Where “47% of patients use AI to find a doctor” came from

In mid-2026 a statistic began appearing across the healthcare trade press: 47% of patients now use artificial intelligence to research providers, up from 31%. TechTarget carried it. So did MediaPost, PR Newswire and Medical Economics. From there it moved into agency pitch decks, where it is doing a great deal of work.

It should not be. Here is where it comes from.

The methodology

The figure originates in a vendor report published by a company that sells reputation-management software to healthcare practices.

The report describes respondents as “representing multiple regions, incomes, and genders”. That is a quota description. It is not a probability sample, and it does not support inference to the US patient population.

What the better-designed research found

Four months before that fieldwork, West Health and Gallup surveyed 5,660 US adults on the Gallup Panel — a probability-based panel recruited by random-digit dialling and address-based sampling, weighted to the Current Population Survey, margin of error ±2.1 percentage points.

They found that one in four US adults, 25%, had used AI tools or chatbots for physical or mental health information or advice of any kind. Not for choosing a provider. For anything.

The Australian figure, smaller again

Australia has something better than either: a peer-reviewed, nationally representative study. Ayre, Cvejic and McCaffery surveyed the Life in Australia panel in June 2024 and published in the Medical Journal of Australia in March 2025. Sample of 2,034, response rate 68.9%, propensity-score weighted to the Australian population, ethics approval from the University of Sydney.

Result: 9.9% (95% confidence interval 8.5–11.4) had used ChatGPT for health information in the preceding six months. Among those who had, the commonest tasks were learning about a condition (48%) and working out what symptoms mean (37%).

“Find a doctor or a clinic” was not among the nine response options. The nearest was “find out if I or someone I know should see a doctor”, at 12% — a question about whether to seek care, not about which clinician to see.

Why it spread

The number spread because it is useful. It gives an agency a reason for a practice to buy something, and it gives a trade publication a headline. The syndication chain carried the figure without the methodology at every step, which is how a SurveyMonkey panel becomes a fact about American patients across four publications inside a month.

This is not an accusation of dishonesty. Vendor research is a legitimate genre and the report discloses its own sample size. The failure is downstream, in the repetition.

What can honestly be said

• A quarter of US adults have used AI for health information of some kind. Well evidenced.
• About one in ten Australians had used ChatGPT for health information as at mid-2024. Peer-reviewed.
• There is no reliable data, in any country, on how many patients use AI to choose a clinician. The best-designed surveys do not ask the question.

We would rather say “no reliable data exists” than pick a number. And there is a measured finding pointing the other way in any case: through December 2025, local and provider searches showed no AI Overview presence at all. The part of health search that is about finding a named doctor is, so far, the part AI has left alone.