ChatGPT’s citations are “wide but shallow”: the top 10 sites are just 12 percent of them

New data on where ChatGPT actually pulls its sources cuts against the common worry that a handful of giant sites now own every answer. Profound analysed about 730,000 US English ChatGPT.com conversations from October to December 2025 and found the citation landscape is, in its words, wide but shallow: Wikipedia is the single most-cited domain at 5 percent of all citations, Reddit is next at 3 percent, and the top 10 domains together account for just 12 percent. The long tail, not a few dominant sites, is where most citations live.

What the numbers actually say

• Wikipedia leads at 5 percent of all citations and appears in 18 percent of the conversations that carry any citation.
• Reddit follows at 3 percent of citations and 13 percent of cited conversations.
• Reuters and the NIH each take about 1 percent of citations, each showing up in roughly 4 percent of cited conversations.
• The 10 most-cited domains combined are only 12 percent of citations, so the other 88 percent is spread across a very long tail of sites.

Why wide but shallow is good news for a smaller site

If ten sites split nearly all the citations, an ordinary business page would have no way in. They do not. Because the citations are spread so thin, being quoted does not require out-ranking Wikipedia; it requires being the clearest, most quotable passage on the specific question the model is answering. A focused page that answers one real question well can be the cited source on that question, even with little domain authority, which is close to the opposite of how classic search rewards big established sites.

The other finding: the first question does most of the sourcing

Profound also found that when ChatGPT cites at all, a conversation averages about six unique sources, and that the opening question is about 2.5 times more likely to trigger a web search than a tenth follow-up. Sourcing is front-loaded: the model reaches for the web early and leans on that gathered context later, rather than searching again on every turn. For anyone trying to be cited, that means the phrasing that matters is the direct, first-question version of a query, the one a person actually opens with, not the narrower follow-ups they type afterward.

The practical takeaway matches what our AI-visibility work keeps finding: you get quoted by stating a clear, self-contained answer near the top of a page, using real headings and structured data, and letting the retrieval crawlers reach it. Our free AI search visibility auditor checks whether your page is set up to be parsed and quoted, and flags the robots.txt rules that quietly lock the answer-engine crawlers out. Our guide on how to get cited by ChatGPT and Perplexity walks the signals in detail, and our piece on why Google’s AI Overviews cut your clicks covers what answer-first search does to your traffic.