ChatGPT and Google Trust Completely Different Sources
Send the same patient to two consultants and you can get two different reads. Each one is working from a different subset of the chart.
One weighs the imaging heavily. The other leans on the labs and the history, and both of them are being straight with you.
They’re just drawing conclusions from different evidence, and the patient walks away with two opinions that both sound authoritative.
Something similar happens when a patient asks ChatGPT and Google’s AI Overviews the same question about a condition, a procedure, or a doctor nearby.
The two systems read off different shelves. One tends to pull from institutional and government-style sources, and the other leans harder on hospital and practice websites.
So a practice that shows up cleanly in one can be a blank in the other, and from the outside there’s no way to tell which situation you’re in until you go and check.
Two systems, two habits of trust
Ask an AI assistant a medical question and it has to decide, in the moment, which sources it trusts enough to summarize. Systems differ on that.
Some lean toward pages that read as institutional: professional associations, government health agencies, established medical references. Others lean toward pages that read as clinical. Hospital systems, practice websites, physician bios written in the first person.
Neither habit is a bug.
Each system built its trust model its own way, the same way two consultants trained in different programs will each default to the evidence they were taught to weigh first.
And the practical effect is exactly what you’d expect from two consultants.
Ask the same question twice, get two different answers built from two different reference sets, and a practice that only ever worked on one of those reference sets is going to be silent in the other conversation entirely.
Where specialists actually get mentioned
This surprises many practices when they finally look. For specialists in particular, a large share of what an AI assistant says about a doctor comes from third-party profiles.
Healthgrades, Zocdoc, WebMD, insurer directories, hospital staff pages. Very little of it comes from the doctor’s own website.
Those profiles are what an assistant finds when it goes looking for who treats a given condition or whether a doctor is any good, and they carry weight all on their own, independent of anything sitting on your site.
The page belongs to somebody else. What’s on it is yours to correct.
A claimed, filled-out Healthgrades profile with the right specialty, the right credentials, and a real description reads very differently to an assistant summarizing it than an unclaimed listing sitting on autogenerated placeholder text does.
The placeholder version still gets cited. It just says nothing useful, or says something slightly off, because nobody ever went in and fixed it.
Most specialists have at least one profile like this sitting untouched somewhere. It’s worth an afternoon to find and fix, precisely because these third-party pages carry more weight with AI assistants than many practices assume.
Why overlap between systems is smaller than it looks
You’d reasonably assume that whatever gets cited by one AI assistant mostly gets cited by the others too, with the differences out at the margins.
In practice the overlap is much thinner. Ask two systems the identical question and a meaningful majority of what the first one cites is missing from what the second one cites.
So a patient using ChatGPT, a patient using Google’s AI Overview, and a patient using some third assistant can end up with three answers built from three different source pools, all describing the same condition or the same search for a doctor nearby.
No one of them is more correct than the others. They’re just different doors into the same information, and your practice has to be reachable through more than one door.
What this means for a practice, concretely
One set of accurate facts, sitting in the places each system actually looks, agreeing with itself everywhere it appears. That’s the whole response. You don’t need a separate strategy per assistant.
- Your own site. Clear, specific pages for what you treat and what you offer, written as plain text an assistant can read and quote. Design elements and PDFs hide that text. A system leaning on practice and hospital websites draws straight from this material.
- Directory and association profiles. Healthgrades, Zocdoc, your specialty association’s member directory, your hospital’s physician finder. Claim what’s unclaimed, correct what’s wrong, keep the details current. A system leaning on third-party and institutional sources draws from this pile instead.
- Consistency across all of it. The same name, the same credentials, the same insurance and hours, everywhere your information appears. An assistant that finds conflicting facts across sources has no good way to work out which one is current, and the visible cost of that conflict is a patient calling about hours you no longer keep.
You never have to guess which AI system a given patient happens to be using.
Treat your digital footprint the way you’d treat a chart that gets pulled by more than one department. Accurate and current everywhere it exists, including the pages you rarely open.
The next time you have a cancellation
Search your own name and your practice name in two different AI assistants, and read past the first sentence.
Notice where each answer seems to be pulling from. Does it read like it came off your own website, or off a directory profile you’ve never once logged into?
Any listing that turns up and that you haven’t touched in a year or more, go claim it and correct it.
Questions practices ask about this
How do I check what ChatGPT and Google are actually saying about my practice?
Start with the questions a patient would ask rather than your own name. Try who treats [condition] near [your town] in two or three assistants, signed out of your accounts, and see whether you come up at all. Then search your name and check which sources each answer links to. Repeat it next month, since answers shift.
Can my office manager claim these directory profiles, or do I need a specialist?
Your office manager can handle most of it. Claiming a Healthgrades or Zocdoc profile mainly takes verification, which usually means a phone call, a fax, or a code sent to your practice line. Bringing in a specialist makes more sense when you've got several providers, more than one location, or old addresses still attached to your name from previous jobs.
How long after I fix a profile before AI answers change?
There's no reliable timeline, and it varies. The directory page itself refreshes quickly, but assistants answer from crawled and cached copies, so the corrected version has to be picked up again before it can show up in an answer. Fix the profile, keep it accurate, and check it again in a month or two rather than watching it daily.
What happens if I just leave an outdated profile alone?
The wrong details keep circulating. A stale listing can send a patient to a number that now rings at a group you left, or drop you out of a who-treats-this answer because the profile still shows a specialty you no longer practice. Correcting your own website does not reach it, since the fix has to happen on the profile.
Photograph: Lucas Oliveira / Pexels