Between March and September, the sources ChatGPT links to in its US answers changed character. PubMed, the US National Library of Medicine’s research database, climbed from 64th to fourth among the domains the assistant cites most, according to data Ahrefs published on 12 October. The US Food and Drug Administration entered the list at eighth.

The sites that summarise other sources moved the other way. WebMD, Healthline, Investopedia and Britannica were among the domains cited less, alongside Reddit and Wikipedia. Reddit appeared in 19.1% of US answers in March and 2.3% by late September, according to Search Engine Journal’s reporting of the Ahrefs data. Wikipedia fell from 14.1% to 2.9%.

For B2B software marketers, the more useful number sits in a separate Ahrefs test. Across 20,000 US prompts, 84.4% of factual questions about pricing, features or company details cited at least one official source, and official websites accounted for 63.4% of all citations on those questions. When ChatGPT needs a fact, it increasingly goes to whoever owns the fact. That puts a direct question to content programmes weighted towards articles that restate what other sources already say.

What did Ahrefs find in ChatGPT’s citations?

Ahrefs drew the comparison from Brand Radar, its AI-visibility tracking tool, covering between 2.4 million and 3.8 million US ChatGPT answers a month from March to September 2026. The analysis, written by Louise Linehan, a content marketer at the company, was published alongside a wider study of how ChatGPT breaks a question into its own web searches.

The gainers fall into recognisable groups. Health and science authorities such as the CDC, the NIH and the NHS rose, as did US regulators including the SEC, the FTC and the Consumer Financial Protection Bureau. Professional bodies gained, among them the American Academy of Pediatrics. So did brands’ own help and support domains, including Apple, Google Support and Microsoft Support, and hands-on testing sites such as RTINGS and RepairPal.

The decliners share a different trait. Ahrefs listed health and finance explainers, encyclopaedias, marketplaces such as Alibaba and syndicating publishers including Yahoo and AOL among the domains cited less, and concluded that ChatGPT seems to be prioritising original sources over the sites that summarise them. Search Engine Journal reported that WebMD dropped from 18th to 79th, and that Britannica, Investopedia and the New York Post fell out of the top 200 altogether.

Ahrefs measures the share of answers that cite a domain. That differs from the share-of-citations figures tracking firm Promptwatch used when it recorded Reddit’s August fall in ChatGPT Search, so the two sets of numbers are not interchangeable, although both point the same way.

Licensing deals add a less tidy thread. Publications with OpenAI content agreements went from 7.5% of citations among the 200 most-cited domains in March to 15.8% in July, then eased to 12.5% by late September. That movement does not fit neatly into the primary-source pattern.

Why do official product pages matter more for B2B software brands?

Ahrefs’ 20,000-prompt test is the closest the study comes to a software-buying scenario. On comparison questions, 68.3% of answers cited at least one official source and official websites took 46.4% of citations. On straight factual questions about pricing, features or company details, the official share of citations rose to 63.4%.

The search data points the same way. Ahrefs examined 15 million of the background queries ChatGPT writes for itself and found the word “official” in 13.1% of English-language queries in September, up from 10.1% in August. Queries containing “price” rose from 6.04% to 8.88% over the same two months. Where ChatGPT limited a search to one named domain, 75.9% of those aimed at the roughly 160 most-searched domains went to brand sites.

That extends the change The Marketing Helm reported in September, when Promptwatch recorded ChatGPT starting to query named websites directly from 8 August. It also sits alongside a single-account traffic analysis by search consultant Suganthan Mohanadasan, which found factual queries routed towards official pages and opinion-led ones towards review sites and Reddit.

Buyers are using the channel. G2 surveyed 1,076 B2B software buyers and decision-makers in March 2026 and reported in April that 51% now start research with an AI chatbot more often than with Google, up from 29% in its 2025 report. A vendor’s pricing page, documentation and integration list now serve a second reader: the assistant a buyer asks what a product costs or whether it connects to an existing CRM. Pages that are out of date, gated or vague give ChatGPT less to cite, and risk leaving the answer to whichever source has documented the product more clearly.

How much weight can the Ahrefs data carry?

Ahrefs’ data describes a measured shift in one sample, and the company’s own caveats are specific. The March-to-September comparison covers US citations only. Because Ahrefs adds prompts to its tracking over time, it checked the movement against about 25,000 prompts that appeared among the 150,000 highest-volume prompts in both months, and found the charted sites moved in the same direction. It also notes that the prompts in any set shape which sites appear, and that OpenAI changes retrieval often enough for a model update to move the picture again.

The wider study is narrower than its headline figure suggests. Its 66.6 million responses came from free, logged-out sessions on a single model, GPT-5.6 Luna, in August and September, with no chat history or memory. Free sessions search less than paid or reasoning modes, Ahrefs says, and only 14.4% of the responses triggered a web search at all.

Independent work supports the direction without confirming the scale. A Semrush study published on 30 June ran 100 buyer-journey prompts, a quarter of them B2B SaaS, through GPT-5.2 in two modes. In high-reasoning mode, government and academic sources rose from 1.9% to 8.8% of citations and official documentation from 12.4% to 17.5%, while user-generated and review sites fell from 14.3% to 6%. It is a small sample measuring a different variable, and both Semrush and Ahrefs sell AI-visibility tracking. Neither study shows that publishing original research earns citations on its own; Ahrefs’ clearest evidence concerns brands’ own product facts.

Of the 66.6 million free ChatGPT answers Ahrefs examined, 85.6% were written without a web search.