By writing the kind of content Claude already trusts: recent, specific, and backed by a source Claude considers independent, such as a government agency, a peer-reviewed database, or an outlet with no commercial stake in the claim. Claude is pickier than ChatGPT or Gemini about when it cites anything at all. Muck Rack's "What Is AI Reading?" study, published in May 2026 after analyzing more than 25 million links cited across ChatGPT, Claude, and Gemini, found Claude includes a citation in only 55% of its responses. ChatGPT cites in 96% of responses, and Gemini in 82%. But when Claude does decide a claim needs a source, it commits harder than either competitor, averaging 13 citations per cited response, more than double ChatGPT's average of five.
That gap between how often Claude cites and how deeply it cites once it starts is the whole story for anyone trying to be visible on it. This article covers what the data shows about what Claude actually links to, and what that implies for a company trying to show up there.
What makes Claude different from ChatGPT and Gemini when it cites?
Claude leans on its training knowledge for routine questions and reaches for live sources mainly when a question needs a number, a date, or a fact that could have changed since training. That selectivity is visible in the Muck Rack data: Claude's top cited domain overall is PubMed Central, the free database of peer-reviewed biomedical research, while ChatGPT's top domain is Wikipedia and Gemini's is Reddit. Wikipedia still ranks in Claude's top three cited domains in 8 of the study's 17 industries, but that is well behind ChatGPT, where Wikipedia appears in the top three for 12 of 17 industries.
Anthropic's own account of this behavior, described in Muck Rack's companion analysis of how Claude cites media, notes that Claude's top 100 most cited outlets have roughly half the average monthly unique visitors of ChatGPT's top 100. Claude is not chasing the biggest sites. It is chasing the source that most directly answers the question, regardless of how much general traffic that source gets.
Which sites does Claude actually cite?
In the Muck Rack dataset, Claude's five most-cited journalism outlets were U.S. News and World Report, Yahoo Finance, Bankrate, Consumer Reports, and The Motley Fool. Every one of those is a personal finance or consumer advice publication, and the study notes that no general news outlet appears in Claude's top five at all. ChatGPT's top five spans news, personal finance, and product journalism, led by Axios, and Gemini's range is the broadest of the three, from business and health coverage to travel and newsletters.
That pattern holds beyond finance. In the study's healthcare category, Claude's top cited domain is PubMed Central at 22.3% of citations, anchored by the same academic sourcing that defines its overall profile. Claude also appears in the rankings covered in this breakdown of which sources AI models cite most, where the pattern of favoring primary research and government data over general web content repeats across subject areas.
For a company trying to be cited by Claude, the implication is specific rather than general. Publishing a well-written blog post is not what gets picked up. Getting the underlying data in front of a publication Claude already trusts for that subject, whether that is a consumer finance outlet, a trade publication, or a peer-reviewed source, is what gets picked up.
Why does Claude ignore video and maps almost entirely?
Across the entire 25-million-link sample, Claude returned zero citations to YouTube. Gemini cited YouTube for about 2.1% of its total citations and ChatGPT for about 2.0%, but Claude's figure was flat zero. Claude does cite other video platforms, including TikTok and Vimeo, but at nowhere near the same volume.
The same pattern shows up with local search. For "best of" queries, such as best hotels or best restaurants in a city, ChatGPT systematically pulls in Google Maps results as citations, at a rate Muck Rack measured at 188 Maps citations per 1,000 such queries, averaging seven map pins per response. Claude, across the same query set, cited Google Maps exactly once.
The practical reading is that Claude treats a map listing or a video as too thin to count as a source on its own. If your business depends on local discovery or video content to drive AI visibility, Claude is the one model in this study that is very unlikely to be the channel. How ChatGPT handles the same local and video-heavy queries is worth reading as the direct contrast.
How recent does content need to be to show up in a Claude answer?
Recent, but not urgent. Muck Rack's companion analysis found Claude's citation "sweet spot" is content published within the past 10 weeks, and that Claude is roughly three times more likely than ChatGPT to cite an article that is two to four weeks old. ChatGPT's citation rate, by contrast, drops off sharply after about one week, favoring content that is fresher but has a much shorter shelf life.
Across the full study, among journalism citations with a known publish date, 57% were published within the past 12 months, a slight increase from the 2025 edition of the same research, where the figure was closer to 50%. The remaining citations spread across three or more years, with a long tail of older articles still picked up occasionally. The practical window for new content to matter to Claude is measured in weeks, not days, which gives a company more time to get a fact, a study, or an announcement in front of the publications Claude already favors.
Source: Muck Rack's "What Is AI Reading?" study, May 2026, linked in the article.
Does paid content or a press release ever work?
Rarely, and the study's numbers explain why. Across all three AI providers combined, paid and advertorial content accounted for just 0.3% of the 25 million citations analyzed, and press releases made up 1.1%, down from 6% in Muck Rack's December 2025 edition of the same study. About 99% of everything the three models cited came from non-paid sources: journalism, third-party coverage, academic research, or government and nonprofit sites.
Whether a press release is worth the spend is a separate question from AI visibility, but the citation data answers the AI-visibility half of it clearly. A release distributed on a wire and never covered by an independent outlet has almost no chance of being the thing Claude, or any of the three models studied, cites back to a user. Earned coverage is already the dominant source behind AI citations generally, and Claude's academic and government-leaning profile makes that even more true for it specifically than for ChatGPT or Gemini.
What should a company actually do differently for Claude?
Three things follow directly from the data. First, aim the pitch at outlets Claude already trusts for your category, which for most consumer and B2B subjects means finance, trade, and research-adjacent publications rather than general news. Second, keep the underlying facts, numbers, and methodology genuinely verifiable, since Claude's heaviest reliance is on PubMed Central and government sources, both of which exist specifically to be checkable. Third, treat the two-to-ten-week window after publication as the period when Claude is most likely to pick a piece up, rather than expecting a same-day spike the way a company might plan for ChatGPT.
None of that guarantees a citation. Claude cites selectively by design, and the same Muck Rack data shows no single AI media hierarchy exists across all three models: the outlets that win with Claude are not the ones that win with ChatGPT or Gemini. Building a PR strategy around what actually gets covered, rather than around any one AI platform, is still the more durable approach, since the earned coverage that wins with a human editor is the same coverage Claude is reading nine weeks later.
The direct answer, for anyone who read this far
Claude cites far less often than ChatGPT or Gemini, 55% of responses against 96% and 82%, but when it does, it goes deeper, averaging 13 sources and leaning on PubMed Central, government data, and a small set of consumer finance and trade outlets rather than general news, video, or maps. Getting cited by Claude means getting covered by the specific, checkable sources it already trusts, within a few weeks of publication, not writing more content and hoping it gets noticed.
If you want help building the kind of earned coverage that shows up across ChatGPT, Gemini, and Claude rather than just one of them, request a quotation and we will walk you through how we would approach it.
