The Sites ChatGPT Cites Most, and Why the List Won’t Hold

Last October this page reported on one study. Ahrefs had taken the 1,000 pages ChatGPT cited most in September 2025 and sorted them by type. Wikipedia sat at the top. Only 32.3% of the pages were ones a marketer could realistically pitch. Just over 28% had no organic search visibility at all.

In September 2026, Green Flag Digital re-ran its own ChatGPT citation study on personal finance answers, eleven months after the first pass. Of the 215 pages ChatGPT had cited most in October 2025, 107 now get no citations at all. Wikipedia and Reddit went from 18% of the top-500 citations in that set to none.

So before any list of domains, one warning. A list of ChatGPT’s most-cited sources is a snapshot of one set of questions in one month. It is useful. It is not a map.

What the named studies agree on

Three studies measured this differently and still overlap. Profound’s analysis of citations across AI platforms found that within ChatGPT’s ten most-cited sources, Wikipedia accounted for 47.9% of citations. Semrush tracked cited domains over three months and found that before mid-September 2025, ChatGPT cited Wikipedia and Reddit roughly five times as often as the next domains. After that point the gap narrowed, with Medium, Forbes and LinkedIn closer behind and Forbes doubling the number of ChatGPT responses that cited it. Semrush lined the shift up with Google’s removal of the num=100 search parameter around September 11, 2025.

BuzzStream then compared several of these lists side by side. Forbes was the most consistently recurring domain after Reddit and Wikipedia. G2 appeared in the SaaS breakdown. And one dataset contained no Reddit or Wikipedia at all, which BuzzStream read as evidence that the questions behind a study shape its results.

Read together, the durable pattern is short. A general reference site at the top. A community site close behind or falling away, depending on the month. Large editorial publishers. Review platforms wherever the question is a purchase. Below that, each category’s own specialists.

The list describes the prompts, not ChatGPT

That last finding decides whether a list is any use to you. Every citation study starts with a set of questions someone chose to ask. Green Flag says so in its limitations: its data comes from one vendor’s prompt set, and a change in those prompts between two dates would look exactly like a change in ChatGPT.

A study weighted toward general knowledge questions will crown Wikipedia. One weighted toward product questions will surface marketplaces, G2 and review publishers. Finance prompts return banks and rate sites. None of them tell you what ChatGPT cites when your buyer asks about your category, because your buyer’s questions weren’t in the set.

The finance re-run shows what a narrow, consistent set can reveal. Between October 2025 and September 2026, first-party pages from banks, card issuers and airlines went from 2% of top-500 citations to 22%. Ask ChatGPT about a specific card and it increasingly goes to the issuer’s own page. A global top-domains list would never show you that shift.

Why these domains keep winning, and what authority buys

ChatGPT can only cite what its search step retrieves, the pipeline walked through in why AI search names your competitor and not you. Domain authority works at the entrance to that pipeline, not at the end of it.

Green Flag’s domain data shows how. Across 999 finance domains, sites rated DR 90 and above took half of all citations. They did it by having more pages cited, not more citations per page: the median per cited page stayed flat across authority bands. Inside Investopedia alone, where authority is constant, referring domains showed no positive link to a page’s citations. How many ways the page ranked, and how many people asked the question, did.

That explains the zero-visibility pages in the Ahrefs data. A page with no rankings can still be cited if it sits on a domain the model already draws from and answers the exact narrow question the model searched for. It also explains the decay. In the finance set, pages with a year in the URL lost nearly all their citations once the year turned, while constantly updated rate and comparison pages gained.

So the reliable way onto the cited list is coverage. One page per distinct question, kept current, on a site with enough authority to be in the pool at all. That is ordinary search work on indexation, rankings and site authority, aimed at the questions a model asks as well as the ones people type.

How to analyze ChatGPT citations for your own category

Since the public lists describe someone else’s prompts, build your own.

Write down the questions your buyers ask before a purchase, in their words. Run them in ChatGPT with search on, and log every cited domain against every question. Then sort the domains into four groups: your own site, competitors’ sites, third-party pages you can reach, such as review platforms, trade publications and comparison articles whose authors update them, and pages you can’t, such as encyclopedias, government sites and app stores. The third group is your outreach list. Questions answered only from the fourth group are ones to leave alone. Questions where a competitor’s own page is cited need a better page of yours, which is the job of content planned around the questions buyers ask.

Repeat it monthly. The finance data says a citation is closer to rented than owned, and a list that was accurate in March can be half wrong by September.

One thing the analysis won’t show is paid placement. Advertising inside ChatGPT is a separate system with its own problems, covered in why ChatGPT’s ads are struggling to win consumers over. Buying an ad doesn’t put your domain in the sources.

A published citation list is a weather report: accurate for its place and week, and a poor guide to where you should build. Your own prompt set, rerun on a schedule, is the closest thing to a map, and it lets you measure AI answers the way you measure the other channels you already run.

Frequently Asked Questions

Which websites does ChatGPT cite most?

In the large published studies, Wikipedia leads ChatGPT’s citations, with Reddit, Forbes, Medium, LinkedIn and review platforms such as G2 recurring near the top. Profound found Wikipedia held 47.9% of citations among ChatGPT’s ten most-cited sources, and Semrush saw Wikipedia and Reddit’s lead narrow sharply after mid-September 2025. Rankings differ by study because each uses its own questions, so treat any single list as a snapshot rather than a ranking to plan around.

Does ChatGPT only cite high-authority websites?

No, but authority helps a site get into the pool. In Green Flag Digital’s finance data, sites rated DR 90 and above took half of all citations by having more pages cited, while their citations per cited page were no higher than smaller sites’. Some low-authority pages that answered one narrow question directly still earned citations. Ahrefs also found that just over 28% of ChatGPT’s top 1,000 cited pages had no organic search visibility.

How often do ChatGPT’s cited sources change?

Often enough that a year-old list can mislead you. In Green Flag Digital’s re-run, half of the 215 most-cited finance pages from October 2025 had no citations eleven months later, and pages with a year in the URL fared worst. If ChatGPT citations matter to your pipeline, check your own category monthly rather than relying on a published ranking.

Can we get our own site cited instead of relying on third parties?

Sometimes, and more often for questions about your own product. In the finance set, first-party pages from banks and card issuers went from 2% to 22% of top citations within a year, as ChatGPT went to the source for product-specific questions. For category questions such as “best tool for X,” independent review and comparison pages still carry most of the weight. You need a clear, current page for every question about your product, and a presence on the independent pages that answer category questions.