There is no audit that opens the most-cited domains and counts which of them publish original data. The closest measurement is the type of the URL that got the citation. On marketing and SEO queries, Fractl classified original research and data at about 0.6% of citations. How-to pages were about a third. The domains that lead the broad US lists are Reddit, YouTube, and Wikipedia. None of the opened studies scored those three for original datasets. A research page can still be the source of a number that a how-to then gets cited for. That path was not counted.
- About 0.6% is the share of cited URLs classified as original research, on marketing and SEO queries. 10,108 answers, 67,951 citations. The tilde is Fractl's.
- The cited URL is usually a how-to, a product page, or a listicle. Those three are about 33%, 21%, and 20%. YouTube, which is a domain rather than a format, is about 11%.
- FAQ markup, an updated date, tables, and length were audited. Original data was not. Resonance's gaps are associations. Pages with an FAQ averaged 5.9 citations against 4.4 without.
- A schema correlation is not this measurement either. AI-cited pages were almost three times as likely to have JSON-LD, and adding JSON-LD to 1,885 pages did not raise citations.
- A 14% original-data share appears in an index that was not opened. It conflicts with 0.6% and it is not in the chart. A 2026 preprint on citation predictors was not read. It is not used as a domain ranking.
The audit that was not done
The question sounds like a domain list. Which of the sites in the head actually publish data of their own. Answering it properly means opening the domains, defining original data, and counting. That count was not in any source opened for this series. Inventing it from reputation, Forbes as a data brand, NerdWallet as a study mill, Wikipedia as a compiler, would be a list of opinions with percentages glued on.
What the domain lists do show is the kind of host that wins a broad US query set. Reddit, YouTube, and English Wikipedia lead or place on the September 2026 Ahrefs pages. A forum thread, a video, and an encyclopedia article are not, in the ordinary sense, original-research publications. That is a description of the format, not a score. Wikipedia's own sourcing rules are a different document, and they were not re-read for this page. The reference shares are on reference domains. The page types under the domains are on most-cited pages.
0.6% of citations, one vertical
Fractl, published 22 September 2026, classified Ahrefs Brand Radar citations from marketing and SEO queries. Six surfaces: ChatGPT, Gemini, Perplexity, Copilot, Grok, and Google's AI Overviews and AI Mode. 10,108 answers. 67,951 citations. The shares were printed with tildes. The method note says the citations were classified. It does not say whether a person or a model did it, and it does not define original research. Partial
| Type of cited URL | Share, as printed |
|---|---|
| How-to and explainer guides | ~33% |
| Brand product and service pages | ~21% |
| Listicles and ranked lists | ~20% |
| YouTube videos | ~11% |
| Opinion and editorial | ~2.3% |
| Comparison | ~1% |
| Original research and data | ~0.6% |
Fractl notes that YouTube is one domain sitting inside a table of formats. The chart keeps it, with that label, because dropping it would hide a tenth of the file. Fractl's interpretation is that research still matters indirectly. It earns rankings and links. The URL that receives the citation is the guide. That interpretation is plausible and untested by the classification. The 0.6% is the share of URLs whose type is the research page. It is not the share of answers that depend on a number first published somewhere else.
The vertical limit is the other half of the sentence. Marketing and SEO queries are a strange place to look for original data, and they are the only place this classification looked. A health, finance, or software cut could differ. Green Flag's finance panel is the nearest contrast that was opened. Government and regulator pages rose to 5.3% of the top-500 citations in that set by September 2026. Those are primary-source pages. They are not "original research" by a publisher, and they are one vertical's head, not a share of all finance citations.
What was counted instead of original data
Resonance audited about 190,000 cited pages in the B2B technology panel and looked for features that separate repeat citations from one-offs. Pages with FAQ markup averaged 5.9 citations, against 4.4 without, about 36% more. Pages showing an updated date were cited about 26% more often. Pages with tables, about 16% more. Pages over 5,000 words averaged about six citations, against about three for pages under 500 words. Article markup made no difference. The authors say these are associations, not causes. They do not report a count of pages that contain a dataset, a methods section, or the phrase original research.
They do say something practical about the research object itself. White papers and reports are cited when the text can be read. A report behind a form, or saved as images, is invisible. That is a retrieval fact, not a ranking of who publishes data.
Ahrefs' schema study is adjacent and easy to drag in, so it is fenced off. Among 6 million URLs, AI-cited pages were almost three times as likely to carry JSON-LD as pages that were not cited. Adding JSON-LD to 1,885 pages, from August 2025 to March 2026, against 4,000 controls, did not raise citations. AI Overviews were 4.6% lower than the controls, relative. AI Mode was 2.4% higher and ChatGPT 2.2% higher, both indistinguishable from zero. A markup correlation is not an original-data count. The test registered on this site has not enrolled a page: does schema markup increase AI citations.
The tactics people attach to "publish original data" are graded, with the evidence named, on how to get cited by ChatGPT. This page does not rerun that argument. It only refuses to pretend the domain head has been scored for original data.
What this does not prove
Open question It does not say original research is useless, and it does not say a how-to outranks a study as a source of truth. It says the cited URL, on one vertical, is rarely the study. It does not name which of Reddit, Wikipedia, YouTube, Forbes, or Healthline publish original data, because that audit was not done. It does not say tables cause citations. Resonance reported a gap and declined the causal verb. So does this page.
How to use it
If a brief says "publish original research, because that is what gets cited," the opened classification says the citation usually lands on the explainer that used the research. Publish the study if the study is the point. Do not expect the study URL to be the footnote as often as the guide that quotes it. And do not cite 0.6% outside marketing and SEO queries. That is the file it came from.
The series rule, one denominator per number, is the same rule as the hub.
On marketing queries, the cited URL classified as original research is about 0.6% of citations. How-to pages are about a third. Nobody has audited the most-cited domains for original data.
Share on XVerification status
Fractl's classification, Resonance's page audit, and Ahrefs' schema experiment were opened. No domain in a top-50 list was scored for original data by this site. Unopened percentages were left out.
Namdev, R. (2026). Which most-cited domains publish original data (v1). Retrieved from https://ritiknamdev.com/blog/most-cited-domains-original-research Published under CC BY 4.0 — reuse freely with attribution.
The page under the domain is classified on most-cited pages. A randomised schema test is registered, and not yet enrolled, at does schema markup increase AI citations.