Only 14.4% of the 66.6 million ChatGPT responses Ahrefs tracked between August and September 2026 triggered a web search — and when the model did search, it averaged just 1.6 fan-out queries per response that searched. Those figures come from what Ahrefs describes as the largest study yet of ChatGPT's retrieval behavior, covering 66.6 million free, logged-out GPT-5.6 Luna responses and 15 million fan-out queries across all countries.

The findings, published by Ahrefs content marketer Louise Linehan, point to a chatbot that answers most prompts from training data and, when it does look things up, increasingly goes straight to the original owner of the information — brand sites, regulators, and research bodies.

Why most ChatGPT answers never search the web

Every web search costs OpenAI money, so the model skips one when it can answer well enough from what it already knows. That aligns with WordLift's earlier research suggesting free-tier models lean more on training data to keep costs down.

The logged-out setup matters here. More than 90% of ChatGPT's weekly users are on the free plan, according to Olivier de Segonzac, co-founder of RESONEO, writing in Search Engine Land. Logging out also strips chat history and memory, exposing ChatGPT's default behavior rather than answers shaped by one person's past chats.

The practical consequence: for many prompts, your chances of appearing in ChatGPT answers depend on what the model already knows about your brand, not on what it can find.

What does ChatGPT search for when it does look?

Of the 15 million fan-outs analyzed, 6.8% used the site: operator to search a single website, spread across roughly 116,000 different domains. Among the ~160 most-searched sites, three in four of those searches targeted a brand site like apple.com or ikea.com. Retailers such as Walmart, Costco, and Amazon made up 11.9%, while Reddit and Facebook together accounted for just 1.9%.

The most site:-searched domains were dominated by big consumer brands:

• apple.com — 9,442 site: mentions
• ikea.com — 7,361
• nike.com — 5,210
• adidas.com — 4,619
• samsung.com — 4,432

Government domains also featured heavily: gov.br, gov.uk, service-public.fr, the NHTSA, the FDA, and the SEC all sit in the top 150. That matches observations from Lily Ray, founder of Algorythmic, that ChatGPT favors official sources for YMYL questions.

The language inside the fan-outs tells the same story. "Official" and "price" were the most common terms after "2026," and both grew month-over-month. "Price" went from 6.04% to 8.88% of fan-outs — a relative rise of nearly half in one month. "Reviews" grew by a third, and "specifications" entered the top 50 for the first time. "Meaning" and "definition" each fell by about a fifth.

Official sources for facts, third parties for comparisons

A separate Ahrefs test of 20,000 US prompts split behavior by query type. For factual questions about pricing, product features, or company details, 84.4% of answers cited at least one official source, and 63.4% of citations went to official websites. For comparison questions like "A vs B" or "which is cheaper," those numbers dropped to 68.3% and 46.4% respectively — most citations went elsewhere.

When ChatGPT has to weigh one option against another, it looks beyond official sources. That's third-party signal.

A running-shoes example illustrates it: of 1,474 US ChatGPT responses with fan-out queries in Brand Radar, "best" appeared 206 times, "2026" 160 times, "review" 117 times, "vs" 105 times, and "official" only 65 times. A running shoe brand, Ahrefs concludes, gains more from review coverage and comparison pages than from product pages alone.

Citations are shifting fast — who's winning?

Drawing on a separate Brand Radar dataset of 2.4 to 3.8 million US ChatGPT answers per month from March to September, Ahrefs documented steep citation declines for user-generated and reference platforms:

• Reddit: from 1 in 5 answers in March to about 1 in 43 by late September
• Wikipedia: from 1 in 7 to 1 in 35
• YouTube: from about 1 in 370 to 1 in 500

These are tool-signal findings rather than confirmed OpenAI statements, but they corroborate third-party reports: Snippet Digital found Reddit citations in ChatGPT dropped 86% in August, and data shared by Aleyda Solís, founder of Orainti, showed social, reference, and news sites losing the most citation share between July and August.

The citation share those sites lost went to research bodies, regulators, brands, and hands-on testers. PubMed climbed from 64th to 4th among ChatGPT's most-cited domains; the FDA entered at 8th. Health and finance explainers — WebMD, Healthline, Verywell Health, Investopedia — fell, as did encyclopedias, marketplaces, and syndicators like Yahoo, AOL, and the New York Post. Ahrefs validated the pattern against roughly 25,000 prompts that appeared among the 150,000 highest-volume prompts in both March and September; every site in the comparison moved in the same direction on identical prompts.

Meanwhile, OpenAI's licensing partners are gaining. Partner titles among ChatGPT's 200 most-cited domains more than doubled their combined citation share, from 7.5% in March to 15.8% in July, and still held 12.5% in late September.

How this differs from other studies

Ahrefs found far fewer fan-outs per prompt than other researchers — some API-based studies logged 7.61 fan-outs per prompt, and logged-in tests recorded 15 to 40 on Thinking mode. Ahrefs attributes the gap to methodology: its prompts stem from broad, real search demand across every topic and country, while many other studies used commercial "best X" prompts that naturally trigger brand lookups. Free, logged-out sessions also search less than paid or reasoning models.

What to monitor next

Ahrefs frames two viable paths into ChatGPT's answers: be well enough known that the model doesn't need to look you up, or be the original source it finds when it does. Pages that only summarize someone else's work are the ones currently losing citations; the pages worth investing in are original research, hands-on testing, and your own product facts, in plain text on pages ChatGPT can find. Ahrefs cautions that these numbers are a snapshot of one model between August and September 2026, and OpenAI changes how ChatGPT retrieves information constantly — so brands should track their presence in fan-out queries monthly rather than discover shifts months later.