AI and Your Doctor

In April 2026, a study published in JAMA Network Open by researchers at Mass General Brigham's MESH Incubator generated a headline that spread quickly across physician forums and social media: AI chatbots misdiagnose in over 80 percent of early medical cases. Physicians shared it as confirmation of what many already believed.

Here is what the headline did not say. The researchers tested 21 large language models on 29 standardized clinical vignettes, feeding information in sequence the way a physician actually encounters it. Models failed to produce an appropriate differential diagnosis more than 80 percent of the time when patient data was incomplete, and were penalized for listing too few diagnoses or even one too many. On final diagnosis, with full clinical information, failure rates fell to less than 40 percent across all models, and the best performers exceeded 90 percent accuracy.¹¹

Co-author Marc Succi said the goal was to help separate hype from reality, and that the results reinforce that large language models in healthcare continue to require a human in the loop and close oversight.¹¹ That is a calibrated conclusion, not a declaration that AI is dangerous.

Most physician commentary defaults to the access argument: the office was closed, the appointment was six weeks out. That is true and it matters. But it is not the whole story. Seven percent of all Google searches, roughly 580 million a day, are health related.²⁵

Patients are not embracing AI because they want less humanity in their healthcare. They are using AI precisely because they want more of it, and the current system is not consistently delivering it.

Editor-in-Chief, Concierge Medicine Today

A joint Massachusetts General Hospital and Harvard University survey of 443,455 U.S. adults across 24 waves found the proportion reporting a lot of trust in physicians and hospitals fell from 71.5 percent in April 2020 to 40.1 percent in January 2024, a decline that held consistently across demographics and political affiliation, and had not meaningfully recovered as of April 2025.³ A study published in Nature Medicine in 2024 found patients who believed identical advice came from an AI chatbot rather than a human physician were less likely to deem it reliable.²⁹ Patients still want their physician's voice on their care. What has changed is their willingness to wait passively for that relationship to show up on its own schedule.

Research published in Frontiers in Psychology examining 1,183 participants found patients still strongly prefer human physicians, particularly for complex or emotionally significant situations.⁴ A 2024 Harvard study found AI made the correct diagnosis 88 percent of the time on standard prompts, compared with 96 percent among human physicians given the same information.¹⁰ That eight-point gap matters clinically, and confirms most patients are not operating under the illusion that AI and physicians are equivalent.

The real risk is narrower: a 2024 KFF survey found 57 percent of U.S. adults do not feel confident determining whether AI-generated health information is true or false.⁸ That is a genuine clinical problem worth addressing directly, educationally, and with patience rather than condescension, not a reason to tell patients to stop using AI.

The commonly cited figure for overall medical diagnostic error rates in the United States is 10 to 15 percent. In primary care specifically, diagnostic errors are estimated to affect 6.3 percent of encounters, more than 12 million Americans annually.¹² A 2024 JAMA study found 23 percent of patients transferred to intensive care or who died in the hospital had a missed or delayed diagnosis.¹³ These are not arguments against physicians. They are arguments against the pretense of perfection, and against a profession that holds AI to a standard of scrutiny it does not uniformly apply to itself.

Caution sounds like: "That's an interesting tool, let's look at what you found together." Dismissal sounds like: "You shouldn't trust that." One of those responses builds a relationship. The other ends one. One in four Americans say they would not visit a healthcare provider who refuses to embrace AI technology.¹⁵ Seventy percent of patients are open to AI tools for researching physicians, and 84 percent check online reviews before appointments.¹⁶

A Sermo poll found 58 percent of physicians believe AI will change the face of healthcare, either diminishing the physician's role or making doctors obsolete.⁵ That fear is understandable. It is also not a strategy. The leadership move is not to resist AI, but to become the physician who understands it better than anyone else in the room, the director, not the reactor.

Nearly half of physicians, 49 percent, reported burnout in 2024.²¹ More than a third of burned-out primary care physicians said they plan to stop seeing patients within one to three years.²² The AMA's own survey found 66 percent of physicians reported using healthcare AI in 2024, a 78 percent jump from the year before.²⁴ The concierge and membership medicine model was built, in part, as an answer to exactly this dynamic: smaller panels, real access, physicians who are genuinely reachable. These models are not anti-technology. They are pro-relationship, and the best physician-patient relationships today are increasingly built by physicians who use every available tool, including AI, to serve patients more completely.