ChatGPT is skipping the explainer page

An explainer page is a page that restates what primary sources say in plainer language, and ChatGPT is citing fewer of them: between March and September 2026, its most-cited US domains shifted toward the sources those explainers summarize.

The data comes from an Ahrefs study published on October 12, 2026, built on its Brand Radar citation set. Ahrefs puts its reading in one line: ChatGPT "seems to be prioritizing original sources over sites that summarize them." That is an interpretation of observed citations, not a statement from OpenAI, and it should be read that way. The pattern is still worth acting on, because it lines up with what the search behavior in the same study shows.

#What moved

According to the Ahrefs study, PubMed climbed from 64th to 4th among the domains ChatGPT cites most in US answers, and the FDA entered the list at 8th. WebMD went the other way, from 18th to 79th, according to Search Engine Journal's write-up of the same data. SEJ also reports that Britannica, Investopedia and the New York Post fell out of the top 200 entirely.

The climbers fall into a few groups. Ahrefs lists government health and research bodies, regulators, professional bodies, brand support pages and hands-on product testers such as RTINGS. The decliners are health and finance explainers (WebMD, Healthline, Investopedia), encyclopedias, syndicators and some marketplaces.

The two community and reference giants dropped too. SEJ reports Reddit was cited in 19.1% of US answers in March and 2.3% in September, and Wikipedia went from 14.1% to 2.9% over the same months. Both are still among the most-cited domains. They are just far less dominant.

A reasonable worry with any month-over-month comparison is that the prompt set changed. Ahrefs checked that. It compared about 25,000 prompts that appeared in both March and September among its 150,000 highest-volume prompts, and the movement held in the same direction for those matched prompts.

#Why the search behavior points the same way

The citation shift is easier to believe once you look at what ChatGPT searches for. In the same study, Ahrefs analyzed 66.6 million free, logged-out responses and 15 million fan-out queries from August and September 2026. A few findings stand out:

• Only 14.4% of responses triggered a web search at all.
• 6.8% of fan-out queries used the site: operator, and 75.9% of the most-searched site: targets were brand sites, against 11.9% for retailers.
• The word "official" rose from 10.1% to 13.1% of English fan-outs between August and September. "Price" rose from 6.0% to 8.9%.
• In a separate 20,000-prompt test, 84.4% of factual prompts cited at least one official source, compared with 68.3% of comparison prompts.

When the engine writes a query that asks for the official page, or restricts the search to a specific domain, the summary page is never in the candidate set. It does not lose a ranking contest. It is filtered out before the contest starts.

#How far to trust it

Ahrefs is clear about the limits, and they matter. The fan-out data is one model, logged out, over two months. Free sessions search less than paid or reasoning models, and OpenAI changes retrieval often. We wrote about that last point in model swaps are the new core updates: a large share of what looks like a trend can be one release.

Other datasets disagree on scale. Nectiv ran about 4,000 prompts through GPT-5.6 Sol via the API and found an average of 7.61 fan-outs per prompt, with 64% of fan-out queries using site:. That is a different model, a different surface and a different prompt mix, so neither number is the true one. The surface difference is the same one we described in logged-out ChatGPT is its own surface.

Citation rank also measures visibility, not traffic. A site can fall in citation share and keep its visitors, or the reverse. The direction across both studies is consistent, though: the engine goes looking for the source that owns the fact.

#What this means if you publish summaries

If your content strategy is "explain the topic better than the primary source does", the ground under it is moving. A clearer rewrite of an FDA page or a manufacturer's spec sheet now competes with the original in a retrieval step that increasingly asks for the original by name.

Google's own guidance has asked the same question for years. Its helpful content page asks whether content provides "original information, reporting, research, or analysis" and whether it avoids "simply copying or rewriting those sources". Ahrefs' data suggests ChatGPT's retrieval now behaves as if it were asking the same thing.

Practical steps:

1. Audit which of your cited pages are summaries. Pull the pages that earn AI citations and tag each one: primary (your data, your product, your testing) or secondary (a restatement of someone else's facts). The secondary group is where the exposure sits.
2. Put your own facts in plain text on your own domain. Prices, specifications, compatibility, support steps and policies should live on crawlable pages, stated directly. When a fan-out query targets site:yourbrand.com, that is the page it has to find.
3. Add something the original does not have. A summary survives when it carries data of its own: a test, a measurement, a dataset, a worked case. If a page has nothing that cannot be found upstream, assume the engine will go upstream.
4. Link to the primary source and say what you added. A page that cites the regulator and then states what it measured on top is easier to justify as a separate source than one that paraphrases.
5. Track citations by source type, not only by domain. Classify the URLs cited for your prompt set as official, testing, explainer, community or marketplace. A move from explainer to official in your own category is the earliest signal that this shift has reached you.
6. Annotate the window. Mark March to September 2026 on your ChatGPT reporting, and keep ChatGPT data separate from other engines so a change in one does not get averaged away.

#The short version

The pages holding up in this data are the ones that own the fact being asked about. If your page explains someone else's fact, the engine now has a shorter path to the owner. The fix is not better paraphrasing. It is publishing something only you can publish, and making your own facts easy to retrieve.