Top ChatGPT Citation Domains: In 7,656 Index Runs, No Site Passed 9%
Across 7,656 runs of ChatGPT models, the Machine Relations Index found 9,679 cited domains. TechRadar led at 9.0% of runs. No run cited reddit.com.
The sites cited by ChatGPT models in the Machine Relations Index do not reduce to a short list. In its release of October 29, 2026, the Index recorded 7,656 returned runs of its questions through ChatGPT models on OpenAI's API, and those runs cited 9,679 different domains. The most-cited, techradar.com, appeared in 687 runs (9.0%); only 18 domains reached 1% of runs; the ten most-cited made up 7.6% of the citations; and 4,428 domains were cited in exactly one run.
That is the answer to "which sites does ChatGPT cite most," and it changes what a top-domains list is good for. A ranked list of ChatGPT's favorite sources describes one surface, one window and one set of questions. The Index is a first-party dataset, published by Machine Relations Research and created by Jaxon Parrott, who founded AuthorityTech; it is not an independent audit. Treat it as a map of where to look, then read the sources ChatGPT cites on your own buyers' questions before you fund a placement.
Key takeaways
- No site dominates. The leader, techradar.com, appeared in 9.0% of ChatGPT runs; youtube.com was second at 7.3%; every other domain was below 2.5%.
- The tail is the answer. The ten most-cited domains made up 7.6% of the 32,024 domain citations in ChatGPT runs and the top 100 made up 18.9%. 4,428 of 9,679 domains (45.7%) appeared in one run.
- Studies of ChatGPT disagree. Semrush found Reddit and Wikipedia were ChatGPT's two most-cited domains in a study ending October 12, 2025; the Machine Relations Index, a first-party dataset created by AuthorityTech's founder, found in its October 29, 2026 release that no ChatGPT-model run cited reddit.com and wikipedia.org appeared in 1.6% of runs. The surface, window and questions differ.
- The same site is cited at very different rates by engine. techradar.com appeared in 9.0% of ChatGPT runs and in 0.1% of Google AI Mode runs. reddit.com appeared in 19.7% of Perplexity runs and in none of ChatGPT's.
- Measure ChatGPT on its own. Take your buyers' questions to the ChatGPT surface they use, record the cited sources, and rank the outlets from that baseline.
Which domains did ChatGPT cite most?
The Machine Relations Index, published by Machine Relations Research at machinerelations.ai/index, asks six AI engines buying and research questions drawn from a fixed basket of 24 subject categories plus a legacy news-topic bucket, and records the domains each answer cites. For ChatGPT, the release of October 29, 2026 covers 7,656 returned runs from May 10 to October 29, 2026. The Index reaches ChatGPT through OpenAI API models, not the consumer app: 5,087 of those runs used gpt-4.1-mini, 1,969 used gpt-4o-mini-search-preview, 384 used gpt-4o-mini and 216 did not record the model. The table lists the fifteen domains cited in the most runs. A domain counts once per run, however many of its pages the answer cites.
| Rank | Domain | Source role in the Index | ChatGPT runs citing it | Share of 7,656 runs |
|---|---|---|---|---|
| 1 | techradar.com | Editorial publication | 687 | 9.0% |
| 2 | youtube.com | Search or media platform | 558 | 7.3% |
| 3 | tomsguide.com | Editorial publication | 181 | 2.4% |
| 4 | techtarget.com | Editorial publication | 171 | 2.2% |
| 5 | arxiv.org | Academic and government source | 159 | 2.1% |
| 6 | axios.com | Editorial publication | 156 | 2.0% |
| 7 | prnewswire.com | Wire and press-release distribution | 144 | 1.9% |
| 8 | itpro.com | Editorial publication | 137 | 1.8% |
| 9 | kiplinger.com | Editorial publication | 122 | 1.6% |
| 10 | wikipedia.org | Academic and government source | 119 | 1.6% |
| 11 | healthline.com | Editorial publication | 118 | 1.5% |
| 12 | linkedin.com | Community and social platform | 117 | 1.5% |
| 13 | nerdwallet.com | Editorial publication | 113 | 1.5% |
| 14 | forbes.com | Editorial publication | 104 | 1.4% |
| 15 | scienceinsights.org | Other observed source | 93 | 1.2% |
The list mixes consumer and general-technology publications with business ones, and the Index's categories include consumer ones. It is a ranking of this question set on these models, not of every question anyone asks ChatGPT. The source roles are the Index's own labels: it classifies a domain as an editorial publication, a wire, a community platform and so on, and files domains without role evidence under "Other observed source." Those domains received 23,581 of ChatGPT's 32,024 citations (73.6%), more than every named role combined. Editorial publications received 3,469 (10.8%), vendor-owned sources 1,924 (6.0%) and wire and press-release distribution 316 (1.0%).
How concentrated is ChatGPT's citation list?
Less than most lists imply. After the top two, the share per domain falls below 2.5% and keeps falling. 443 domains were cited in ten or more runs; 4,428 were cited in one. A pitch list built from the top fifteen covers a small part of what ChatGPT drew on in this set of questions.
A smaller benchmark reports a higher figure, and the two are not like-for-like. Gadex put 50 B2B software prompts to ChatGPT through its API on 11 August 2026 and counted each distinct URL once per answer: 209 ChatGPT citations on 80 distinct domains, with the ten most-cited domains accounting for 34.4%. The Index counts each domain once per run, over 32,024 citations on 9,679 domains. With only 80 domains in the pool, the ten most-cited of them must carry at least 12.5% of the total, more than the 7.6% the ten most-cited carry among 9,679. The Index's 7.6% describes its own question set and models, not every ChatGPT question.
Why do studies of ChatGPT's top sources disagree?
Each study measures a particular surface in a particular window with its own prompts. Semrush analysed more than 230,000 prompts over 13 weeks, July 14 to October 12, 2025, on ChatGPT search, Google AI Mode and Perplexity, and reported that Reddit and Wikipedia were still ChatGPT's two most-cited domains after a sharp drop in mid-September 2025: Reddit from close to 60% of prompt responses in early August to around 10% by mid-September, Wikipedia from roughly 55% to under 20%.
The Index finds something else a year later. It queries OpenAI API models, mostly gpt-4.1-mini (see the method note). No ChatGPT-model run in the release cited reddit.com, and wikipedia.org appeared in 119 runs (1.6%). Jaxon Parrott, who created the Index and founded AuthorityTech, published an analysis of the October 14 release that found that seven of the eleven top-100 domains with no ChatGPT citation, Reddit among them, served a robots.txt that disallowed OpenAI's search crawler that day. The essay calls this an association, not a switch. OpenAI's crawler documentation says sites that opt out of its search crawler will not be shown in ChatGPT search answers, though they can still appear as navigational links.
The data cannot say which difference explains the gap. None of this makes either study wrong. It means a "top ChatGPT domains" list needs a surface, a window and a question set stated beside it, and a plan built on one should be rechecked against your own questions.
Does the same domain get cited the same way across engines?
No. The table gives the share of each engine's runs that cited the domain, with counts, in the same release. Engines also differ in how often a run cites anything at all, and the Index marks some engines' coverage as degraded, so read it as rates in this release, not as engine preference.
| Domain | ChatGPT | Claude | Gemini | Google AI Mode | Google AI Overviews | Perplexity |
|---|---|---|---|---|---|---|
| techradar.com | 9.0% (687 of 7,656) | 0.6% (46 of 7,692) | 0.2% (11 of 6,843) | 0.1% (1 of 1,966) | 0.2% (2 of 1,218) | 2.1% (165 of 7,772) |
| youtube.com | 7.3% (558 of 7,656) | 2 of 7,692 | 4.9% (334 of 6,843) | 12.9% (253 of 1,966) | 25.0% (305 of 1,218) | 8.6% (672 of 7,772) |
| reddit.com | none in the release | none in the release | 7.3% (498 of 6,843) | 16.9% (333 of 1,966) | 18.1% (221 of 1,218) | 19.7% (1,533 of 7,772) |
| linkedin.com | 1.5% (117 of 7,656) | 0.5% (36 of 7,692) | 0.0% (1 of 6,843) | 23.8% (468 of 1,966) | 13.3% (162 of 1,218) | 10.8% (842 of 7,772) |
| forbes.com | 1.4% (104 of 7,656) | 1.9% (149 of 7,692) | 2.5% (174 of 6,843) | 5.6% (111 of 1,966) | 3.0% (37 of 1,218) | 7.8% (606 of 7,772) |
| wikipedia.org | 1.6% (119 of 7,656) | 0.8% (64 of 7,692) | 0.4% (29 of 6,843) | 0.5% (10 of 1,966) | 0.2% (2 of 1,218) | 1.1% (86 of 7,772) |
techradar.com is a ChatGPT story in this release: 9.0% of ChatGPT runs and 2.1% of Perplexity's, and under 1% on the other four engines. reddit.com and linkedin.com run the other way: strongest on Perplexity and Google's AI Mode and AI Overviews, weak or absent on ChatGPT and Claude. A source-list strategy that averages the engines describes none of them. The engine-by-engine comparison and the Citation Gap analysis cover why rankings and citations part ways.
What to do with a ChatGPT top-domains list
- Start from your buyers' questions, not a ranking. Write the questions they ask about your category, run them in ChatGPT, and record every cited source. The ChatGPT citation guide covers the baseline.
- Check the surface. A result from ChatGPT search, from the API and from a logged-in app can differ. Record which one you measured.
- Read the outlet's crawler rules before you pitch it for ChatGPT. An outlet that disallows OpenAI's search crawler is a weak ChatGPT bet however well it ranks elsewhere.
- Pick outlets per category. The publications index ranks the editorial publications engines cite in each buyer category, and a fixed list of fifteen sites does not.
- Match the outlet to the job. Reach, credibility and AI citation are different jobs, as the paywalled-publications analysis shows.
- Re-run the same questions on a schedule and compare with the baseline. A later citation after a placement is an observation, not proof the placement caused it.
This is the work AuthorityTech, The World's First Machine Relations Agency, does for clients, on the discipline machinerelations.ai defines. Start your visibility audit → to see which sources the engines cite on your category's questions.
Frequently asked questions
Which website does ChatGPT cite most?
In the Machine Relations Index release of October 29, 2026, techradar.com: 687 of 7,656 ChatGPT runs (9.0%), ahead of youtube.com at 7.3%. A different study at a different time found Reddit and Wikipedia at the top, so treat any single ranking as a snapshot.
Does ChatGPT cite Reddit?
It depends on the surface and the date. Semrush reported Reddit cited in close to 60% of ChatGPT search prompt responses in early August 2025 and around 10% by mid-September. In the Machine Relations Index release, no ChatGPT-model run cited reddit.com. See the method note.
Does ChatGPT cite press release wires?
Sometimes. prnewswire.com appeared in 144 ChatGPT runs (1.9%) and businesswire.com in 83 (1.1%); wire and press-release distribution as a role received 316 of 32,024 ChatGPT citations (1.0%). The release does not show what a given press release earned.
How many domains does ChatGPT cite?
In the Machine Relations Index release of October 29, 2026, 9,679 different domains across 7,656 runs of ChatGPT models, an average of 4.2 cited domains per returned run. Almost half appeared in a single run.
Method note. All figures are from Machine Relations Index release mri_score_v2.1+2026-10-29+225b998b091e, window 2026-05-10 to 2026-10-29, read from the public file at machinerelations.ai/data/machine-relations-index.json and its manifest; the file updates daily, so later releases will differ. Rates are the Index's per-engine rate: runs citing the domain over that engine's runs (ChatGPT 7,656, Claude 7,692, Gemini 6,843, Google AI Mode 1,966, Google AI Overviews 1,218, Perplexity 7,772). Shares of citations divide by the engine's total domain citations (ChatGPT 32,024). The Index reaches ChatGPT through OpenAI API models; a run is one returned answer, and the 7,656 runs split as gpt-4.1-mini 5,087, gpt-4o-mini-search-preview 1,969, gpt-4o-mini 384 and model not recorded 216, per the file's per-category model counts. The statement that no ChatGPT-model run cited reddit.com means the domain's per-engine record in this release lists no entry for that engine; it says nothing about other releases, surfaces or dates. The Index is published by Machine Relations Research and created by Jaxon Parrott, who founded AuthorityTech; it is not an independent audit. Domains are root domains, except that the Index lists YouTube channels as separate youtube.com/@handle rows (20 of the 9,679 ChatGPT-cited rows).