No single domain runs ChatGPT search. On Profound's share of every citation from August 2024 to June 2025, Wikipedia led at 7.8% and Reddit was second at 1.8%. On Ahrefs' United States list updated 2 September 2026, which divides by the top sources only, Reddit led at 16.8% and English Wikipedia was second at 7.0%. Read the list that matches the question you are asking. Do not average them.
- Wikipedia was 7.8% of all ChatGPT citations in Profound's August 2024 to June 2025 window, and 47.9% of that window's top 10. Same row. Two denominators. Graded partial.
- Reddit was 16.8% of the US top-source pool in Ahrefs Brand Radar, snapshot updated 2 September 2026, with 810,887 distinct pages cited. That is not 16.8% of every ChatGPT citation.
- The top 10 domains were 12% of all citations in Profound's US English sample of about 730,000 cited conversations, October to December 2025. The head of the list is a minority of the links.
- Germany is a different list. In Ahrefs' August 2026 German snapshot, German Wikipedia was 18.67% of the top 50's citations and Reddit was third, behind ADAC.
- About 18% of US ChatGPT conversations in that October to December 2025 sample triggered a web search at all. A domain rank describes the conversations that cited the web, not every chat.
The two lists, side by side
People ask for the most-cited domains in ChatGPT as if one table existed. Two serious public tables do, and they answer different questions. One counts a domain's links against every citation the engine emitted. The other counts a domain's links against the links of the domains that already won.
The first is Profound's, published 5 June 2025 and marked updated in August 2025. The article describes a corpus of 680 million citations across ChatGPT, Google AI Overviews, and Perplexity. It does not give a ChatGPT-only citation count, and it does not say the sample is US-only. Grades on this page stay Partial for that reason.
| Rank | Domain | Share of all citations | Share of the top 10 only |
|---|---|---|---|
| 1 | Wikipedia | 7.8% | 47.9% |
| 2 | 1.8% | 11.3% | |
| 3 | Forbes | 1.1% | 6.8% |
| 4 | G2 | 1.1% | 6.7% |
| 5 | TechRadar | 0.9% | 5.5% |
| 6 | NerdWallet | 0.8% | 5.1% |
| 7 | Business Insider | 0.8% | 4.9% |
| 8 | NYPost | 0.7% | 4.4% |
| 9 | Toxigon | 0.7% | 4.1% |
| 10 | Reuters | 0.6% | 3.4% |
The second list is newer, American, and defined differently. Ahrefs Brand Radar, all topics, page updated 2 September 2026. Mention share is a domain's citations as a percentage of the summed citations of the top sources, not of every citation. The page does not state how many queries. Pages cited are distinct URLs.
| Rank | Domain | Mention share of top sources | Pages cited |
|---|---|---|---|
| 1 | reddit.com | 16.8% | 810,887 |
| 2 | en.wikipedia.org | 7.0% | 129,667 |
| 3 | consumerreports.org | 3.7% | 36,674 |
| 4 | forbes.com | 3.1% | 36,505 |
| 5 | engineerfix.com | 2.8% | 93,427 |
| 6 | legalclarity.org | 2.8% | 72,741 |
| 7 | healthline.com | 2.5% | 16,502 |
| 8 | youtube.com | 2.5% | 114,596 |
| 9 | pmc.ncbi.nlm.nih.gov | 2.4% | 69,801 |
| 10 | walmart.com | 2.3% | 144,045 |
Just outside that ten: Tom's Guide, Merriam-Webster, and TechRadar at 2.1%, then Edmunds and Home Depot at 1.9%. LinkedIn was 38th at 1.1%. Facebook was 26th at 1.4%. The community domain is Reddit. The reference layer is Wikipedia plus the dictionaries. The retailers are real and they are not the story of the top ten.
A later Profound cut, US English users on ChatGPT.com from October to December 2025, about 730,000 conversations that contained at least one web citation, did not reprint a full top ten. It did name four domains. Wikipedia was 5% of citations and appeared in 18% of conversations that cited anything. Reddit was 3% and 13%. Reuters and NIH were each 1% of citations and 4% of those conversations. The top ten domains together were 12% of all citations. The concentration was real. It was not a monopoly.
Why 47.9% and 7.8% are one row
The number that travels is 47.9%. It is Wikipedia's slice of ChatGPT's top ten in that Profound window, not its slice of the engine. Divide 7.8 by the top ten's combined share and you are back at 47.9. Quoting the larger figure as "almost half of ChatGPT citations" is how a head-of-list share becomes a market share.
Semrush measured something else again: whether a domain showed up in the answer, not what share of the links it took. From 14 July to 12 October 2025, on more than 230,000 prompts, Reddit was in close to 60% of ChatGPT responses in early August and around 10% by mid-September. Wikipedia fell from roughly 55% of responses to under 20%. Both were still the two most-cited domains at the end of that window. Forbes roughly doubled its response count after the drop. A presence rate and a link share can move together and still not be the same statistic.
Profound's own collaboration with Reddit, data through late October 2025, put the September break in the same place: Reddit's ChatGPT citation share fell from about 7% to about 1%, then rebounded to about 3%, and Reddit stayed second behind Wikipedia. That post is vendor-aligned. The direction matches Semrush. The levels should not be pasted onto the Ahrefs 16.8%.
Wikipedia's share of all ChatGPT citations, August 2024 to June 2025. The 47.9% figure is this row as a share of the top 10.
Reddit's share of the US top-source pool. Not a share of every citation. 810,887 distinct pages.
Share of all citations held by the top 10 domains, in a US sample of about 730,000 cited conversations.
The US list is not the German list
Ahrefs Germany, August 2026, divides each domain by the top 50's 408,723 citations. German Wikipedia led at 18.67%. ADAC, the motoring club, was second at 5.67%. Reddit was third at 5.53%, just ahead of Duden, the dictionary, at 5.19%. English Wikipedia was fifth at 4.65%. Stiftung Warentest, Idealo, FAZ, the consumer-advice centres, and Welt filled out the ten.
Japan moved the other way in the same season. Ahrefs' September 2026 Japan note says Reddit led the June ChatGPT list there and had left the top ten by September, with Kotobank first. The percentages were not in the text. Country is not a footnote on a citation league table. It is often the whole table.
An older Ahrefs cut, June 2025, about 957,000 ChatGPT prompts, had Wikipedia at a 16.3% mention share, with YouTube, Reddit, and Quora outside the top ten. Reuters sat at 4.3%. That is a third denominator and a third year. It is useful as history. It is a poor description of September 2026.
What a domain rank does not prove
Open question Nothing here is a quality ranking. Citation frequency is how often a retrieval system picked a host, on the prompts that vendor chose to run. EngineerFix and LegalClarity outrank YouTube on the September US mention-share list. That is a fact about that sample. It is not a finding that those sites are better sources than YouTube.
It also does not say what ChatGPT does when it does not search. In the US sample, about 18% of conversations triggered at least one web search, steady across October, November, and December 2025. Among conversations that did cite, there were about six unique citations per conversation. Turn one was 2.5 times as likely to carry citations as turn ten. A publisher who "ranks in ChatGPT" is visible in a minority of chats.
And it does not say the citation was right. The Tow Center's March 2025 test asked ChatGPT Search to identify a real article from an excerpt. It was wrong on 134 of 200. That test is old, and it is a lookup, not a sample of ordinary answers. It does not belong in the domain table. It does belong next to any plan that treats a citation as an endorsement. The retrieval behaviour is on ChatGPT citation statistics.
How to use the list
If you publish reference material in English for a US audience, Wikipedia and Reddit are the two hosts you are sharing the answer with, and you will not out-accumulate either of them. The opening is the other 88% of links in the late-2025 US cut, which is where Forbes, Healthline, PubMed Central, the retailers, and ordinary brand domains actually live.
If you publish in German, start from the German table. ADAC and Duden are not a curiosity. They are what a local list looks like when the encyclopedia is local and the institutions people trust are local.
Match the page type to the row you are trying to sit near, and then go one level down. A Forbes rank is often a buying guide, not a blog post. That cut is on most-cited pages, not domains. The movement of this list from one quarter to the next is on how the list changes.
Wikipedia was 7.8% of ChatGPT citations, and 47.9% of the top 10. Same row. The larger number is the one that travels.
Share on XVerification status
Every rank above is copied from the named study, with the denominator that study used. This site has not run its own ChatGPT citation panel. Profound's query mix is not public, so those shares stay partial. Ahrefs states a US, all-topics sample and a top-source denominator, and does not state the query count on the September ChatGPT page. Semrush's figures are response presence, and they are labelled as such. No row was filled in from a secondary roundup.
Namdev, R. (2026). Most-cited domains in ChatGPT search (v1). Retrieved from https://ritiknamdev.com/blog/most-cited-domains-chatgpt Published under CC BY 4.0 — reuse freely with attribution.
The cross-engine argument, and why these lists should not be pooled, is on most-cited domains in AI search. Retrieval behaviour sits on ChatGPT citation statistics.