Inc.
AI is developing healthcare, not by replacing doctors, but by making care faster, smarter, and more accessible.
America Is Running Out of Primary Care Doctors. AI Is Filling the Gap
Author: Chase Feiger
This week OpenAI released ChatGPT for Clinicians, a version of the AI company’s product designed to help doctors with tasks like documentation and research. It will be free to “verified” physicians, nurse practitioners, physician assistants and pharmacists in the U.S.
The notion of an AI Primary Care Physician may sound far-fetched to individuals accustomed to traditional doctor-patient relationships. And if you’ve read the horror stories about AI chatbots providing misleading or dangerous medical advice, it’s fair to be dubious. But experts who work at the intersection of medicine and technology are starting to view AI’s role in healthcare as something close to an inevitability.
“We know that although the ‘community standard’ for AI is augmentation of humans, there will be use cases and geographies where the choice is between no care or trusting an AI that is deemed good enough,” says John Halamka, MD, president of the Mayo Clinic Platform. He told me that “in LMICs and rural areas, we’ll see AI use without a human nearby, because that is the safest and best alternative to no care.”
Why an AI Primary Care Doctor Suddenly Feels Inevitable
To be sure, plenty of questions remain. “In the vision of an AI doctor, how do we actually bridge the gap from librarian to clinician?” asks Daniel Nadler, Ph.D., founder and CEO of OpenEvidence. “How does the tech move from just citing the literature to actually applying it to a messy, complicated human patient?
But that skepticism is starting to give way to an acceptance of – and enthusiasm for – AI’s potential role in the primary-care mix.
“I think we’ll be thinking about individual tasks and deciding which can be done safely by Al alone and which can supplement/augment the physician’s practice,” says Bob Wachter, MD, professor and chair of the Department of Medicine at the University of California, San Francisco, and the author of “The Digital Doctor.” “In primary care, for example, I’m guessing the Al will manage the cholesterol and the Wegovy, while a PCP will still be managing complex patients on multiple meds and with multiple diseases. It’s tricky to organize this and get it right, but it seems inevitable, given the impossibility of being a PCP today and the expense of off-loading such tasks to APPs [advanced practice providers].”
The Primary Care Crisis
So how did we get here? It starts with the ongoing decline of primary care in the United States.
Roughly 17 percent of American adults have no primary care physician, not because they haven’t looked for one, but because such practitioners are in short supply. In Massachusetts, one of the most medically dense states in the country, patients are routinely being told that the next available appointment is a year and a half out.
And that shortage is only going to get worse. The Health Resources and Services Administration projects a shortage of 141,160 full-time-equivalent primary care physicians by 2038, while a Commonwealth Fund analysis says rural supply is on track to meet only 68 percent of physician demand by 2037.
At the same time, the adoption of AI by physicians and patients is proceeding apace. Eighty-one percent of physicians report using it professionally and one-third of adults say they have tapped AI for health information or advice in the past year.
The trends have collided in a way that has made the notion of an AI Primary Care Physician sound less like science fiction. To be clear, it’s not happening because people are eager to replace doctors. It is happening because the current version of primary care leaves too many patients underserved.
The term “AI primary care physician” spans a wide spectrum. At one end sit tools that can assist a human clinician with tasks like drafting notes, summarizing labs and flagging care gaps. At the other end sits something closer to an autonomous clinician: A system that takes your history, asks the right questions, forms an assessment and tells you what to do next.
The Companies Building AI Assistants for Doctors
“We know AI can help reduce administrative burden, surface critical insights and streamline workflows, enabling care teams to spend more time with patients and less time on documentation,” says David Feinberg, MD, chairman of Oracle Health and Life Sciences. “But we believe this can only be done by strengthening the role of clinicians, not replacing them.”
Ambient AI has become the first clinical AI category to reach broad deployment, with systems like Houston Methodist reporting that ambient documentation tools reduced documentation time by 40 percent, increased time spent with patients by 27 percent and cut after-hours work by a third. Then there’s the experiment currently playing out in Utah, which permits AI-driven prescription renewals for consumers who request them. The program covers 192 drugs and includes strict safety and privacy protections.
Then there’s the just-announced ChatGPT for Clinicians, the goal of which is to automate necessary and time-consuming clinical tasks like documentation and medical research. Should it catch on among clinicians, it will free them to devote more of their time and energy to patient care. The platform is free for verified doctors, nurse practitioners, physician assistants and pharmacists in the U.S. OpenAI for Healthcare, launched in January, is also pushing into enterprise clinical workflows with hospital partners and HIPAA-oriented deployment.
Distribution is the name of the game in health tech, and Microsoft knows it.
Instead of burning cycles building yet another standalone “AI doctor” app that forces a workflow change, they’ve made the ultimate distribution play by partnering with Epic. By embedding agentic AI directly into the EMR, they’re automating clinical notes and patient comms right where the data lives. It’s about meeting doctors where they already are to actually solve the burnout problem at scale.
These are examples of what supplementary AI looks like at its best: An invisible infrastructure that gives physicians their lives back. It is consequential, but it is not a replacement.
The Biggest Companies Setting the Stage for AI Doctors
The more disruptive question, and the one upon which the success or failure of AI Primary Care Physicians likely hinges, is what happens when the AI starts talking directly to patients and making decisions. That is the frontier several A-list tech players are trying to inhabit.
Of course, every AI Primary Care Physician needs to be thoroughly assessed, both around what risks are involved and how it integrates into workflow. “I have observed agentic AI products working autonomously for revenue cycle and supply chain use cases—getting pre-authorization for treatment or completing a nurse call for follow-up or appointment-making,” Halamka explains. But for now, he adds, diagnosis and treatment still requires a human to be in the loop.
Amazon
Of the major players, Amazon has a clear case for building the first version patients will tolerate because it is bundling AI into an already legible care experience.
The company says that its Health AI can explain records and labs, answer symptom questions, renew medications, book appointments and connect users to One Medical providers. It also has a few things the others do not offer in a single place: a national primary care brand, a membership model, pharmacy integration and scheduling and pricing that people can understand. As an introductory offer, eligible Prime members get up to five free direct-message care consultations for common conditions. After that, Amazon charges $29 per telehealth visit, or a discounted $99 annual One Medical membership for Prime users.
Google, by contrast, looks less like the first consumer winner and more like the company most likely to supply the underlying clinical brain. Its AMIE program has moved beyond simulated physician comparisons into real-world feasibility work in ambulatory primary care, where 100 adult patients used the system for pre-visit medical history-taking under physician supervision. Google says zero safety stops were required in that study, and that its newer guardrailed version of AMIE is explicitly designed to gather history and draft a summary without giving individualized medical advice.
Google is also launching a nationwide randomized study to test conversational AI in real-world virtual care. Put more simply, Google has the most serious evidence-generation posture in the field. But its likeliest role is as the intelligence layer inside other clinical experiences, not as the branded AI family doctor patients choose first.
OpenAI
OpenAI still lacks what Amazon has: a national, owned primary care delivery network. It’s more likely to become the conversational layer people use before, between and after visits than the first fully trusted AI PCP brand.
That’s because it already sits where millions of people instinctively go when they want an answer fast. The company reports that more than 230 million people globally ask health and wellness questions in ChatGPT every week, and ChatGPT Health, released at the beginning of 2026, offers a dedicated space where users can connect medical records and wellness apps with stronger privacy controls.
Meta and Snap
Meta and Snap matter too, but mostly as interface bets rather than care-delivery bets. Meta is pushing AI glasses deeper into health-adjacent territory with prescription-ready wearables, nutrition tracking and memory-support use cases. Snap is betting that AI will move off the phone and into lightweight AR glasses, with Specs designed to bring AI assistance into three-dimensional space.
But neither company has solved the core primary care problem of clinical accountability, and Snap’s own support materials still warn users that AI can be incorrect or misleading and should be independently verified. That is a useful reminder: The future interface for AI in health may be glasses, voice or ambient assistance, but the first palatable AI PCP will still need a clinician somewhere behind the curtain.
The Problems of Scale and Access
The path for scaling in the AI healthcare category will likely begin with the boring, repetitive work that consumes time but does not require independent clinical authority: intake, history-taking, patient verification, chart synthesis, scheduling, documentation, after-visit summaries, medication management and routine follow-up.
If that sounds less dramatic than “AI doctor,” that’s exactly the point. The real transformation in primary care will not start with AI replacing a physician’s clinical identity; it will start with AI stripping away the administrative sludge around the visit.
We already have evidence for how much that matters. A JAMA Network Open study across six healthcare systems found that after 30 days of ambient AI scribe use, burnout among ambulatory clinicians fell from 51.9 percent to 38.8 percent. That came with improvements in after-hours documentation and the time and attention clinicians felt they could give patients.
Could that help eliminate primary care deserts? Maybe, but it is not a magic elixir. The Commonwealth Fund study estimates that 42.6 million people lived in rural primary care shortage areas in 2023 and that 92 percent of rural counties were designated primary care health professional shortage areas. KFF’s latest polling also suggests why AI will be attractive in underserved settings: Among people who used AI for health information, 18 percent say a major reason was that they did not have a regular provider or could not get an appointment, while 19 percent cited affordability.
AI can mitigate the impact of shortages by stretching scarce clinicians further, supporting asynchronous care and making low-touch follow-up more realistic. But it cannot replace local labs, imaging, hands-on exams, broadband, transportation or the trust that comes from a health system actually existing in a community.
Who pays for this is less mysterious than the hype suggests. My read is that the money will come from familiar buckets, not from some brand-new “AI doctor” reimbursement category. Consumer-facing players like Amazon will use membership and pay-per-visit models. Health systems will buy AI because it reduces labor costs and friction inside existing care operations. Providers will try to fit some of this work into payment structures that already reward continuous primary care management and virtual care.
The Centers for Medicare and Medicaid Services says advanced primary care management services already bundle existing care-management and communication technology-based services, and the Department of Health and Human Services says several Medicare telehealth flexibilities now run through the end of 2027. That makes it much more likely that AI is paid for as part of the primary care stack than as a standalone robotic clinician.
As Halamka puts it, “In a value-based purchasing world, those who use AI will deliver higher-quality care to more people in less time, which will result in appropriate compensation for maintaining wellness.”
Who Will Win the AI Doctor Race?
The winner in AI primary care will not be the company that asks patients to trust a machine with everything. It will be the company that first asks them to trust AI with the paperwork, the prep and the routine, then builds the thing people can actually say yes to: An AI layer attached to real clinicians, scheduling, prescriptions, pricing and accountability. By doing so, that company will make the human doctor feel closer.
“A doctor possesses wisdom and human context that an AI, even one trained only on medical knowledge, doesn’t have,” Nadler says. “I’ve talked about how I think the future of knowledge work is moving past data recall and into creativity and wisdom. In our vision of the future, the doctor is always there, making better, more informed decisions on behalf of the patient.”
Credits: TCA, LLC.