When a patient asks an AI assistant who to see, it answers with a short list of names, usually four to six in my tests. In the local-business tests I ran, those names tracked what reviews, forums, local blogs, and local press said about a business, more than who advertised the most or how large the company was.
Think about the last time a friend asked you to recommend a doctor. Did you say “a good one near you”? Probably not. You said something like: “My daughter has had a cough for three weeks and our pediatrician is booked. Who do you trust, and do they take Aetna?”
Now think about the last thing you asked ChatGPT. Notice how long it was. It was almost certainly longer than anything you would type into a search box. People ask assistants the way they ask friends: the whole situation, in full sentences.
Some patients have started doing exactly this, and a practice that is not on the assistant’s short list is, for that patient, invisible. I run a company that studies how these assistants choose, so I tested it. This piece covers what I found, where it stops, and what a practice can check for itself in an afternoon.
Real physician voices, twice a week
Free, and one click to unsubscribe.
Where I am coming from
I am not a clinician, and I want to be clear about what I did and did not test. I got here through a different industry. I started selling feeder insects online, in a market that was decades old, full of established brands with strong websites. Winning the short, obvious Google searches looked unrealistic for a newcomer. So I went to the AI assistants instead, read what the platforms publish about how they retrieve and cite sources, and formed a working theory: An assistant looks for the source that actually knows the answer to the specific question, then names it. If that is true, the opening for a smaller business is the long, specific questions customers ask before they buy, which most competitors answer poorly or not at all.
That store grew from 915 pages on Google to 4,089 in about a month. I did not run a controlled test of how often assistants named it before and after, so I will not say the pages caused the mentions. But it led me to a more direct question that applies to every local business, medical practices included: When a person asks an assistant who to go to, who does it name, and why?
What I tested, and what I did not
I asked ChatGPT and Google’s AI Mode the questions a customer would type, signed out in a private window, and asked Claude the same questions from my own account. I did this for a town in Florida and for a larger city in Virginia, for plumbers, coffee shops, pizza places, air conditioning repair, and oil changes.
I did not test physicians. I did not test a single medical question, specialty, or outcome. I chose trades where trust decides the choice, because those are the closest stand-ins I could test without touching health advice. Whether the same mechanism applies to medicine is my inference. Health queries may well get extra caution from the assistants, and I do not know.
Two places, one day. Treat it as an illustration, not a study.
What the assistants did
In the Florida town, a plumber with 19 reviews was named in the answer. Companies with roughly 4,600 and 10,400 reviews, including national franchises, appeared only in the paid ads beneath it.
In the Virginia city, plumber answers led with local companies, and two national names, Roto-Rooter and Service Experts, showed up only as sponsored ads. For coffee, pizza, and air conditioning repair, no national chain appeared in any answer. Across five coffee answers there were twenty-seven picks and none was a national chain.
With oil changes, chains did lead. But the assistants chose them one location at a time, based on that location’s reviews, and always next to independent shops. One assistant told its user to skip a specific location over a single bad review.
Different assistants gave different answers to the same question. The five coffee answers named four different number ones. There is no single ranking to optimize for.
What they leaned on was almost entirely outside the businesses: review sites, community forums, local blogs, niche guides, local polls, and local press. When asked about a company’s own description of itself, one assistant said, without prompting, that it is a lead and not an endorsement.
The pattern: Size and advertising spend got a business the slot next to the answer. What got it into the answer was what other people said about it, for that exact job.
Why the assistants pick who they pick
I cannot see inside these systems, so this is a working theory built from what the platforms publish about how they retrieve sources, and from what I saw in the answers.
An assistant asked a long question does not hand over a list of every provider. It looks for sources that appear to know the answer to that exact question, then names whoever those sources point to. The closer a page or a review matches the specific need, the more likely it gets used. And a source that other people back up counts for more than one that only describes itself.
That is consistent with the 19-review plumber. I do not know what he had that the larger companies did not. But the assistants kept citing review sites, forums, and local press, so a small business with a few reviews that describe the work, plus a mention in the right place, can look like a better source for one specific job than a large one with thousands of generic ratings.
It is also why a practice website alone is not enough. It is the one source you control, and the assistants treat it as a claim. What other people say is the evidence.
What this suggests for a practice
If the same mechanism applies, a practice is chosen on what independent sources say about a specific service, not on how large it is. Here is what I would check, in order.
Look first, in the patient’s words
In a private window, ask two or three assistants the questions a patient would ask, written the way patients talk:
- Who treats sciatica near me, and do they take my insurance?
- My eight-year-old snores and cannot focus at school. What kind of specialist, and who is good near here?
- Pediatric dentist near me that is gentle with anxious kids.
Save the answers. Write down who is named and which sources the assistant cites. You will learn more in ten minutes than from a month of guessing. Repeat it monthly, because the answers change, and ask your front desk what patients say when they call: New patients often mention what they asked an assistant before they picked up the phone.
Then read the answers closely, because they hand you a free lesson. Look at the exact words the assistant uses to describe each practice it names. If it says “known for treating anxious children” about another office, that phrase came from somewhere: their pages, their reviews, or an article. Find it. Those are the words patients and reviewers will need to use about you.
Keep it small and regular. Fifteen minutes a month is enough: one person, five patient-style questions, three assistants, one spreadsheet with a row per answer. In six months, you will have a record of who gets named, who dropped off, and what changed, which is more than most practices know today.
Publish one clear page per service
An assistant answers a specific question, so it looks for a specific page. A services list gives it nothing to quote. For each service, write what it is, who it is for, what to expect, how long it takes, and how insurance and payment work. Plain language, by someone who does the work. If a treatment has a trade-off, state it. Pages that read like honest explanations are easier to trust than pages that read like brochures.
Make the listings agree
Use the identical practice name, address, phone number, hours, and clinician names on your website, your Google profile, and every directory. Assistants cross-check. A mismatch, such as an old address on one listing or a clinician who left two years ago on another, gives them a reason to skip you.
Let reviews name the service
A review that says “treated my knee pain after two other clinics did not help” tells an assistant what to recommend you for. A bare five stars says very little. Invite patients to share their experience in their own words. Offer no incentive, and do not pressure anyone. Most platforms prohibit rewards for reviews, and patients should never feel their care depends on it.
Reply to reviews without confirming anyone is a patient
Privacy rules apply even when a reviewer names themselves in public. A reply that confirms the person was treated can create a problem. A general reply about how you approach care protects the patient and still shows readers how you work. Check the specifics with whoever handles your compliance, because I am not the right source for that.
Earn mentions you did not write
The assistants kept leaning on sources that had nothing to gain from the answer. For a practice, that could be a local news feature, a hospital or society affiliation, a community health event, a profile in a local magazine, or a clinician quoted in a story. One mention from a source people trust is worth more than another page you wrote about yourself.
A realistic first month
A practice does not need a marketing department for this, and the work is mostly ordinary. Week one, run the check and save the answers. Week two, fix every listing so the practice name, address, phone number, hours, and clinician names match everywhere, and ask whoever manages your website which service pages are thin or missing. Week three, write or improve the two service pages patients ask about most, in plain language, and invite recent patients to describe their experience in a review, without any incentive. Week four, run the same check again and compare it to week one. Small, steady steps beat a big push that gets dropped after a month, and the comparison tells you whether any of it moved the answers.
What I would not do
I would not buy ads hoping to land inside the answer. In my tests the ads sat beside the answer, not in it. Ads can still bring in patients, and some practices will want them, but they are a separate channel.
I would not publish a pile of thin pages to look big, either. Search platforms have policies against low-value, mass-produced pages, and an assistant has no reason to cite one that does not help a patient. What worked for me was one real question, answered well, per page.
What I did not test
I did not test any medical question, any specialty, or any patient outcome. I do not know whether assistants apply extra caution to health queries, how they handle a question about a symptom versus a question about a provider, or how this changes by region. The assistants also change often, and a result today may not hold next month.
The point of this piece is not that I know how a particular assistant ranks a particular doctor. It is narrower: In the businesses I could test, the choice rested on independent evidence about a specific job, and that is something any practice can check for itself.
Bottom line
Patients are asking assistants the way they ask friends: long, specific, with their situation in it. The assistants answer with a short list of names. In my tests, those names were not the biggest companies or the biggest advertisers. They were the ones independent sources vouched for, for the exact job. For a practice, that points to one clear page per service, reviews that describe the care in the patient’s own words, listings that match everywhere, and a few mentions from sources you did not write. Start by asking the question the way a patient would, and see who the assistant names.
Frequently asked questions
Do AI assistants take money to recommend a doctor? In the tests I ran on other local businesses, paid listings were separate ads shown next to the answer and labeled as sponsored. I did not test medical queries, and I cannot speak to every product or future change.
Is this the same as ranking on Google? No. Different assistants gave different answers to the same question, so there was no single ranking. Good local basics still matter, since the assistants appeared to lean on reviews and local sources, but there is more than one list now.
Can a small practice compete with a large health system? In my tests of other trades, a small business with 19 reviews was named while companies with thousands were not. Size alone did not decide it. I cannot say whether the same holds for medical care.
Should we ask patients for reviews? Yes, in a way that lets them describe their experience in their own words. Do not offer rewards, do not pressure patients, and do not reply in a way that confirms who is a patient.
How many pages does a practice need? You do not need hundreds. Start with one page per service that patients ask about most, each answered completely and honestly.
How often should we check what the assistants say? Monthly is a reasonable rhythm. Answers change, so write down what you see each time and compare.
Matt Goren is a technology executive.
Your KevinMD author page is one of the top results for your name. Turn it into a premium profile, publish in 1 to 3 days, and get a guaranteed podcast slot.
