AI Search & GEO

How Query Fan-out Works in ChatGPT?

By BacklinkPrice Team · Nov 19, 2026 · Updated Nov 20, 2026 · 12 min read
Illustrated cover for How Query Fan-out Works in ChatGPT?

Query fan-out in ChatGPT rewrites your prompt into short follow-up searches before it answers. This guide explains each step, the numbers behind it, and how to match the searches ChatGPT runs.

Query fan-out in ChatGPT means ChatGPT rewrites your prompt into one or more short follow-up searches before it answers. It runs those searches, reads the pages they return, and then writes one answer with citations to the pages it used.

Key takeaways

  • ChatGPT rewrites a prompt into follow-up searches, then cites pages that match those searches, not only the first prompt.
  • In an Ahrefs study of 66.6 million free ChatGPT responses from August and September 2026, only 14.4% contained fan-out queries, at about 1.6 each.
  • Fan-out queries often ask for official facts, prices, and reviews, and 6.8% used a site search for one website.
  • Pages at the top of the search results ChatGPT reads are cited far more often than pages at position 10.
  • You cannot see ChatGPT's fan-out log, so you measure coverage by testing prompts and tracking citations over time.

What query fan-out means in ChatGPT

Query fan-out is the step where ChatGPT turns one prompt into several short searches. OpenAI says ChatGPT search typically rewrites your query into one or more targeted queries that it sends to search providers. After it reviews the first results, it may send more specific queries to other providers.

That rewrite is the fan-out. The user sees one question and one answer.

Behind that answer, ChatGPT may have run a search for the price, a search for the official product page, and a search for reviews. Each of those searches is a fan-out query.

The term itself comes from Google. Google Search Central says AI Overviews and AI Mode may use query fan-out, which means issuing several related searches across subtopics and data sources.

ChatGPT does a similar rewrite, even though OpenAI does not use the name fan-out in its help page. For the general term, see what query fan-out is.

Why fan-out matters for your pages

Fan-out matters because a page can win a citation for a follow-up search, even if it does not match the first prompt word for word. ChatGPT cites pages that answer the searches it actually runs. A page that covers the price, the specs, and the review angle has more chances to match one of those searches.

It also matters because most ChatGPT answers do not search the web at all. In the Ahrefs study of 66.6 million free, logged-out responses from August and September 2026, only 14.4% contained fan-out queries. The other 85.6% were answered from the model's training data, with no live page cited.

That split sets the job. Your page cannot be cited when ChatGPT does not search.

When it does search, it usually runs only about 1.6 fan-out queries per response, so your page must match one of those one or two searches. For more on how those choices are made, see how ChatGPT chooses which sources to cite.

How fan-out starts in ChatGPT

Fan-out starts with a decision and a rewrite. OpenAI describes the first part, and the studies below fill in the rest. The two steps below cover the start of the process.

ChatGPT first decides if the question needs current facts. OpenAI says it may search the web automatically when your question would benefit from current information.

A user can also start a search by hand in the tools menu. If the question is general, ChatGPT may answer from training data and cite nothing.

Step 2: ChatGPT rewrites the prompt

ChatGPT turns the prompt into one or more targeted searches. OpenAI gives an example where a question about a cancer drug becomes a search for the drug and the year. Then a second search covers a conference on the same topic.

Memory and location can shape the rewrite too. OpenAI says saved memory may change the search, and an IP address may add a city name.

How fan-out ends in ChatGPT

Fan-out ends with retrieval, reading, and one cited answer. The three steps below cover the end of the process. Each step decides which pages reach the final answer.

Step 3: ChatGPT sends the searches to providers

ChatGPT sends the rewritten queries to search providers. OpenAI says it sometimes partners with other search providers, and it names Microsoft in its privacy notes for those queries.

It does not share your IP address or account details with those providers. The results come back as titles, URLs, and short text for each page.

Step 4: ChatGPT reads the best pages

ChatGPT reads the pages that look most useful for each search. A page must be crawlable by OAI-SearchBot to be eligible for ChatGPT search, as OpenAI states on its help page.

That crawler is separate from GPTBot, which collects training data. Blocking the training crawler does not block search. For the crawler itself, see what OAI-SearchBot is.

Step 5: ChatGPT writes one answer with citations

ChatGPT joins the results into one answer and cites the pages it drew on. OpenAI warns that search results and citations can be incomplete, outdated, or incorrect.

Open a cited source and check that it supports the claim. Placement is not guaranteed.

What ChatGPT searches for first

ChatGPT's fan-out queries lean toward official facts, prices, and reviews. In the Ahrefs study of 15 million fan-out queries from August and September 2026, the word "official" appeared in 13.1% of English fan-out queries in September.

That was up from 10.1% in August. The word "2026" appeared in 22.9% of fan-out queries in September.

Price searches rose fast. The word "price" went from 6.0% of fan-out queries in August to 8.9% in September, a rise of 47% in one month.

Reviews rose by about a third in the same period, and "specifications" entered the top 50 words for the first time. Definition searches fell by about a fifth.

Many fan-out queries go straight to one site. In the same Ahrefs study, 6.8% of fan-out queries used a site search for a single website, across about 116,000 domains. Among the most searched sites, 75.9% of those site searches pointed at a brand site, not a retailer, forum, or review site.

Apple, IKEA, and Nike led that list, with 9,442, 7,361, and 5,210 site searches in the study. Retailers made up 11.9% of the most common site searches, while Reddit and Facebook made up 1.9% between them. The pattern shows that ChatGPT often goes to the brand itself for facts.

How many fan-out queries to expect

The count depends on the sample, the model, and the prompt type. The table below shows three published counts for three setups.

SetupSearches per promptSource and date
ChatGPT, free and logged outAbout 1.6 fan-out queriesAhrefs, 2026
Google AI Mode5 to 11 searchesAhrefs, 2026
ChatGPT Deep Research, one test420 searchesAhrefs, 2026

The Ahrefs study of 66.6 million free, logged-out responses found 14.4% contained fan-out queries, with about 1.6 each. That is a floor, not a ceiling. Ahrefs notes that free, logged-out sessions search less than paid or reasoning models, and it calls the figures a snapshot of one model over two months.

Other studies of harder prompts found more searches. Ahrefs cites research from Seer Interactive and Nectiv that found an average of 9 to 11 fan-out queries per prompt. In that research, 59% of prompts triggered 5 to 11 searches, and 24% triggered 12 to 19, up to 28.

Deep Research is the extreme case. In one test cited by Ahrefs, it ran 420 searches for a red phone case, with 200 searches spent on the phone model alone.

Tools and time needed to track fan-out

You need three tools to track fan-out, and none of them shows the full log. Ahrefs Brand Radar shows fan-out queries for ChatGPT and Perplexity prompts in its AI responses report. You can filter that report by words such as price, review, or site, then export the rows to a sheet.

Google Search Console shows how your pages perform in search, and its Performance report covers AI features in Google Search. That is a Google measure, not a ChatGPT measure, so use it as a check on the same pages. Do not read it as proof of ChatGPT citations.

Bing matters too, because ChatGPT search uses search providers that include Bing. A page that is not indexed in Bing has fewer chances to be retrieved. See how to get your site indexed in Bing.

A first audit takes 1 to 2 hours. Pick 20 to 30 prompts buyers ask, run each one in ChatGPT with search on, and record the citations.

Repeat them every 2 weeks, and allow 30 minutes per check. Brand Radar is a paid Ahrefs add-on, while Search Console and Bing Webmaster Tools are free, as of November 2026.

How to write pages that match fan-out queries

Write one focused page per question, with a short section for each fact ChatGPT is likely to check. AirOps tested this in its 2026 study of 16,851 queries and 353,799 pages. Pages that covered 26% to 50% of a query's sub-topics were cited more often than pages that tried to cover all of them.

Match the heading to the question. In the same AirOps study, pages with headings that closely matched the user's query were cited 41% of the time, against 29% for weak matches. Put the price, the date, and the key fact in plain text near the top of the section, so the model can quote it without guessing.

Use official facts where they exist. In an Ahrefs test of 20,000 US prompts, 84.4% of factual questions about pricing, product features, or company details cited at least one official source.

For those factual questions, 63.4% of citations went to official websites. For comparison questions, that share fell to 46.4%.

Put your own prices, specs, and dates on your own site in plain text. A fact that lives only in a sales deck or a video cannot be cited from a web search. For the wider method, see what generative engine optimization is.

What fresh, data-rich pages gain

Fresh pages with many facts do better in ChatGPT citations. SE Ranking found this in its 2025 study of 129,000 domains and 216,524 pages across 20 niches.

Pages updated in the past 3 months averaged 6 citations, against 3.6 for outdated pages. That is a clear gain from a quarterly refresh.

Depth helps too. In the same SE Ranking study, articles over 2,900 words averaged 5.1 citations, against 3.2 for articles under 800 words.

Pages with 19 or more data points averaged 5.4 citations, against 2.8 for pages with few facts. Add sourced numbers, not filler, to reach that level.

Authority still counts. SE Ranking found that sites with over 32,000 referring domains were 3.5 times more likely to be cited by ChatGPT than sites with up to 200 referring domains.

Citations jumped from 2.9 to 5.6 at that mark. Earn links and mentions over time, and keep the page itself fresh and fact-rich.

How to measure results

Measure citations, not clicks, because a citation is the unit ChatGPT shows. Count how many of your 20 to 30 test prompts cite your site, how many cite a rival, and how many cite no site at all. Record the date and the model, since the same prompt can give different answers on a new run.

Track the same prompts over 4 to 8 weeks. A page that moves from no citation to one citation on 3 of 30 prompts has made real progress.

A page that is retrieved but never cited needs a closer title match or fresher facts, not more words. For a full method, see AI citation tracking explained.

Watch your referral traffic in your analytics tool as a second check. ChatGPT referrals are small next to search traffic, and a cited page may get few clicks. That is normal when the answer itself gives the fact.

Judge the work by citation share first, then by visits from ChatGPT. For the analytics setup, see how to measure AI referral traffic in GA4.

Common mistakes

  • Chasing full coverage of every sub-topic. AirOps found that pages covering all sub-topics were cited less often than pages covering 26% to 50% of them, once relevance was held steady.
  • Hiding prices and specs in images or scripts. ChatGPT searches for price and official facts in text, so a fact it cannot read in plain text cannot be cited.
  • Blocking OAI-SearchBot by mistake. Check your robots file, since OpenAI says that crawler is what makes a page eligible for ChatGPT search.
  • Reading one run as proof. Fan-out queries change between runs, so test the same prompt at least 3 times before you change a page.
  • Counting a retrieval as a citation. ChatGPT reads far more pages than it cites, so being read is not the same as being shown. For the difference, see what an AI citation is.

Frequently Asked Questions

Does ChatGPT use query fan-out for every prompt?

No. In the Ahrefs study of 66.6 million free, logged-out responses from August and September 2026, only 14.4% contained fan-out queries.

Most prompts were answered from training data, with no web search and no citation. Paid and reasoning models search more often, so the rate varies by plan and prompt type.

Can I see the fan-out queries ChatGPT ran for my prompt?

No, ChatGPT does not show a fan-out log to users. You can infer the likely searches from the citations in the answer and from the words in the prompt. Ahrefs Brand Radar shows fan-out queries for tracked prompts, which is the closest public view of the searches behind an answer.

How many fan-out queries does ChatGPT run?

It depends on the prompt and the model. In the Ahrefs study of free, logged-out ChatGPT, searching responses averaged about 1.6 fan-out queries each.

Harder prompts and Deep Research run far more, with one test reaching 420 searches. Use the 1.6 figure for simple prompts, not for research tasks.

Does a page need to rank in Google to be cited by ChatGPT?

No, a Google rank is not required. ChatGPT pulls pages from search providers that include Bing and from its own crawl, so a page indexed in Bing and open to OAI-SearchBot can be cited. A strong Google rank still helps, since it signals trust and relevance to the same kind of content.

Why does ChatGPT cite a rival instead of my page?

ChatGPT cites the page that best matches the fan-out search it ran. That is often the official source for a price or a spec, or a page with a closer heading match.

Check which search your rival won, then add the missing fact or tighten your heading. One missing price can lose the citation.

Is query fan-out the same in ChatGPT and Google AI Mode?

No. Both split one question into several searches, but the counts and models differ.

Google says AI Overviews and AI Mode may use fan-out across subtopics and data sources. Ahrefs reports 5 to 11 searches for AI Mode and about 1.6 fan-out queries for free ChatGPT, so the two systems should be tracked separately.

Sources

BacklinkPrice Team
Written by
BacklinkPrice Team
Link building and backlink pricing research

The BacklinkPrice Team researches backlink prices, publisher sites and link building tools. We check prices and plans against each provider's own pages and date them, so you know how current the numbers are.

More from BacklinkPrice Team

Related guides