What Happens to the Doctor-Patient Relationship When Both of You Have Already Asked the AI

A keynote angle for health systems and pharma events: what changes in the exam room once patients and clinicians both consult AI before the appointment, and what that means for trust.

Before my last three specialist appointments, I had already asked a language model what I thought the specialist would say. Before at least one of those appointments, I have good reason to believe the specialist had done something similar with the case notes.

Neither of us mentioned it. That silence is the actual subject of this article, and it is a more useful keynote topic for a health system or pharma audience than the usual AI-diagnosed-me-fast framing, because it describes something happening in every exam room right now rather than something that happened to one patient once.

The asymmetry nobody names out loud

When I arrive having consulted an AI, I am read one of two ways. Either I am the informed patient, which is flattering and rare, or I am the difficult patient with a printed differential, which is exhausting and common. The same behaviour, read oppositely, depending entirely on how the folder lands.

When the clinician has consulted an AI — for a differential, for a drug interaction check, for a second read on an image — that is simply called practising medicine well. Nobody asks the clinician to disclose it. Nobody frames it as a crutch.

The tool is the same. The status of the person using it determines whether the same act reads as diligence or desperation. That is not a technology story. That is a story about who gets believed, wearing a technology costume, and it is exactly the kind of finding that makes a room sit up, because everybody in a clinical audience has felt both sides of it.

What patients are actually using these tools for

Not diagnosis, mostly. Translation.

Translation of what a symptom might mean before it is dismissed as nothing. Translation of a diagnosis, after the appointment, back into language a frightened person can actually hold. Translation of what a drug interaction warning means for a life that already has four other medications in it. Translation, sometimes, of the appointment itself — the difference between what was said and what was understood is enormous, and I have gone home and asked an AI to explain what my own doctor just told me, in words simple enough that fear did not distort them.

That is not a rejection of the clinician. It is a patient buying themselves a second pass at information they were given once, badly, under stress, in a twelve-minute appointment. Framed that way, it stops being a threat to clinical authority and starts being obviously useful — and a keynote that frames it that way gets the room instead of losing it.

Where it actually goes wrong

Not where people assume. The dangerous case is not the patient who trusts the AI completely. That patient is usually easy to redirect, because they will show you the output and you can correct it together.

The dangerous case is the patient who has learned, from repeated experience, that neither source is reliable, and now trusts neither the clinician nor the tool, and makes decisions from instinct and fear instead. That patient does not show you the AI conversation. They do not mention the symptom they googled and decided was probably nothing. Silence, in a system that has trained people to expect not to be believed, is the actual risk — and it long predates any of this technology. The tool did not create the mistrust. It gave the mistrust somewhere new to go.

What this means for the room

For clinicians in the audience, the useful reframe is that the patient's AI conversation is data, not a threat. Ask what they asked it. Ask what it told them. That single question — what did you already look into — does two things at once: it surfaces the actual fear driving the visit, and it tells the patient they are allowed to have looked something up, which is the opposite of what most patients currently expect to hear.

For health system and pharma leadership, the useful reframe is about where to spend the next investment. Nearly all current effort is going into clinician-facing diagnostic tools. Very little is going into helping patients ask better questions of the tools they are already using unsupervised, at scale, tonight, on their own phones. That is the gap, and it is where the next genuinely useful patient-facing product sits.

The talk I actually give

I open with the fact that a machine found what thirty-three doctors did not, because it is true and because it gets the room's attention. Then I spend the other forty minutes taking that fact apart, because the interesting material is underneath it: what it says about whose word gets trusted, what it changes about the balance of power in the room, and what it means that the tool is now a third party in a relationship that used to have two.

The talks that only celebrate the technology or only warn about it are both, by this point, doing something the audience has heard before. The talk that describes what is actually happening between two people in a room, with a third party neither of them is naming, is the one clinical and pharma audiences tell me afterwards they have not heard put that way.

If you are booking an AI in healthcare keynote and want the version built around the encounter rather than the capability, tell me about your event →