price," "review" and "official"). Real fan-out sets change from run to run.
The last column describes an illustrative composite: a made-up mid-sized software company we'll call Taskfold. Like many companies its size, it has feature pages, a pricing page, a general blog and a help center.
| # | Illustrative sub-query | What the AI wants to learn | Content type that answers it | Does Taskfold have it? |
|---|---|---|---|---|
| 1 | best project management software for agencies 2026 | The current shortlist | "Best for agencies" roundup or listicle (often on third-party sites) | No (not on any roundup it knows of) |
| 2 | project management tools for small teams 10 to 50 users | Which tools fit this team size | Use-case page by team size | Partly (a generic "small teams" page) |
| 3 | agency project management client portal | Can clients see progress? | Feature page for client access | Yes |
| 4 | project management software with time tracking and billing | Can agencies bill hours? | Feature page plus integrations list | Partly (time tracking yes, billing only via an add-on with no page) |
| 5 | Taskfold pricing per user | Cost for 20 seats | Plain-text pricing page | Partly (prices sit in an image, not text) |
| 6 | Taskfold vs [main rival] | Head-to-head differences | Honest comparison page | No |
| 7 | Taskfold review agency | What real users say | Third-party review sites, case studies | Partly (one case study, not agency-specific) |
| 8 | resource management for creative agencies | Who is free to take on work? | Feature page or guide on capacity planning | No |
| 9 | project management software integrations Slack Google Drive | Does it fit existing tools? | Integrations directory with one page per integration | Yes |
| 10 | how long to set up project management software for a team of 20 | Effort to switch | Onboarding guide or migration page | No |
| 11 | site:taskfold.com agency templates | Official proof of agency features | Templates gallery on the brand's own site | Yes |
| 12 | Taskfold official security SOC 2 | Is client data safe? | Security or trust page | Yes |
Only a few of the twelve sub-queries look like the keyword a marketer would normally target. The rest are about pricing, setup time, billing, security and comparisons. Those are the questions a buyer would ask a salesperson. The AI asks them in the background instead.
Some sub-queries (rows 1 and 7) are mostly answered by other websites, like review sites and roundups. For those, the job is to be included on the pages that answer them.
The fan-out coverage score: a simple method
You can turn that table into one number, which we call the fan-out coverage score. It shows what share of the likely background searches your content can answer. It is a planning tool, not a ranking signal.
How to calculate it
- Pick one important buyer prompt, ideally one you already track.
- Write out 10 to 15 likely sub-queries. Use the patterns from the research: add the year, "best," "vs," "price," "review," "official," and the specific needs in the prompt (team size, industry, budget). You can also ask an AI assistant which searches it would run. Treat its list as a sample, not the truth.
- For each sub-query, find the page that answers it. This can be on your site or on a third-party site that mentions you.
- Score each sub-query.
- Fully covered (a clear, crawlable page answers it in plain text) = 1 point
- Partly covered (the answer exists but is thin, hidden in an image, or buried on an unrelated page) = 0.5 points
- Not covered = 0 points
- Add up the points and divide by the number of sub-queries.
Taskfold's score
From the illustrative table above, Taskfold has:
- 4 fully covered (rows 3, 9, 11 and 12) = 4 points
- 4 partly covered (rows 2, 4, 5 and 7) = 2 points
- 4 not covered (rows 1, 6, 8 and 10) = 0 points
So the calculation is:
(4 × 1 + 4 × 0.5 + 4 × 0) ÷ 12 = 6 ÷ 12 = 50%
The cheapest win is usually fixing the "partly" rows. Putting prices in plain text instead of an image takes an afternoon. That alone would lift the score to about 54% (6.5 ÷ 12). Fixing all four partial rows would take it to 67% (8 ÷ 12). New pages, like a comparison page or setup guide, take longer.
How to read the score
| Score | What it usually means | Next step |
|---|---|---|
| Below 40% | The AI has few ways to find you for this prompt | Start with plain-text facts: pricing, features, security, setup |
| 40% to 70% | Core pages exist, but comparisons and use cases are thin | Fix partial rows, then add comparison and use-case pages |
| Above 70% | Your own site covers most background searches | Focus on third-party roundups and reviews you are missing from |
These ranges are rules of thumb, not tested cut-off points. A high score does not guarantee a citation. It gives the AI more chances to find you.
How to plan content around fan-out
Google says no special optimization is needed.3 What changes is which questions you decide to answer.
- Put key facts in plain text. Prices, plan limits, features and security details should be readable text on crawlable pages. Ahrefs found that 75.9% of the most common "site:" searches in ChatGPT pointed at a brand's own site.5 When the AI checks your site directly, it needs to find the answer there.
- Write the pages buyers' follow-up questions need. Comparison pages, setup guides, use-case pages by team size or industry, and pricing explainers all match common fan-out patterns.
- Keep pages current. The current year was the top term in ChatGPT's English fan-out searches in September 2026, at 22.9%.5 Clear dates and up-to-date facts help.
- Make each section answer one sub-question clearly. A long page can serve several sub-queries if each section has a clear heading and a direct answer.
- Earn a place on third-party pages. Some sub-queries, like "best X for Y 2026," are mostly answered by roundups and review sites. You need to be on those pages.
Be realistic about results. In a test published in September 2025, Semrush tried this on four of its own blog articles. It researched 10 to 20 likely fan-out queries per article and updated the content to cover them. After one month, the articles went "from being cited two times to five times."9 But share of voice (the share of AI answers in a topic that mention your brand, compared with others) fell from 23.4% to 20.0%, and brand mentions fell from 18 to 10.9 The test was small (four articles) and ran for one month, and the authors called the citation gain "directionally indicative of success."9 Covering more sub-queries seemed to help citations, but it is not a guaranteed win.
How to adapt your prompt tracking for fan-out
Prompt tracking means running a fixed set of questions through AI platforms on a regular schedule and recording whether your brand is mentioned or cited. Fan-out changes how you should read the results.
- Track the whole prompt, not only the keyword. Keep details like "for a 20-person agency" in your tracked prompts. They shape the sub-queries.
- Record which of your pages get cited, not only whether you were cited. If your pricing page or security page gets cited for a "best software" prompt, that tells you which sub-query you won.
- Map each cited page back to a row in your coverage table. Over time this shows which content types earn citations.
- Note which competitor pages get cited, and what type they are. If a rival's comparison page keeps showing up, that points to a sub-query you don't cover.
- Run each prompt more than once. One check is a snapshot. Look at how often you are cited across several runs.
- Group results by intent, not exact wording. Semrush advises tracking trends in sub-query intent rather than exact wording, because the wording keeps changing.6
- Re-score your coverage every quarter. After adding or fixing pages, check whether citations follow.
- Re-check after big model updates. Ahrefs stresses that its numbers are a snapshot of one model in August and September 2026.5 A new model can change fan-out patterns quickly.
The honest limits of all this
Fan-out is confirmed by the platforms. But a lot of it is hidden from you.
- You usually can't see the real sub-queries. Google does not show its fan-out queries to site owners.8 AI Mode and AI Overviews traffic is mixed into the normal "Web" numbers in Search Console, so you can't separate it out there either.3 Any list you build is an educated guess.
- The sub-queries change every time. Semrush notes that "the same prompt can produce different fan-out queries each time you run it."6 It also says fan-out queries "shift from one run to the next, and most carry little or no search volume of their own."6 So don't chase exact phrasings or judge sub-queries by search volume.
- The number of searches varies a lot. Free, logged-out ChatGPT searched for only 14.4% of prompts.5 Google's Deep Search can run hundreds of searches.1 Results differ across platforms and plans.
- The evidence linking fan-out to citations is mostly correlation. Ahrefs presents fan-out as the likely explanation for citations outside the top 10, but frames it as a possibility, not a proven cause.8 Semrush's optimization test was small and its results were mixed.9
- Coverage is not the only factor. Being crawlable, indexed and eligible for a snippet still comes first for Google.3 For ChatGPT, OAI-SearchBot must be allowed to crawl your site.4 Trust, freshness and what other sites say about you also matter.
The coverage score works within these limits. It doesn't guess the exact searches. It checks whether your content answers the kinds of questions the AI is likely to ask.
See which sub-queries you are winning and losing
Your coverage table is a guess until you see which of your pages AI platforms actually cite, and for which prompts. Ripplix tracks your buyer prompts across AI platforms and shows which pages get cited and which competitor pages win instead.
Get your free AI Visibility Report →
Sources:
- Google: AI in Search: Going beyond information to intelligence, Elizabeth Reid (May 20, 2025)
- Google: Expanding AI Overviews and introducing AI Mode, Robby Stein (March 5, 2025)
- Google Search Central: AI features and your website (last updated December 10, 2025)
- OpenAI Help Center: Searching the web with ChatGPT (accessed October 19, 2026)
- Ahrefs: How ChatGPT Does Query Fan-Out (15 Million Queries Studied), Louise Linehan (October 12, 2026)
- Semrush: What is query fan-out? How to find & optimize for subqueries, Zach Paruch (August 17, 2026)
- Ahrefs: Only 12% of AI Cited URLs Rank in Google's Top 10 for the Original Prompt, Louise Linehan (August 11, 2025)
- Ahrefs: Update: 38% of AI Overview Citations Pull From The Top 10, Louise Linehan (March 2, 2026)
- Semrush: We Tested Query Fan-Out Optimization (Here's What We Learned), Zach Paruch (September 26, 2025)
