How to measure a hospital’s visibility in AI answers

A ranking report cannot tell a hospital whether ChatGPT, Gemini, Copilot or Google’s AI Overviews name it when a patient asks where to have a knee replaced. Measuring that needs a different method, because AI answers change from one run to the next.

One check proves very little

In a study published in January 2026, SparkToro’s Rand Fishkin and Patrick O’Donnell of Gumshoe.ai had volunteers run 12 recommendation prompts a combined 2,961 times across ChatGPT, Claude and Google’s AI results (AI Overviews, or AI Mode when no Overview appeared) in November and December 2025. One of the prompts asked about cancer care hospitals. They put the chance of ChatGPT or Google’s AI returning the same list twice at under 1 in 100, while how often a brand appeared across runs was steadier than where it appeared in the list.

The study is not peer reviewed, it was run with a company that sells AI tracking tools, and its authors say open questions remain, including how many runs reliable data needs. Its practical lesson still holds for hospitals: measure how often you appear across many runs, and ignore the order of any single answer.

Build the question set from patients’ questions

• Start from the questions patients ask at each stage: understanding a condition, comparing treatments, choosing where to go, cost and insurance, travel and recovery.
• Write provider-choice questions the way patients ask them, with the place and the procedure: “Which hospitals in Surabaya do keyhole knee surgery?” Leave the hospital’s name out of these, so you measure whether the assistant brings it up unprompted.
• Run the questions in the languages your patients use, such as Bahasa Indonesia, Malay, English and Chinese.
• Add a smaller set of questions that name the hospital and ask what it offers, where it is and which doctors practise there. These check what the assistants say about it.
• Start with 30 to 50 questions across the two or three service lines that matter most, and keep the wording fixed so results compare over time.

Run each question several times, the same way

• Cover the assistants your patients use: ChatGPT, Gemini, Microsoft Copilot, Perplexity, and Google’s AI Overviews and AI Mode where they are available.
• Run each question several times per assistant in each cycle, with the same settings every time: a fresh session, the same language, and the location written into the question.
• Save the full answer text, every linked source, the date and the assistant, plus the model where the assistant shows it.
• Repeat every month, and compare quarters rather than single months, since one month’s runs can move by chance.

What to count

• Mention rate: the share of runs that name the hospital, by question, service line and assistant.
• Citation rate: the share of runs that link to one of the hospital’s own pages.
• Share of voice: the hospital’s mentions against the other providers named in the same runs.
• Cited sources: the domains the assistants link to for each question, such as directories, news sites, health portals and government registers. They show where wrong or missing information comes from.
• Accuracy: whether the answer gets the services, doctors, opening hours and location right. A wrong opening time, or a service the hospital no longer offers, is a patient safety problem as well as a marketing one.

Leave out a hospital’s position within an answer, for the reason the study gives.

Add what the platforms report

• Google Search Console includes traffic from AI Overviews and AI Mode in the Performance report, under the Web search type, without a separate breakdown.
• Bing Webmaster Tools has an AI Performance report, in public preview since 10 February 2026. It counts citations of your pages in Microsoft Copilot, AI-generated summaries in Bing and some partner services, shows which pages were cited, and gives a sample of the grounding queries that retrieved them. It does not report clicks.
• Web analytics show visits referred by assistant domains such as chatgpt.com, perplexity.ai, gemini.google.com and copilot.microsoft.com.

None of these covers every assistant, so the question runs stay at the centre of the measurement.

Act on the findings within the rules

When a hospital rarely appears for a service it offers, check first whether its pages state that service plainly: the procedure name, the specialty, and the doctors who perform it.

When an answer gets a fact wrong, look at the sources it cites. Correct the hospital’s own pages first, then the listings it controls, such as its Google Business Profile.

Don’t answer a low score with superlatives or comparison pages. They break the advertising rules in Indonesia, Malaysia and Singapore. Kevin Cho sets out what to write instead in What hospitals can publish for AI search in Indonesia, Malaysia and Singapore.