ChatGPT Query Fan-Out Statistics (2026)

What do ChatGPT query fan-out statistics show?

They show that ChatGPT searches less often, and more narrowly, than most SEOs assume. In Ahrefs' study of 66.6 million free, logged-out ChatGPT responses from August and September 2026, only 14.4% contained fan-out queries, the background searches ChatGPT writes to find pages for its answer. Those responses averaged about 1.6 fan-out queries each. The Fan-Out Phrasebook in Figure 1 shows what those queries say: in one month, "official" rose from 10.1% to 13.1% of fan-out queries and "price" from 6.0% to 8.9%, while "meaning" fell.

The supporting cards follow the queries to the sites they target and the sources ChatGPT ends up citing. Both point the same way: away from sites that summarize information and toward the brands, regulators and researchers that produce it.

How often does ChatGPT use query fan-out?

About 1 in 7 times: 14.4% of the 66.6 million responses Ahrefs analyzed contained fan-out queries, roughly 9.6 million responses holding 15.0 million queries. Ahrefs first described this as the share of responses that searched the web and corrected it in October 2026, so the figure counts fan-out queries, not every kind of retrieval. The data covers one model, GPT-5.6 Luna, in free, logged-out sessions, which strips out chat history and memory.

How many fan-out queries does ChatGPT run per prompt?

Published averages range from 1.6 to 7.61 fan-out queries, depending on who measures and how. Ahrefs counts about 1.6 per response that has any fan-outs. Peec AI measured 2.1 per prompt across 5 million fan-outs in April 2026, against 1.4 for Perplexity and 6.8 for Grok. Nectiv, rerunning about 4,000 fixed prompts through GPT-5.6 Sol via the API, found 7.61 per prompt in 2026, up from 2.17 in its 2025 test, with one prompt producing 29.

The queries also got longer. Peec found the average ChatGPT fan-out grew from about 6 words in October 2025 to about 12 by January 2026, and Nectiv's 2026 test averaged 6.82 words per query, up from 5.48.

What does ChatGPT search for in fan-out queries?

Mostly the current year, official information and prices. In September 2026, "2026" appeared in 22.9% of fan-out queries, "official" in 13.1% and "price" in 8.9%, all up from August. Measured more precisely, "price" went from 6.04% to 8.88%, a 47% relative rise in a month, and "reviews" grew by a third. "Specifications" entered Ahrefs' top 50 for the first time, while "meaning" and "definition" each fell by about a fifth.

Question type matters. In a separate Ahrefs test of 20,000 US prompts, 84.4% of factual questions about pricing, features or company details cited at least one official source, against 68.3% of comparison questions like "A vs B." For comparisons, most citations (53.6%) went to sites other than official ones.

Which websites does ChatGPT search with site:?

Mostly brand sites: 6.8% of fan-out queries used the site: operator, and among the ~160 most-searched domains, 75.9% of those searches targeted a brand's own website. Retailers took 11.9%, government and health authorities 7.3%, and Reddit and Facebook 1.9% between them. Apple (9,442 site: mentions), IKEA (7,361) and Nike (5,210) lead the list; consumer brands hold 8 of the top 10 spots. Reddit is ninth, and the only one of OpenAI's 17 licensing partners in the top 150.

Why is ChatGPT citing Reddit less?

ChatGPT cited Reddit in 2.3% of US answers in September 2026, down from 19.1% in March, an 88% drop, in Ahrefs' Brand Radar data. Wikipedia fell 80%, from 14.1% to 2.9% of answers. Ahrefs found the lost citations went to research bodies, regulators, brands and hands-on testers: PubMed climbed from 64th to 4th among ChatGPT's most-cited domains and the FDA entered at 8th, while explainer sites like WebMD and Verywell Health fell. Promptwatch tied a sharp drop in Reddit's citation share to a change in how ChatGPT fans out queries on August 8, 2026, though it calls the size of the drop provisional and OpenAI says it does not set visibility levels for individual sites.

Licensed publishers moved the other way. OpenAI's partner titles grew from 7.5% of citations among ChatGPT's 200 most-cited domains in March to 15.8% in July, and still held 12.5% in late September, even though ChatGPT rarely targets them with site: searches.

Methodology & caveats

Ahrefs' fan-out figures come from 66.6 million free, logged-out GPT-5.6 Luna responses in all countries, August and September 2026, built from prompts that stem from real search demand. Its citation figures come from a separate dataset, Brand Radar's US ChatGPT citations from March to September 2026 (2.4 to 3.8 million answers a month), and its official-source figures from a third, a test of 20,000 US prompts. Ahrefs checked its citation shifts on ~25,000 prompts that appeared in both March and September. Ahrefs sells Brand Radar, Peec AI sells AI visibility tracking, and Nectiv is an SEO agency.

Several values are read from Ahrefs' chart labels rather than its text: the site: source-type shares for government sites (7.3%), review platforms (1.2%) and other (1.9%), the partner citation shares for April to June and August, and the March and September citation rates for Reddit (19.1%, 2.3%), Wikipedia (14.1%, 2.9%) and YouTube (0.3%, 0.2%). Derived values: 9.6 million responses with fan-outs is Ahrefs' own figure (14.4% of 66.6 million); 53.6% of comparison-question citations to non-official sites is 100% minus 46.4%; the multiples in Figure 4 are each average divided by the lowest. Fan-out averages in Figure 4 come from different models, account types and prompt sets and are not a time series. Promptwatch's figures are cited through Semrush's report. All shares are as published, with no rounding or updating.