The sources AI cites most
80,614 answers from 3 answer engines, measured every day on real commercial prompts. Every percentage here has a denominator you can check.
What this meansHide
One answer from one engine on one day. Every percentage on this page is a fraction of this number, including answers that cited nothing at all.
What this meansHide
Individual links, so an answer citing four sources contributes four. The ranking counts answers rather than links, so one page linked repeatedly cannot inflate its own standing.
What this meansHide
Domains cited at least 5 times, and cited for several independent businesses rather than only one. The long tail below that is measured but not published, so no single niche can be mistaken for the web.
What this meansHide
The share of all analysed answers that cited this domain at least once, across every engine in the index.
The engines do not agree with each other
A single blended ranking hides the most useful thing in this dataset: the same source can be a quarter of one engine's citations and a rounding error on another.
Cited 26× more often by Google AI Overviews
Google AI Overviews
- 21.6%
ChatGPT
- 0.8%
Cited 6× more often by Google AI Overviews
- 17.4%
- 2.7%
Cited 25× more often by Google AI Overviews
- 9.8%
- 0.4%
Why Wikipedia is not at the top of this list
Every widely-quoted study of AI citations puts Wikipedia at or near the top for ChatGPT. This one does not, and the reason matters more than the ranking.
Published studies have put Wikipedia’s share of ChatGPT citations anywhere between roughly 8% and 48%. Those studies disagree with each other by a factor of six, and none of them states the mix of questions behind the number.
That mix is the whole answer. Wikipedia is what an engine reaches for when asked what something is. It is not what it reaches for when asked which product to buy. Every prompt in this index is the second kind, and Wikipedia lands at rank 81, in 0.7% of ChatGPT answers.
The check that the method is sound is Reddit: this index puts it in 17.4% of Google AI Overviews answers, against the roughly 21% widely reported elsewhere. Close agreement where the question types overlap, and a wide gap exactly where they do not.
The 20 most cited domains in AI search
Ranked by citation share: the percentage of analysed answers that cited each domain at least once, across every engine. Click a source for its daily history.
| # | Source | Share of all answers | ChatGPT | Google AI Overviews | Google AI Mode |
|---|---|---|---|---|---|
| 1 | 11.5% | 0.8% | 21.6% | 13.1% | |
| 2 | 9.4% | 2.7% | 17.4% | 8.8% | |
| 3 | 4.1% | 0.4% | 9.8% | 2.8% | |
| 4 | 4.0% | 2.5% | 6.4% | 3.2% | |
| 5 | 2.7% | 2.3% | 3.2% | 2.6% | |
| 6 | 2.5% | 2.5% | 2.9% | 2.1% | |
| 7 |  instagram.com | 1.9% | 0.1% | 4.6% | 1.4% |
| 8 | 1.9% | 4.4% | 0.8% (provisional, below the reporting threshold) | 0.5% (provisional, below the reporting threshold) | |
| 9 | 1.8% | 4.0% | 0.8% (provisional, below the reporting threshold) | 0.6% | |
| 10 | 1.6% | 1.5% | 1.6% | 1.9% | |
| 11 | 1.6% | 1.3% | 2.5% | 1.2% | |
| 12 | 1.6% | 3.7% | 0.8% | 0.3% | |
| 13 | 1.6% | 1.1% | 2.6% | 1.1% | |
| 14 | 1.6% | 0% | 4.4% | 0.6% | |
| 15 | 1.4% | 0.9% | 2.2% | 1.3% | |
| 16 | 1.1% | 1.2% | 1.3% | 0.8% | |
| 17 | 1.1% | 1.7% | 1.0% | 0.5% | |
| 18 | 1% | 2.5% | 0.2% | 0.2% | |
| 19 |  amazon.com | 0.9% | 0.9% | 1.1% | 0.7% |
| 20 | 0.9% | 1.7% | 0.7% | 0.3% |
Share is the percentage of analysed answers that cited the domain at least once, counting each answer once however many times it linked there. A domain enters the table on at least 5 citations, and only once it has been cited for several independent businesses. Cells marked † fall under that on the individual engine and are shown for contrast rather than as a rate.
How the leaders move, day by day
Seven-day rolling average across all engines. A single day is a few hundred answers per engine, so the raw line is noisier than the change it would imply.
Break it down by answer engine
Each engine has its own citation habits. These pages rank sources by that engine's own rate, not by a blended average.
ChatGPT
27,242 answers analysed, averaging 19.4 cited sources each. Cites sleepfoundation.org in 4.4% of answers.
See ChatGPT sourcesGoogle AI Overviews
24,713 answers analysed, averaging 7.9 cited sources each. Cites youtube.com in 21.6% of answers.
See Google AI Overviews sources
Google AI Mode
28,501 answers analysed, averaging 5 cited sources each. Cites youtube.com in 13.1% of answers.
See Google AI Mode sourcesTake the data with you
The whole index is free to download and free to republish with a link back. Charts are embeddable on any site, and they update themselves as the data does.
Questions about AI citation data
What the numbers mean, how they were measured, and how they compare with what else has been published.
What are the most cited domains in AI search?
Across 80,614 AI answers measured over the last 51 days, the five most cited domains are youtube.com, reddit.com, facebook.com, linkedin.com, play.google.com. youtube.com leads, appearing in 11.5% of all analysed answers. The full ranking is the table above, and it is rebuilt every day.
Do ChatGPT and Google cite the same sources?
No, and the gap is the most striking thing in this data. The engines agree on almost nothing at the top of the ranking. Google's AI Overviews and AI Mode lean heavily on user-generated video and forums, while ChatGPT reaches for vendor documentation and technology press. A single blended list of "sources AI cites" averages that difference away and tells you very little.
Does ChatGPT cite Reddit?
Rarely, on the commercial questions in this index. Reddit appears in 2.7% of ChatGPT answers here, against 17.4% of Google AI Overviews answers. Reddit is one of the most cited domains on the web overall, but that reputation comes mostly from Google's AI surfaces rather than from ChatGPT.
Why is Wikipedia not near the top of this ranking?
Because these are buying-intent prompts, not general-knowledge ones. Published studies have put Wikipedia's share of ChatGPT citations anywhere between roughly 8% and 48%; in this index it appears in 0.7% of ChatGPT answers and ranks 81 overall. Wikipedia is what an engine reaches for when asked what something is. It is not what it reaches for when asked which product to buy.
How does an AI engine choose which sources to cite?
Each engine runs its own retrieval step before it writes an answer, then attributes the pages it drew on. What that step favours differs sharply by engine and by the kind of question asked, which is why the same domain can hold a quarter of one engine's citations and almost none of another's. This index measures the outcome rather than the mechanism: what actually got cited, how often, on which surface.
How is citation share calculated?
Share is the percentage of analysed answers that cited a domain at least once. An answer linking to the same source four times counts once, so a verbose citation block cannot inflate a domain's standing, and answers that cited nothing at all stay in the denominator. A domain has to clear at least 5 citations across several independent businesses before it is published.
How often is this data updated?
Every day. Each engine is queried daily and the whole index is rebuilt from scratch, so the tables reflect the most recent measurement rather than a study run once and left to age. The current window covers 80,614 answers.
Can I use this data in my own work?
Yes. The full dataset is available as JSON and CSV, free to reuse and republish with a link back, and every source page carries an embeddable chart that updates itself.
How these numbers are measured
Every figure on this page has a denominator you can check. Here is exactly what was counted, and what this dataset is not.
What one measurement is
RankSpot runs its customers’ tracked prompts against each answer engine every day and records the sources each answer cites. One measurement is one answer from one engine on one day: 80,614 of them, Sep 1, 2026 to Oct 21, 2026.
What the percentage means
Share is the proportion of analysed answers that cited a domain at least once. An answer linking to a source four times counts once, so a single verbose citation block cannot inflate a domain’s standing. Answers that cited nothing at all stay in the denominator.
Which domains appear
A domain has to be cited at least 5 times, and cited for several independent businesses rather than only one, before it is published. Subdomains fold into their registrable domain, so mobile and desktop hosts count as one source. App-store listings stay separate because being cited there is a different fact. Google.com itself is left out: Google’s AI surfaces link back into Google as part of how they render an answer, which describes the interface rather than which publishers earn citations. YouTube stays, because it is cited as a source in its own right.
Where the prompts come from
These are real commercial prompts written by real businesses tracking their own visibility, not synthetic queries invented to fill a leaderboard. That is the strength of the dataset and also its shape: it skews toward buying-intent questions in the categories RankSpot customers sell into, across many businesses and several languages. It is not a random sample of everything people ask.
How often it updates
Every engine is queried daily and the whole index is rebuilt from scratch each day, so the tables reflect the most recent measurement rather than a snapshot taken once and left. Daily figures on a single engine move on small numbers, so the charts switch to a seven-day rate once there is enough history to compute one honestly.
Updated daily. Last refreshed October 21, 2026. The full dataset behind every table is available as JSON and CSV, free to reuse with attribution.
Millions of people will ask AI about your category this week
RankSpot researches, writes and publishes daily, and sends you the short list of what's left. Free for 3 days.
Start free trial