PubMed jumped from 64th to 4th place in ChatGPT’s most-cited domains between March and September, while WebMD dropped from 18th to 79th. That’s from Ahrefs’ analysis of 2.4 million U.S. ChatGPT answers in March and 3.8 million in September, published October 12, 2026. Ahrefs concludes that ChatGPT “seems to be prioritizing original sources over sites that summarize them.”
What Did Ahrefs Actually Measure?
Ahrefs tracked two separate things: which domains ChatGPT cited most often in answers, and how ChatGPT searches the web before answering. The citation data covers the 200 most-cited domains in U.S. ChatGPT answers. The search behavior data comes from 66.6 million free, logged-out GPT-5.6 Luna responses and their 15 million fan-out queries (background searches ChatGPT runs before generating an answer) from August and September.
To check whether the shifts were real or an artifact of new prompts entering the dataset, Ahrefs compared roughly 25,000 prompts that appeared in both March and September. Every site on the chart moved in the same direction for those prompts.
Which Sites Rose and Which Fell?
Most of the climbers were government health and research sites, regulators, and hands-on product testers. Most fallers were health and finance explainers. Britannica, Investopedia, and the New York Post fell out of the top 200 entirely.
- PubMed: 64th to 4th
- WebMD: 18th to 79th (Ahrefs describes it as a health site that summarizes research)
- Healthline and Investopedia: also declined, grouped by Ahrefs with health and finance explainers
- FDA and RTINGS (a hands-on product testing site): increased in citation rank
- CFPB and FTC: appear among sites that gained ground as regulators
Reddit’s presence in answers fell sharply. Reddit was cited in 19.1% of U.S. ChatGPT answers in March and 2.3% in September, an 88% drop in six months, according to Ahrefs. Wikipedia dropped from 14.1% to 2.9% over the same period. Ahrefs notes both are still among ChatGPT’s most-cited domains overall.
How Often Does ChatGPT Search the Web?
Less often than many assume. Only 14.4% of the 66.6 million free, logged-out responses triggered a web search. When a search did happen, ChatGPT ran about 1.6 fan-out queries on average. Ahrefs notes that more than 90% of ChatGPT’s weekly users are on the free plan, citing Olivier de Segonzac, co-founder of RESONEO, in Search Engine Land, so free sessions are closer to how most people use ChatGPT.
For comparison, Nectiv measured an average of 7.61 fan-out queries per prompt from roughly 4,000 prompts processed through GPT-5.6 Sol via the API. Ahrefs notes the two datasets differ in model, prompts, and collection methods, so the numbers are not directly comparable. Paid and reasoning models also search more than free, logged-out sessions, per Ahrefs.
What Does ChatGPT Look for With Site: Searches?
When ChatGPT uses a site: operator (a search restricted to a single website) to look something up, it overwhelmingly targets brand sites. Of the roughly 160 most-searched domains, 75.9% of site: searches went to brand sites, defined by Ahrefs as company websites like apple.com or ikea.com. Retailers like Walmart, Costco, and Amazon made up 11.9%. Reddit and Facebook together accounted for just 1.9%.
The language inside fan-out queries reinforces the trend. “Official” appeared in 13.1% of English fan-out queries in September, up from 10.1% in August. “Price” and “review” fan-outs rose up to 47% in a month, according to Ahrefs.
In a separate test of 20,000 U.S. prompts, Ahrefs found that 84.4% of factual questions about pricing, product features, or company details cited at least one official source. For comparison questions such as “A vs B” or “which is cheaper,” that share dropped to 68.3%. Among citations for factual questions, 63.4% pointed to official websites; for comparisons, it was 46.4%.
Reddit was still the ninth most site:-searched domain in Ahrefs’ worldwide sample, consistent with what Suganthan Mohanadasan, co-founder of Snippet Digital, has observed: ChatGPT still fetches Reddit when it wants real users’ opinions, even as it cites it less often in final answers.
Of OpenAI’s 17 current publisher and platform licensing partners, Reddit is the only one in the top 150 most site:-searched domains. Ahrefs offers one possible explanation: ChatGPT may already have partner content pre-loaded. David McSweeney, founder of QueryBurst, found that some high-authority publications including Time, Business Insider, and Forbes appear to have pre-synthesized, page-level summaries dropped directly into ChatGPT’s context window (the working memory an AI model uses to generate a response).
What This Means for AI-Search Visibility
The practical shift is simple: if your content mainly explains what a government agency or research database already published, ChatGPT may be more likely to cite that original source instead, Ahrefs says.
In health, for example, WebMD, which summarizes medical research, fell from 18th to 79th, while PubMed climbed to 4th and the FDA moved up, per Ahrefs. For a brand in an information-heavy category, the useful question is less am I ranking well in Google? and more am I the source, or the explainer of someone else’s source?
Three patterns from the Ahrefs data carry the clearest implications:
- Most free, logged-out responses skip web search. Only 14.4% of the responses in Ahrefs’ sample involved a search. In Hingewise’s reading, that means visibility for many questions depends on what the model already “knows” about your brand, not only on what it can find on the web at that moment.
- Brand sites are the primary target for site: searches. Ahrefs found that three in four of the most common site: searches targeted a brand site, so having your own pricing, specs, and support details in plain, crawlable text on your site matters when ChatGPT looks for first-party facts.
- Citation rank is not traffic. As Search Engine Journal notes, Ahrefs’ citation ranks don’t measure traffic. A site can rank well in ChatGPT citations and still see little measurable traffic, so the business value depends on whether those citations appear for questions that influence decisions.
In Hingewise’s view, the Reddit numbers are the most instructive part of this report because they show how far apart searching and citing can be. Reddit went from 19.1% of U.S. answers in March to 2.3% in September, yet in Ahrefs’ separate worldwide sample it was still the ninth most site:-searched domain. Ahrefs links this to Suganthan Mohanadasan’s finding that ChatGPT still fetches Reddit for real users’ opinions. The two figures come from different samples, so they are not a direct comparison, but they suggest that a brand tracking only final citations may miss part of what ChatGPT reads before it answers.
Ahrefs is direct about the content implication: focus on pages only you can create, such as original research, hands-on testing, and your own product information.
Before your next content audit, check these:
- Ask ChatGPT your core product or service questions and note which sources it cites.
- Check whether your pricing, specs, and support details are in plain, crawlable text on your brand site.
- Identify whether your content is original or primarily summarizes third-party sources.
- Determine whether you are a primary source for any data, study, or hands-on result others in your industry cite.
- Compare your citation share before and after the most recent ChatGPT model update.
- For topics where you have dropped in citations, check which sources now appear instead.
Ahrefs notes that its fan-out data is a snapshot of a single model from August and September, and that these patterns could change in future updates. The 25,000-prompt check applies to the citation rankings, not the fan-out data, so the next model update is a natural point to recheck both.
Sources: Search Engine Journal, October 12, 2026; Ahrefs, October 12, 2026.
Lam Nguyen · Hingewise
This article was drafted with AI assistance from the sources cited above and checked against them before publishing. The "What this means for AI-search visibility" section is Hingewise's editorial analysis, built on the frameworks and positions we use in our GEO work. How we work: Editorial policy.