AI for a medical practice, explained for a clinician
Updated 2026-08-23 · Atiba de Souza
What actually changed in how patients find you, what an AI assistant can and cannot see about your practice, which of the popular advice is measurably wrong, and the one boundary a practice must not cross.
Most of what a physician gets told about AI is either vague enthusiasm or vague alarm. The part that matters to a practice is narrow, concrete, and mostly about who is doing the answering when a patient goes looking.
What actually changed about how patients find you?
A growing share of people never see a list of results. They ask a question and read a paragraph.
That paragraph is assembled on the spot from sources the system can read and has reason to trust. Nobody scrolls it, nobody compares ten options, and the practices named in it get considered while everyone else is simply absent — not ranked lower, absent. There is no second page to be on.
The practical consequence: being findable and being cited are now different problems. A practice can rank respectably in a traditional search and never appear in an assistant's answer, because the assistant is drawing on material that answers the question directly and can be corroborated somewhere else.
What can an AI assistant actually see about your practice?
Only what is written down, publicly readable, and consistent with what other sources say.
That is a shorter list than most practices assume. It can see your site if the text is really there in the page rather than assembled by a script after the fact. It can see directory entries, professional profiles, news mentions, transcripts, and anything anyone else has written about you. It cannot see your outcomes, your reputation among colleagues, the quality of your consultations, or anything a patient told a friend.
Two things follow, and they surprise people:
- Consistency matters more than volume. A practice described three different ways across its site, its listings and its profiles gives a system nothing solid to say, so it says nothing. Making those agree is unglamorous and it is often the highest-value hour available.
- What others say about you counts more than what you say. Your own site asserting you are excellent carries structurally less weight than a third party mentioning you. This is why coverage, interviews and genuine mentions outperform another page of self-description.
Which popular AI advice is measurably wrong?
Several things sold hard right now do not survive measurement.
- Special files that "tell AI about your site." A study of 137,000 domains found that of those publishing one, 97% were never fetched at all (Ahrefs analysis, reported by Search Engine Journal). It is not harmful; it is simply not the lever.
- Adding structured markup for a quick citation boost. Measured across 1,885 pages against a control group of about 4,000 over 30 days, the effect on AI citation was within noise (same analysis). Markup remains worth having for other reasons. It is not a shortcut.
- Writing your own "best in the area" lists. Tracked across 9,886 answers from 34 such lists, the articles themselves were cited while competitors named inside them got recommended instead (same analysis). Writing your own rankings mostly promotes the people you listed.
Those three findings come from one body of analysis, published by a search-data company that also sells a visibility product — strong evidence, not settled fact. The uncomfortable finding underneath all three: the things that actually correlate with being named are ordinary and slow — being genuinely findable, being described consistently, and being mentioned by other people.
Does traditional search still matter?
Yes, substantially, and the practices treating it as obsolete are making an expensive mistake.
The systems producing those paragraphs largely draw on the same underlying index. Analysis of 1.4 million ChatGPT prompts found roughly 88% of citations came from the general search index rather than from specialist sources (same analysis). Being unfindable in traditional search remains the most reliable way to be absent from an AI answer.
What has changed is that ranking in the top handful of results no longer guarantees being cited — the overlap has fallen sharply year over year. Breadth of genuinely useful coverage now matters more than the rank of any single page. That is a strategy change, not an abandonment.
What is the boundary a practice must not cross?
Never put patient information into a general AI tool, and never let a machine speak clinically in your name.
Both are simple and both are violated constantly. Consumer AI tools are not appropriate places for identifiable patient information, whatever the interface implies. And material published under a physician's name carries that physician's professional exposure — generated text that drifts into asserting what something does, or how well it works, is a liability that arrives quietly and with confidence.
The workable rule: AI is excellent at helping you say what you already know, and dangerous the moment it is asked to know things on your behalf.
What is AI actually good for inside a practice?
Turning what is already in your head into material, faster than writing it.
The strongest use by a distance is extraction. Talk through an explanation you give patients weekly; have it transcribed; have it shaped into something readable; check it carefully; publish it. You have produced something genuinely yours in a fraction of the time, and it sounds like you because it was you.
That is a real productivity change and it is available now. It also happens to produce exactly the material that gets cited, because it answers a real question in the words a real person uses — which is the whole of why clinical excellence does not transfer online solved from the other end.
What should a practice actually do about this?
Make yourself readable, make yourself consistent, and get mentioned by other people. That is the entire list.
Everything else being sold under this heading is either a version of those three or is not supported by evidence. For where this sits in the wider work, see what authority-building for physicians actually is, and for the top of the guide, how a physician builds authority outside the exam room.