ChatGPT’s citation playbook shifted between March and September, and the pattern benefits brands and publishers that generate first-hand evidence.
Ahrefs compared 2.4 million U.S. answers in March with 3.8 million in September to identify the 200 domains ChatGPT cited most. PubMed, the biomedical research database, jumped from 64th to fourth place. WebMD dropped from 18th to 79th, and other health and finance explainers such as Healthline and Verywell Health lost ground. Britannica, Investopedia and the New York Post fell out of the top 200.
Original sources are the new winners
The climbers were mostly government health and research sites, regulators and hands-on product testers. The CFPB and FTC moved up, support pages from Google and Microsoft gained visibility, and testing sites like RTINGS climbed. Ahrefs put it this way: ChatGPT “seems to be prioritizing original sources over sites that summarize them.”
Ahrefs also checked around 25,000 prompts that appeared in both months; the same pattern held, so new questions alone don’t explain the shift.
Reddit and Wikipedia still matter, but less
Reddit appeared in 19.1% of U.S. ChatGPT answers in March and just 2.3% in September. Wikipedia dropped from 14.1% to 2.9% over the same period. Both remain among ChatGPT’s most-cited domains, but the share of answers they influence has shrunk dramatically. In Ahrefs’ separate worldwide sample, Reddit still ranked ninth among domains ChatGPT searched with site: queries, so the platform remains a research shortcut.
Most free answers skip search
Out of 66.6 million free, logged-out GPT-5.6 Luna responses from August and September, only 14.4% included a web search. When search did happen, ChatGPT averaged around 1.6 fan-out queries, which are background searches it runs before giving an answer. Among 15 million fan-outs, 6.8% were site: searches across 116,000 domains. Of those, 75.9% went to brand sites such as apple.com or ikea.com.
Ahrefs cautions that fan-out rates from other studies aren’t directly comparable. One API-based study found 7.61 fan-outs per prompt, but different models, prompts and collection methods explain part of the gap. Free, logged-out use tends to search less than paid or reasoning models.
What to do about it
Citation rank is not traffic, but it reveals what ChatGPT considers credible. If your visibility depends on explainer pages, recheck which sources ChatGPT cites for your core topics after each OpenAI model update.
- Publish original research, hands-on testing and product information only you can create.
- Audit the sources ChatGPT cites for your priority topics, and note what it cites instead of you.
- Rebuild or reposition summary-heavy pages as first-hand evidence hubs.
Source: Search Engine Journal
