One healthcare brand Yolando tracks shows up in roughly a third of ChatGPT answers about its category. Its own website? ChatGPT has cited it almost never. That is not a broken site or a robots.txt mistake. It is what ChatGPT does.
Learning how to show up on ChatGPT means separating two things most reporting jams into one number: whether the engine names you, and whether it sources your pages. That split is the core of Generative Engine Optimization (GEO) for healthcare, and it plays out differently on each engine.
Why you need to track mention rate and citation rate separately
Two different events get flattened into one "AI visibility" score, and on ChatGPT they can point in opposite directions.
Mention rate: how often the brand is named
The share of relevant answers where the AI names your brand, with or without a link. If a patient asks ChatGPT for fertility clinics in Austin and your clinic appears in the prose, that counts. High mention rate means the model already associates you with the category.
Citation rate: how often your pages are the source
The share of answers where your own pages are cited as the source behind the claim. Not "we got mentioned," but "the engine pulled from our site and said so." A brand can be named constantly and cited almost never.
Why one number hides the problem
A blended score cannot tell a source-layer problem from a website problem. Mentioned but not cited? The engine is pulling from Reddit, directories, and roundups instead, so you work the source layer: everything written about your clinic that you do not own. Neither mentioned nor cited? The model does not associate you with the category yet, which is a problem of entity data (the basic structured facts about your organisation) and corroboration (whether anyone independent confirms them). Same dashboard number, two different projects.
What the numbers look like across four engines
Across three healthcare brands Yolando tracks (90 days to 31 August 2026), ChatGPT recorded the lowest citation share on every single brand. Last, every time.
Brand | Engine | Reach (discoverability) | Citation share |
|---|---|---|---|
Brand A | ChatGPT | 33.2% | 2.7% |
Brand A | Claude | 37.1% | 9.7% |
Brand A | 31.8% | 6.2% | |
Brand A | Perplexity | 15.2% | 1.4% |
Brand B | ChatGPT | 3.1% | 0.4% |
Brand B | Claude | 8.3% | 14.2% |
Brand B | 19.6% | 4.3% | |
Brand B | Perplexity | 18.9% | 4.7% |
Brand C | ChatGPT | 13.7% | 0.4% |
Brand C | Claude | 11.4% | 1.3% |
Brand C | 21.3% | 2.2% | |
Brand C | Perplexity | 14.5% | 1.7% |
ChatGPT is last on citation even when it leads on reach
Brand A appears in 33.2% of ChatGPT answers but is cited in only 2.7%. On Claude, the same brand sits at 37.1% reach and 9.7% citation. Blend the two and ChatGPT looks fine. Split them and the story flips.
No engine leads across all three brands
Claude led citation share on two of the three brands (9.7% and 14.2%); Google led the third (2.2%). No universal winner. The engine that matters most for your clinic is an empirical question about your data, not a headline.
The mention-to-citation gap is a diagnostic
Read the distance between the two numbers, not either one alone. A wide gap (named a lot, cited almost never, like Brand A on ChatGPT) means the engine is finding you through other people's pages. A narrow gap means your own pages are doing the work.
Look at Brand B for the sharpest version: 3.1% reach on ChatGPT against 18.9% on Perplexity, and a citation share that is 35 times higher on Claude than on ChatGPT. One brand, four completely different problems.
The four engines at a glance
Each engine treats reach and citation differently. Reach is how often the engine names brands in your category; citation yield is how often it sources brand-owned pages.
Engine | Typical reach | Typical citation yield | What wins it | What does not move it |
|---|---|---|---|---|
ChatGPT | High (up to ~33%) | Lowest of the four | Community threads, directories, roundups, structured blog content | Generic or thin pages without citable claims |
Claude | Moderate to high | Highest in our set | Corroborated entity data, few sources | Reddit volume (near-zero use) |
Consistently moderate | Brand-specific | Strong service and condition pages | Thin or duplicate location pages | |
Perplexity | Moderate | Citation-forward, brand-specific | Fresh, fact-dense sourced pages | Unsourced brand claims |
Why ChatGPT behaves this way
ChatGPT reads widely and quotes narrowly. In Yolando's analysis of 27,812 AI answers (June 2026), ChatGPT consulted roughly 32 sources per answer, against about 7 for Claude. It synthesises into prose rather than handing you a numbered citation list, which is why a brand can shape the answer without ever getting a visible link.
Only 15% of the pages ChatGPT retrieves earn a citation in the final answer, per AirOps research covering 217,508 retrieved pages, reported by Search Engine Land in April 2026. Even sources that influenced the response may never appear. Compare that to Perplexity, which searches the web on every query and attaches inline numbered citations linking each claim to its source.
A caution: nobody outside OpenAI can see exactly how ChatGPT selects and weights sources. Treat "32 sources per answer" as a measured average from our dataset, not a spec sheet.
How to actually show up on ChatGPT: work the source layer
Work both sides: your own site and the sources you do not control. Blog and content pages account for around 53% of all AI citations across engines, per BuzzStream's analysis of 4 million citations (January 2026), which makes article-shaped content the single most-cited page type there is. Most of that 53% sits on domains other than the brand's own, and that is the point rather than a caveat: strong blog content is what earns the third-party pickup, and third-party pickup is where the bulk of citations live. It is also where most healthcare brands have the widest gap.
The scale of that gap is easy to underestimate. In Yolando's own tracked citation set across 23,052 domains, 317,080 citations were earned (published by someone else) and 49,287 were social. Owned citations numbered 73. Whatever you publish, the overwhelming majority of what an engine reads about a brand was written by somebody else.
Community discussion
Reddit is the most-cited domain in Yolando's tracked set: 41,785 citations, 10.7% of all citations. In the June 2026 study it was the single most-cited domain in every healthcare vertical. You cannot buy your way into a thread. What you can do: make sure accurate information about your clinic exists where patients already discuss your category, respond as a verified brand where platforms allow it, and correct outright errors. Everything past that is monitoring, not influence.
Directory and aggregator records
Review platforms like Yelp and G2 appear often in AI recommendation queries, per a Peec AI analysis of 30 million sources (Search Engine Land, March 2026). Unlike Reddit, your directory record is yours to fix. Claim every relevant listing, make the name, address, specialty, and hours identical across all of them, and kill duplicate or stale profiles. Conflicting records teach the model to trust someone else's summary of you.
Third-party roundups and coverage
Listicles, articles, and product pages drive over half of all AI citations, per Search Engine Land's coverage of 75,000 AI answers (March 2026). You influence these the legitimate way: pitch the journalist, give the reviewer real data, be the clinic worth listing. Any vendor promising placement in editorial roundups is selling something you do not want attached to your brand.
Reference-layer entity data
Entity data is the set of basic structured facts that identify your organisation: legal and trading name, locations, specialties, clinicians. Keep them consistent everywhere: your site, your listings, and public knowledge bases. When those facts match across sources, engines treat you as a confirmed entity in the category, which lifts mention rate even where citation stays low.
How to win the other three: work your own pages
On Google, Perplexity, and Claude, own-site citation is achievable, so the investment flips from influence to construction.
Two decisions carry most of the weight. First, the depth of your service and condition pages, written to answer the question directly. See our take on condition-and-city pages that engines actually cite. Second, the split between your own site and directories, covered in directory versus own-site coverage.
How to choose where to spend
Pull your mention rate and citation rate per engine, then match your situation to one of three profiles.
Nobody has heard of you (low mention across engines): prioritise directories, community accuracy, and entity data. Building service pages nobody is sent to is premature.
Named but never cited (the Brand A pattern): tighten your listings, fix conflicting records, and earn third-party coverage so the model has your account to draw from.
Cited on one engine, invisible on another: measure per-engine before spending a dollar. Your winning engine is not guessable from our set alone. See our guide to AI visibility tracking for healthcare.
Track two scorecards, not one
Retire the single "AI visibility" number. Track mention rate and citation rate per engine, separately, because a tactic that moves one may not touch the other. A Reddit correction can lift your ChatGPT mentions while citation stays flat, and that is a win you would miss on a blended score. Two columns, four engines, on a cadence.
Yolando measures both for healthcare brands, so you can tell a source-layer problem from a website problem before you spend against the wrong one. Book an AI visibility audit and see your two scorecards side by side.
