What we measured ·
Study data: 331 ChatGPT fan-out searches
Method: Fan-out queries returned by DataForSEO's ChatGPT LLM responses API (model gpt-5.4-mini, web search on, US) for 25 SEO tool prompts run twice each on 22 October 2026 (UTC). We counted operators, terms, and site: targets across all 331 queries.
ChatGPT ran between 3 and 10 background searches per answer, with a median of 7, averaging 7.6 words each, far longer than typical human searches.
145 of 331 searches, 44 percent, used a site: operator to search a specific domain, led by ahrefs.com in 11, semrush.com in 10, and developers.google.com in 9.
258 searches, 78 percent, contained the word 'official', and 59 contained the current year, which shows a strong preference for primary, current vendor sources.
| Domain | site: searches |
|---|---|
| ahrefs.com | 11 |
| semrush.com | 10 |
| developers.google.com | 9 |
| screamingfrog.co.uk | 7 |
| surferseo.com | 7 |
| clearscope.io | 6 |
| moz.com | 5 |
What fan-out is
Instead of running the user's question as one search, AI search systems generate several related searches, retrieve results for each, and write an answer from the combined sources. Google describes the same technique for AI Mode. Fan-out explains why a page can be cited for a question it does not directly target: it may answer one of the sub-searches.
ChatGPT looks for primary sources
The dominant pattern was verification. After identifying candidate tools, ChatGPT searched each vendor's own domain with site: plus words like 'official', 'pricing', or 'features'. 28 searches included pricing terms and 14 included 'review'. None included 'reddit'. For software companies, this means your own product, pricing, and documentation pages are directly searched when you are a candidate, so they must state facts clearly and currently.
An example
For 'What are the best programmatic SEO tools for a B2B SaaS company?', one run produced seven searches: three broad searches combining programmatic SEO with terms like 'official documentation' and 'internal linking', then site: searches on zapier.com, airtable.com, make.com, and notion.so to check automation and database features. Tools that did not surface in the first, broad searches were not checked in that run.
Variation between runs
The exact searches almost never repeated: across the 25 prompts, the two runs shared about 1 percent of identical fan-out queries. The themes were similar, but wording and the vendors checked changed. Combined with the 50 percent overlap in brands named, that variability is why AI visibility must be measured over repeated runs.
What to do with this
Make sure your product and pricing pages answer the questions a verification search would ask: what it does, who it is for, current prices, integrations, and limits, in crawlable HTML. Be present in the broad sources the first searches surface, such as comparison articles and review sites. Our query fan-out explorer shows the kinds of sub-questions a broad query tends to produce.
Limitations
One model through an API, 50 answers, one market and date. The consumer ChatGPT app may behave differently. Fan-out queries are as reported by the API. SerionFlow does not monitor fan-out at scale; this was a one-off study.
Clear answers
Frequently asked questions
What is query fan-out in ChatGPT?
+
The set of background searches ChatGPT runs to gather sources before writing an answer. In our study it ran a median of 7 per answer.
Does ChatGPT read vendor websites directly?
+
Often. 44 percent of the background searches in our study used site: to search a specific vendor's domain.
About the author
Shadrack Kumi · Founder, SerionFlow
Shadrack builds SerionFlow's site research, page generation, and Search Console tooling, and runs the free tools these guides pair with. Search Console figures cited on these pages come from serionflow.com's own property.
Continue exploring
Related SerionFlow resources
More in Research
- What ChatGPT recommends and cites when asked about SEO tools
- SEO search demand in 2026: which topics grew and which shrank
- How sites that rank for SEO queries treat AI crawlers in robots.txt
- llms.txt adoption among sites that rank for SEO queries
- People Also Ask in SEO search results: what 541 questions show
- SERP features and ranking page types by search intent
- Homepage schema markup adoption among SEO-industry sites
- What ranks for '[tool] alternatives' searches in the SEO market
Make the next move obvious
Let SerionFlow turn your market evidence into momentum.
Confirm the market, rank the opportunities that matter, create brand matched pages on your domain, and keep moving with controlled weekly intelligence.