Physicians' Newest Competitor Never Sleeps, Never Bills, and Is Frequently Wrong

Let us begin with a number so large it needs its own parking structure: 300,000,000. That is how many health and wellness questions get routed through ChatGPT every single week, per OpenAI’s own head of health, with roughly 70% of them fired off outside clinic hours — which is to say, at the exact moment a human being is lying in bed at 1 a.m. deciding that a mole has “changed” and that this is, medically speaking, an emergency. I am not making this up. Three hundred million times a week, a chatbot is getting the first appointment. And it does not take Tuesdays off.

This is a different AI story than the one everyone has been telling. For two years the plot was AI as the thing happening to doctors — the scribe in the corner, the documentation robot, the software quietly grading your note speed. That story is real, but this week the data spun around and pointed at the other end of the exam room: the patients. AI has stopped being merely the doctor’s uninvited assistant and started being the patient’s uninvited first opinion.

The generational split is where it gets loud. Aflac’s latest survey found that 76% of Gen Z and 61% of millennials now consult AI before consulting an actual professional — and here is the kicker — fewer than half of Gen Z (48%) even have a primary care physician, versus 87% of boomers. So the youngest patients are building their entire “something is wrong” reflex around a language model, at precisely the age when they were historically supposed to be meeting a doctor and starting a thirty-year relationship. You cannot be someone’s trusted second opinion for three decades if a chatbot got there first and you were never in the running. It’s hard to win a race you didn’t know had started, at 1 a.m., in someone’s bedroom.

And by the time these patients show up — if they show up — the plot has already been written without you. The Wall Street Journal reported this week that patients are feeding their own lab and imaging results into AI even though their doctors have politely begged them not to, and are doing it anyway, because “your doctor would prefer you didn’t” has never once stopped a human with an internet connection and a worrying number. Worse, the pre-game narrative is often flatly wrong. As one physician put it this week, “medical misinformation arrives before the physician does” — repeated and reframed and algorithmically re-served until it feels like consensus — so you now walk into the room already losing an argument to a TikTok the patient trusts more than you, your degree, and the entire concept of the pancreas. This is a genuinely new and completely unbillable part of the job: professional myth-rebuttal.

The health systems paying attention are not fighting this, because you cannot fight a chatbot with a memo. They’re building around it. Cleveland Clinic Florida is standing up what it calls a “digital corridor” staffed by, and I promise this is a real job title, “inboxologists” — in-basket clinicians whose entire mission is to intercept the tsunami of patient messages before it reaches a physician’s inbox and drowns them. And the tsunami is real: portal messages are up 153% in five years, ambient AI scribes are running at 90% adoption, and these inbox specialists are already clearing 70–75% of a doctor’s pending work while the doctor is, say, sleeping. With the country staring down a shortage of nearly 90,000 primary care physicians by 2037, this is not innovation theater. It is a system triaging the one resource it cannot manufacture — physician attention — by shoving everything that doesn’t strictly require an MD somewhere, anywhere, else. (”The Inboxologists,” for the record, is an outstanding name for a band, and I will not be taking questions.)

Now the part that is actually good news, because I don’t want you to close the tab and go lie down. When AI credibly swallows the low-stakes first touch — the “is this rash normal,” the “explain my potassium,” the midnight symptom spiral — the physician’s work does not evaporate. It concentrates. It floats upstream to the stuff a chatbot structurally cannot do: complex judgment, high-acuity calls, procedures, and the single asset that just got both more valuable and harder to keep, which is trust. A world where every patient arrives pre-informed and pre-suspicious is a world that pays a premium for the one clinician who can actually take responsibility, sit in the uncertainty, and repair the relationship the internet spent all week vandalizing. The robot is happy to answer. It is notably less happy to be accountable.

The catch — because there’s always a catch — is that “being the patient’s convenient first stop” used to be a perfectly good foundation for a physician career, and that foundation is now quietly competing with something free, instant, infinitely patient, and open at 2 a.m. The demand isn’t disappearing. It’s relocating toward the high-trust, high-complexity end of the field. Physicians who plant themselves there will be fine. The ones whose whole pitch is “I am the front door” are about to meet a front door that answers in four seconds and never asks them to fill out a clipboard first.

The first appointment, in short, has already been booked, and you weren’t the one who got it. The physician’s job is increasingly everything that has to happen after it — which, on reflection, was always the hard and irreplaceable half of medicine anyway. Let the machine have the rash at midnight. You keep the part that requires a spine.

P.S. PhysEmp is an AI-powered job board for physicians and advanced practice providers — which mostly means we spend our time on the half of medicine a chatbot can’t have: matching a specific clinician to a specific role where their actual judgment is the whole point. If you’re a physician wondering where your work stays durable, or a system trying to staff for what medicine is becoming instead of what it was, that’s the thing we built for.

All sources are analyzed and curated from PhysEmp’s industry alert network. AI assists with synthesis and pattern recognition; editorial judgment stays with the PhysEmp Editorial Team. [How we make this newsletter →]