Where AI Overviews actually appear in health search

The pitch most practices hear is that artificial intelligence is taking their patients. It is a good pitch. It is also, in the one part that matters most to a clinic, not what the measured data shows.

AI Overviews have taken over the part of health search where people ask questions. They have almost entirely left alone the part where people look for a doctor.

What the measurements show

Type of health query	AI Overview present	Measured by
Informational health queries	66.9%	WebFX · 130,070 US healthcare queries · data July 2025
Long queries, seven words or more	73.9%	WebFX
Treatment and procedure queries	100%	BrightEdge · December 2025
Symptoms and conditions	93%	BrightEdge
Queries with a place name	35.8%	WebFX · e.g. “dermatologist chattanooga”
Navigational queries	14.1%	WebFX
Local and provider searches	0%	BrightEdge · Dec 2025. Was 100% in Dec 2023
Self-harm, suicide, eating disorders, crisis, addiction	0%	BrightEdge · suppressed entirely, three years running

Read the bottom half of that table twice. Provider-finding is the least disrupted segment of health search, and it went from fully covered to not covered at all in two years. Google appears to have deliberately pulled AI Overviews out of the queries where a person is trying to reach a named clinician, and out of the queries where a wrong answer would be dangerous.

The studies disagree, and you should know why

What we do not claim

We are an answer engine optimisation practice. It would suit us to tell you that patients choose their surgeon through ChatGPT. The evidence does not support it.

The figure circulating in the trade press — that 47% of patients use AI to research providers — comes from a survey of 992 people run on SurveyMonkey by a company that sells reputation-management software. We have taken that number apart separately. Four months before it was fielded, West Health and Gallup surveyed 5,660 US adults on a probability panel and found that 25% had used AI for health information of any kind.

In Australia the best figure is peer-reviewed and smaller again. A nationally representative survey of 2,034 people, published in the Medical Journal of Australia in March 2025 from fieldwork in June 2024, found 9.9% had used ChatGPT for health information in the previous six months. Finding a doctor or a clinic was not among the nine things respondents were asked about.

So what is actually worth doing

The honest version is this. AI is consuming the education layer — what is this condition, what does this scan involve, what are the options, is this urgent. That layer sits before provider choice, and it is where a patient forms a view about who sounds like they know what they are doing.

Which brings us to the number that does justify the work, stated in full because half of it is usually left out.

Seer Interactive analysed 53 brands, 5.47 million queries and 2.43 billion organic impressions between January 2025 and February 2026. Where an AI Overview appears, being cited inside it is associated with 120% more organic clicks per impression than not being cited — 20,743 against 9,445 per million impressions.

That is the realistic objective: be the source the answer is built from, in the education layer, before the patient starts looking for a name. Not because it restores what search used to be, but because it is the best available position in what search has become.