It's 2 a.m. and someone is typing a symptom into a chat window: what this lab result means, whether the lump is anything, if the new medication is safe to take with the old one. Every answer the engine writes back is grounded in a handful of cited sources, and those citations decide who gets the reader. So we studied 8 major health publishers and dissected what AI actually uses when real people ask real health questions. The results surprised us, and if you publish in health, they'll probably surprise you too. Before we get to the leaderboard, one distinction does a lot of work.
Is Being Mentioned by AI the Same as Being Cited?
No. A mention names your brand in the answer text; a citation links your page as a source. Only the citation sends the reader, the referral traffic, and the authority your business runs on. For a publisher, citations are the metric to watch. (Gist GEO tracks both, as Share of voice and Share of citations.)
A mention is exposure. A citation is traffic.
This distinction is where most “AI visibility” reporting stops short. It counts mentions and calls it a win. But a health publisher doesn't get paid when an AI paragraph name-drops it. It gets paid when the engine links its page. An answer can praise your coverage by name and still send every click to the CDC. So we measured the thing that actually pays: when a real person asks AI a health question, whose pages get cited?
The zero-click crisis makes citations the prize
And that question matters more every month, because search itself has become a conversation and the click is caught in the zero-click crisis: about 68% of US searches already end without one, and roughly 83% when an AI summary appears (SparkToro, 2026). When a summary is present, people click a traditional result just 8% of the time, versus 15% without one (Pew Research Center, 2025). The click is disappearing. The citation is what's left.
Who Does AI Cite Most for Health Questions?
Clinical institutions, by a wide margin. Across 104 health questions and more than 20,000 AI answers on three engines, Mayo Clinic (31.6%), Cleveland Clinic (18.7%), and Harvard Health (10.7%) took the top three spots for Share of citations. The biggest consumer health publishers, despite far larger content libraries, landed in the single digits: the best of them nearly five times behind first place.
How we ran the test
We took 104 real health questions, drawn from 13 verticals ranging from symptoms and conditions to nutrition and mental health, and asked them repeatedly across ChatGPT, Perplexity, and Google AI Overviews, logging which sources every answer cited. No finding in this post rests on a single run. The full methodology will be in the forthcoming deep-dive report; the short version is that nothing here is a one-off screenshot.
The leaderboard
Each figure below is the share of all AI answers in the study that cited that publisher at least once.
| Health publisher | Cited on % of AI answers |
|---|---|
| Mayo Clinic | 31.6% |
| Cleveland Clinic | 18.7% |
| Harvard Health | 10.7% |
| Leading commercial publisher (Healthline) | 6.8% |
The full 8-publisher leaderboard, plus the breakdown across all 13 health verticals will be available in the forthcoming larger report.
Strong SEO didn't carry over
The publishers with the biggest content libraries and the best Google rankings were not the ones AI cited most. That's the finding a rank tracker can't show you. I've watched brands own the top Google results and still get left out of the AI answer entirely, and this study confirms the pattern at scale: strong SEO does not automatically translate into AI citations.
Why Don't the Biggest Health Publishers Get Cited?
For health questions, AI engines cite the institutions behind the guidance (the primary sources), not the publishers who summarize it. Government and research domains (.gov, PubMed, CDC) took the largest share of all citations in the study, and community platforms like YouTube and Reddit took another sizable slice. When an engine can cite the institution that wrote the clinical guidance, the site that summarizes it has to give the engine a reason to add a second link.
Where health authority actually goes
Scan the source list on any AI health answer and it smells clinical: white coats, .gov domains, journals. For a big slice of health questions, the citation was always going to a primary source, which makes the questions where publishers can win worth knowing precisely. That's worth sitting with, because it reframes the competition.
A health publisher isn't just competing with other publishers for the citation anymore; it's competing with the CDC, PubMed, and a YouTube explainer, all in the same answer.
The engine you ask changes the answer
Citations are also a moving target: in our runs, the same question rarely returned the same sources twice, and one engine is far stingier with publisher links than the others. That tracks with where each engine pulls its sources. One major publisher, for example, is cited on 13.4% of Google AI Overviews answers but just 1.4% on one of the chat engines. One answer is an anecdote; the average is the data. That's the whole argument for measuring continuously, per engine, instead of spot-checking a blended score.
What Is the “Credit Gap” in AI Answers?
The credit gap (also called the attribution gap) is when an AI engine reads your page to build its answer but doesn't link it: value extracted, no attribution. In this study the credit gap was real, sizable, and almost entirely concentrated on a single engine. (Tracked via Share of found links: how often your pages help form an answer without making it into the citations.)
In plain terms: the engine did the reading, used the work, and skipped the byline.
Why the credit gap is recoverable
That concentration is good news, oddly enough. It means the credit gap isn't a vague, boil-the-ocean problem. It's a specific list of pages that one engine already reads and trusts but doesn't yet credit. These aren't questions where a publisher has to earn trust from zero; the engine is already pulling the page. The gap is attribution, not authority, and attribution is fixable. The exact questions, publishers, and engine involved (the recoverable-citations list) will be in the forthcoming deep-dive report — we'll link it here the moment it's live.
How Do You Measure Your AI Citations?
You measure them per engine, per topic, and over time, because AI answers shift daily: one check is a data point, and the pattern is what pays. Gist GEO is AI visibility measurement: it tracks how often AI engines mention, cite, and recommend your brand versus competitors across ChatGPT, Perplexity, Google AI Overviews, and Claude, then turns the gaps into prioritized fixes. This study was run on Gist GEO.
The nine metrics, in two groups
Gist GEO reports on nine metrics. Five response-text metrics read what the answer says: Share of voice, Share of recommendations, Placement, Average ranking in lists, and Sentiment. Four citation metrics read the sources the answer uses: Share of citations, Citation rate, Earned media score, and Share of found links. The leaderboard in this study lives in Share of citations and Citation rate; the credit gap lives in Share of found links. If you only have bandwidth to watch two, watch those citation metrics first. They're the ones tied to traffic.
Turning measurement into a to-do list
The workflow behind it is Queries, Reports, Opportunities. You define the Queries real people ask, Reports measure the nine metrics across engines and over time, and Opportunities scores the fixes by impact and effort, from content changes to earned-media outreach with named target sites. (What Is Gist GEO? walks through the full tool.) It's one piece of a larger system: Gist GEO measures your AI visibility, Gist Answers puts a cited answer engine on your own site, and Gist Ads amplifies it. Three products, one system.
