Most Cited Health Sources in AI Search: 2026
Signal Hill Strategies regrouped 824,997 US health citations from five AI platforms, collected from mid-April to mid-July 2026, into the organizations behind them, then checked the results against a live November 2026 tracker and a Northwestern University study of Google AI Overviews. As of the fourth quarter of 2026, the data shows that AI health answers lean on a few government research sites and large platforms far more than on hospitals.
Key Findings
|
A few years ago, a patient who searched for a new prescription or a stubborn rash got a list of links and decided which one to trust. AI assistants now make that choice first. They read the web, write a short answer, and attach a handful of citations. Those citations decide which organizations get the credit, and the visits, for health information.
Our analysts started from LLM Pulse's study of US health citations on five AI platforms and combined domains that belong to the same organization, such as NIH's three research databases and Cleveland Clinic's two content sites. We then compared platform habits using LLM Pulse's live tracker and set Google's AI Overviews against its regular results using a Northwestern University preprint.
In the following sections, we compare the organizations AI platforms cite most for health questions, the types of sources behind those citations, how the five platforms differ, and how Google's AI Overviews choose sources compared with its own top results.
NIH Is the Most Cited Health Source in AI Answers
The same organizations show up again and again in AI health answers. Some are medical institutions, and others are video, social, and app platforms that rarely appear on a doctor's reading list.
AI Health Citations by Organization (April to July 2026)
Rank | Organization | Type | Citations | Share |
1 | National Institutes of Health (NIH) | Government and research | 41,541 | 5.04% |
2 | Google (Search and Play Store) | Platform | 30,560 | 3.70% |
3 | YouTube | Social and video | 26,992 | 3.27% |
4 | Social and video | 11,119 | 1.35% | |
5 | Meta (Facebook and Instagram) | Social and video | 7,815 | 0.95% |
6 | ScienceDirect | Academic journals | 6,841 | 0.83% |
7 | Healthline | Health publisher | 6,456 | 0.78% |
8 | GoodRx | Commercial health | 4,337 | 0.53% |
9 | Cleveland Clinic | Health system | 4,169 | 0.51% |
10 | Apple App Store | Platform | 4,168 | 0.51% |
11 | Medical News Today | Health publisher | 3,490 | 0.42% |
12 | Mayo Clinic | Health system | 3,475 | 0.42% |
13 | Social and video | 3,283 | 0.40% | |
14 | US Food and Drug Administration | Government and research | 2,524 | 0.31% |
15 | WebMD | Health publisher | 2,321 | 0.28% |
Key Findings
NIH's three research databases, PubMed Central, PubMed, and NCBI, together drew about 1.5 times as many citations as YouTube. NIH also received more than five times the citations of Mayo Clinic and Cleveland Clinic combined.
The 25 most-cited domains accounted for 20.32% of all citations. The other four in five citations were spread across thousands of smaller sites.
Health systems compete for AI credit with app stores and pharmacy price sites, not only with other hospitals. Cleveland Clinic and Mayo Clinic each held about half a percent of citations, behind GoodRx and level with the Apple App Store.
Commercial and Platform Sites Receive Three in Four AI Health Citations
Sorting citations by the kind of site behind them tells a different story than a top 15 list. The categories here are LLM Pulse's own classification of each cited domain.
AI Health Citations by Source Type (April to July 2026)
Source type | Share of citations | Average position |
Commercial and platform | 75.4% | 7.25 |
Academic and journals | 8.3% | 6.56 |
UGC and social | 6.2% | 8.53 |
Medical publishers | 4.6% | 5.71 |
Government and institutional | 3.4% | 5.51 |
News media | 1.2% | 6.83 |
Hospitals and providers | 1.0% | 7.38 |
Key Findings
Government and institutional sites earned a small share of citations but held the best average position, 5.51, so AI platforms tend to list them near the top of their sources. User-generated and social sites sat furthest down, at 8.53.
Hospitals and providers received the smallest share of any source type, below news media. Academic journals were the largest authoritative group.
Most links attached to an AI health answer lead patients to platforms and company sites, not medical institutions. That held on every platform in the study, where the authoritative share stayed under one in four.
ChatGPT Cites Medical Authorities More Often Than Google's AI Tools
Each AI platform builds answers from its own search index and ranking rules. Those differences show up clearly in the health sources each one leans on.
Health Citation Mix by AI Platform (April to July 2026)
Platform | Citations | Authoritative medical | Social and UGC | Commercial and platform |
ChatGPT | 157,708 | 24.1% | 1.9% | 72.9% |
Perplexity | 43,334 | 18.4% | 3.3% | 77.7% |
Gemini | 50,595 | 16.8% | 1.2% | 80.9% |
Google AI Overviews | 249,784 | 15.8% | 8.5% | 74.4% |
Google AI Mode | 323,576 | 15.1% | 7.6% | 76.1% |
Key Findings
Google AI Overviews gave social and user-generated sites more than seven times Gemini's share. In LLM Pulse's live tracker, YouTube's presence in AI Overview health answers was the highest share any domain held on any platform.
ChatGPT's most-cited health domains in the November tracker were nih.gov (31.83% of answers), fda.gov (13.19%), and doi.org (10.95%), all research or government sources. Perplexity's list put linkedin.com (13.63%) and reddit.com (13.13%) right behind nih.gov.
The same health question can return very different sources depending on the tool. Google AI Mode produced the most health citations in the study, so its source choices reach the largest share of these answers.
Google's AI Overviews Cite WebMD and Healthline More Than Its Top Results Do
AI Overviews sit above Google's regular links on the same page, which makes a direct comparison possible. A Northwestern University team ran thousands of medical questions through Google and recorded both sets of sources.
Share of Sources by Domain, AI Overviews vs Traditional Top 5 Results (September 2026)
Domain | AI Overview sources | Traditional top 5 | Difference (points) |
nih.gov | 12.5% | 12.1% | +0.4 |
clevelandclinic.org | 4.7% | 8.0% | -3.3 |
mayoclinic.org | 3.9% | 5.8% | -1.9 |
cdc.gov | 2.9% | 3.4% | -0.5 |
webmd.com | 2.0% | 1.4% | +0.6 |
healthline.com | 1.8% | 1.3% | +0.5 |
nhs.uk | 1.5% | 3.1% | -1.5 |
ubiehealth.com | 1.4% | 0.9% | +0.5 |
harvard.edu | 1.0% | 1.2% | -0.2 |
medlineplus.gov | 0.8% | 0.9% | -0.1 |
hopkinsmedicine.org | 0.7% | 1.2% | -0.5 |
healthdirect.gov.au | 0.7% | 1.3% | -0.5 |
Key Findings
Cleveland Clinic lost the most ground, from 8.0% of traditional top 5 results to 4.7% of AI Overview sources. Mayo Clinic and the UK's NHS also had smaller shares inside the summaries than in the links below them.
WebMD, Healthline, and Ubie Health each gained about half a point in AI Overviews. NIH held a near-identical share in both, the steadiest position of any domain in the table.
A top five ranking no longer guarantees a place in the summary above it. The same study estimated that 56.7% of cited statements were supported by the cited page, after correcting for classifier error, so being cited does not always mean being quoted accurately.
For patients, the sources behind an AI health answer depend heavily on which tool they use. A question asked in ChatGPT tends to come back with government and research citations, while the same question in Google's AI features is more likely to point to YouTube, Reddit, or a company site.
For hospitals and health publishers, AI citations are spread thin. Even NIH held about one in twenty citations, and the largest health systems held half a percent or less each, well behind the big platforms.
The broader trend is that AI tools are becoming a second front page for health information, each with its own rules for whom to credit. Those rules already differ from Google's long-standing ranking of the same pages.
So if you want to know which health sources AI search trusts today, start with NIH and the large platforms, then check the specific tool, because the list changes from one AI product to the next.
Sources
Signal Hill Strategies analysis of US health citation data from LLM Pulse and Northwestern University, conducted November 7, 2026
LLM Pulse, Top US Health Sources cited by AI (updated November 6, 2026)
LLM Pulse, Top US Health Sources cited by ChatGPT (updated November 6, 2026)
LLM Pulse, Top US Health Sources cited by Perplexity (updated November 6, 2026)
LLM Pulse, Top US Health Sources cited by Google AI Overviews (updated November 6, 2026)
Methodology
We combined domains belonging to the same organization within LLM Pulse's list of the 25 most-cited domains: PubMed Central, PubMed, and NCBI for NIH; google.com and play.google.com for Google; facebook.com and instagram.com for Meta; and my.clevelandclinic.org and health.clevelandclinic.org for Cleveland Clinic. Organization types in the first table are our own classification.
LLM Pulse data covers 824,997 third-party citations in answers to generic, non-branded US health questions on ChatGPT, Perplexity, Gemini, Google AI Overviews, and Google AI Mode, collected from mid-April to mid-July 2026. Pages owned by brands that LLM Pulse tracks were excluded. Source types and the authoritative medical grouping are LLM Pulse's.
Tracker shares are the percentage of US health answers in a rolling window that cited at least one link on each domain, as published on November 6, 2026. They are not comparable with the citation shares in the first three tables.
The Northwestern University preprint, which has not been peer reviewed, submitted 3,611 HealthBench questions to Google with a US English locale from September 16 to 18, 2026, and 3,481 returned an AI Overview. Domain shares are of 23,657 AI Overview sources and 14,443 traditional top 5 results.
Derived figures, such as organization totals, shares, ratios, and percentage-point differences, were calculated from these data.
Comments