Which Sites ChatGPT Cites for Healthcare Prompts: A Buyer's Reading of the NIH Data

In Promptwatch's healthcare citation report, published October 9, 2026, NIH domains account for 5.46% of all citations ChatGPT gave to organic healthcare prompts. That is across six nih.gov subdomains, with PubMed Central at 2.38% and PubMed at 2.33%. The report notes that NIH earns more than FDA, NHS, CMS and SAMHSA combined.

If you buy GEO software for a hospital group, a clinic network, a pharmacy brand or a health publisher, this chart changes what you should ask a vendor. It does not change what you are allowed to say about health, and nothing here is medical advice. This guide is about content quality and measurement only.

What the chart shows

The window runs September 9 to October 8, 2026, and covers organic healthcare prompts in ChatGPT. Past NIH, the next entries are:

FDA at 1.76%, including the accessdata.fda.gov drug database
NHS at 1.1%
American Society of Plastic Surgeons at 0.64%, the top non-government source
American Academy of Dermatology at 0.54%
Mayo Clinic at 0.49% and Cleveland Clinic at 0.4%, the only hospitals in the chart
Boots at 0.28%, the only pharmacy
Doctify and WebMD doctor profiles at 0.23% each, the only review or directory sites

The 20 domains in the chart together hold 15.54% of all healthcare citations. By source type, government and public health sits at 7.06%, NIH and research databases at 5.46%, medical associations at 1.39%, hospitals and clinics at 0.89%, reviews and directories at 0.46%, and pharmacy at 0.28%. The report also points out that the dermatology and plastic surgery associations together (1.18%) beat Mayo Clinic and Cleveland Clinic combined.

What it does not mean

Read this as one vendor's dataset for one month in one engine. A few limits matter for a buyer.

First, it is ChatGPT only. Google AI Overviews, Perplexity and Copilot may draw on different sources, and the report says nothing about them. Second, it is aggregate organic healthcare. A dermatology practice and a pharmacy chain ask very different questions, and their prompt sets will pull from different places. Third, the 20 charted domains are a minority of all healthcare citations. A long tail of sites sits outside the chart, and your own domain is probably in it.

Fourth, this is correlation. The data shows who was cited. It does not show that any specific page change caused a citation, and no change guarantees one. A page aligned with public guidance can still go uncited.

What to do without any tool

The report's own advice is practical, and most of it is editorial work. Attributing it properly: the report suggests backing claims with peer-reviewed studies linked on PMC or PubMed, aligning pages with FDA, CDC, NHS and NICE guidance where relevant, working with medical associations, and keeping Doctify and WebMD doctor profiles complete.

A buyer-minded team can start from that list this week:

Audit your highest-traffic clinical or product explainers. Each factual claim should point to a primary source your reviewers can name.
Have a qualified reviewer sign every health page, and show the review date.
List the associations that cover your specialty. Check whether you are a member, a contributor or a quoted expert.
Open your doctor or clinic profiles on directory sites and fix anything stale: addresses, specialties, hours.

None of this needs software. It does need someone to own it, and that is usually the weak point.

What this means for your shortlist

The healthcare mix is unusual because the biggest slice belongs to public institutions that no brand can compete with. So the realistic questions for a healthcare team are narrower. Which public and association sources shape answers for our specific prompts? Where does our content show up next to them? What does the answer say about us when it does?

That points to a short checklist for any citation tool you evaluate:

Domain-level and page-level citations for your own prompt set. A public chart cannot tell you whether NIH or an association dominates your service line. Your prompts can.
Source-type grouping you can edit. If a tool lumps a hospital, a government site and a directory together, you cannot reproduce the split this report makes.
Tracking of specific pages over time. You want to know whether an updated article gets cited more, less or not at all across repeated runs.
An accuracy check on what the model says about you. For regulated brands this matters more than a visibility score.

The measurement step: Promptwatch

Promptwatch is the tool we would test first against that list. Its citation analytics show page-level and domain-level citations with URL positions, so you can see which government, association or directory pages sit beside yours in answers to your prompts. Page Tracker follows specific URLs, yours or a competitor's, which is how you test whether a revised article behaves differently. Brand Narrative, available on every plan, pulls out the claims AI makes about you, groups them by theme and scores them. On Business, Agency Scale and Enterprise, Fact Check reads your crawled pages against a claim you mark as inaccurate and quotes the pages that state it or the opposite. That is a content-quality tool, not a clinical one, and your reviewers still decide what is correct.

On cost: the free Explore plan covers 10 prompts in ChatGPT only. Essential starts at $95 a month with 50 prompts. Crawler logs, which show whether AI bots can fetch your clinical pages at all, start on Professional at $245 a month, and Essential has none. Our Promptwatch review has the full plan picture.

For related reading, see our guides to GEO tools for healthcare providers, the AI visibility tool for healthcare brands and what domain-level citation data shows.

The short version: the NIH number is a reminder that in health, source quality is the entry ticket. Fix your evidence trail first, then use Promptwatch to see whether it shows up where your patients' questions are answered.