Which AI will Medical Doctors Use?

OpenEvidence vs ChatGPT vs. Gemini vs Epic vs.... up to date?

OpenEvidence reports that, as of May 2026, two-thirds of US physicians are using their platform. OpenEvidence provides a free service where doctors can ask clinical questions, and makes ends meet by showing targeted advertisements from the biopharmaceutical industry.

ChatGPT, from Open AI, is rolling out a healthcare focused LLM among healthcare providers.

Google has deployed AI results into their search engine, and has already transformed daily searches, including those performed by doctors.

EPIC— the predominant medical record— has debuted AI tools with many more to come.

And some stragglers continue to believe that UpToDate will remain dominant.

All of this begs the question:

Who will win this race?

I can’t answer the question, but I wish to offer 6 thoughts.

1. People who think Uptodate— or any website— where you look up articles by topic and read the full essay— and/or pay a lofty subscription fee to use— will remain dominant, are delusional. They remind me of….

   

2. Controlling the electronic medical record is a remarkable advantage. (Obviously Epic owns this) If, as a doctor is typing the note, or while their phone is using an AI powered app to listen to the encounter and transcribing the summary, relevant ideas are suggested to the physician about clinical care— the doctor will be highly unlikely to look beyond those ideas, and the doctor’s attention will be captured and sated. Just as Gmail suggests email responses, an EMR that suggests you order x,y and z, and explains why, will have a huge, perhaps insurmountable, advantage. Large players interested in this space, with the right advisors, may seek to develop an AI/ EMR and dethrone Epic entirely.

3. Speed of result is the single most important factor that will govern use. Every time you ask AI a question, there is a lag time. Even if one company gains users early, the market share can be forfeited if another product is much more responsive. Doctors, like all people, have no attention span and no patience. A 600ms delay in response is a death sentence. Already google Gemini has blown my mind in terms of speed. How is it possible to return such fast AI results with perhaps a million queries made each second globally? Companies must prioritize this.

4. Doctors will want AI to speculate/ and guidelines are not based in science. I think there is a misconception that Medical AI must hew to well worn paths of knowledge and not to lead into experimental and improvisational medicine. And, don’t get me wrong, I have long been a skeptic of such improvisations, but doctors want to read both well accepted ideas, as well as ambitious and daring ideas. Doctors may not follow those courses of action, but they want an AI that is willing to suggest new hypotheses and treatment approaches.

   Second, and most importantly, AI companies must recognize that current expert guidelines are (a) not scientific, (b) highly speculative, and (c) should not be treated as gospel. A great example is the new cholesterol management guidelines. A group of like-minded individuals, laden with financial conflicts, and entrenched career biases, enters a room and produces a document that is loaded with very specific recommendations that lack randomized evidence. This document has been criticized. E.g. John Mandrola here. Adam Cifu here. Anish Koka here. And many others. The purpose of this essay is not to unpack that criticism, but to suggest to AI companies that different sets of reasonable, smart, dedicated doctors may write very different sets of guidelines. Training AI on the guidelines endorsed by professional societies when evidence is weak will result in inappropriate rigidity, and lead many doctors to bristle at AI use— particularly when the guidelines run counter to their interpretation of the evidence, and give them more work to do. Finally, note, that all the rebuttals I linked to are on platforms (Substack) that would be deprioritized by some of these Medical AI engines.

5. What students need/ what attendings want. AI companies should remember that novices and experts need and want different things. Medical trainees often don’t know where to start when it comes to processing the huge volume of medical information, and experts often simply need a quick reminder of a few details of the protocol. Current Medical AI often strikes me as a very competent 3rd year student— broad knowledge of the canonical thinking— who can do a PhD on any detail you ask about, but that’s very different than an expert physician who has synthesized all of this into a pragmatic and simple flow. AI companies should consider offering a product in medical student vs. attending mode, just like my Tesla’s auto drive can be timid or quite daring.

6. Will the advertisers control the product? Even great medical products like the New England Journal and JAMA or the major medical conferences are controlled ultimately by the advertising companies. Disagreement is curated, heavily redacted, and often dismissed. Studies are allowed to be published with massive missing data and companies are not pressed to provide it. I get it— you can only put so much pressure on your funder. Medical AI however may soon face the challenge that doctors are interested in knowing the limitations and shortcomings to therapies— that the advertisers may wish to de-emphasize. Unlike journals or conferences, there is more competition here, and as such, the focus should be on the user. I would strongly suggest they resist such pressures, and happy to elaborate on that at a future date.

Uptodate has never lived up to its name. Its articles lag new results by weeks, months or years. I loved it in medical school, but it is sadly no longer fit for purpose. Medical AI is the future of doctors and trainees learning and refreshing and refining their clinical thinking. He who controls the EHR, controls the universe. And after that, speed and creativity will dominate preferences. Technology folks and medical folks have different dispositions to new ideas, suggestions, and information— my advice for companies is to be careful that the ‘expert’ advisors you employ are not leading you astray, towards reinforcing entrenched opinions, rather than democratizing them.