Artificial intelligence has not entered the clinical encounter at one point. It has entered at three discrete points, and they’re rarely discussed as one problem.
Before the visit, patients are using AI to organize symptoms, interpret test results, and decide whether a concern is worth seeing someone about. During the visit, clinicians are using AI to search evidence, broaden a differential, and check guidelines. After the visit, ambient tools are turning the encounter into a note.
Consider one patient moving through all three. She checks her symptoms in an app the night before, arrives having already framed her fatigue as “probably stress.” Her clinician, mid-visit, pulls up a decision-support tool that ranks stress-related causes highest, and the visit follows that ranking. An ambient scribe writes a fluent, well-organized note built around that frame. Nothing in this chain was reckless. Each step, taken alone, looks reasonable. But no one had authorship of the whole story, because no one was asked to.
Each of these has been written about. Almost none of them have been written about together, and treated separately, they miss the thing they share: At all three moments, AI can produce something that looks finished, an organized symptom list, a ranked differential, a clean note, while the underlying clinical thinking was structurally limited. That risk is well documented at this point, in the literature and increasingly in commentary like this. It is not necessary for me to re-argue it here.
Real physician voices, twice a week
Free, and one click to unsubscribe.
What’s had far less attention is what actually has to happen at each of those three points to prevent it, and I believe the discourse has settled on the wrong word to describe it. The usual answer is that the clinician reviews and signs, and that “signature” is where responsibility supposedly lives. That’s a legal fact. It is not a sufficient description of the work.
Authorship, not signature
Medicine has actually already answered a version of this question, just not for the clinical encounter. Every major medical journal now states, explicitly, that AI cannot be listed as an author on a scientific paper, not because it can’t produce fluent text, but because authorship requires something AI cannot hold: accountability for what the words claim. An author has to be able to stand behind every sentence, defend it, and correct it if it’s wrong. A tool that generated the sentence can’t do any of that. Only a person can.
That standard was written for publishing. It applies with even more force to the chart, and to the reasoning that produces it, and medicine hasn’t said so out loud yet.
Clinical authorship means the same thing the journals mean: not reviewing a draft someone, or something, else produced, but being the person who can account for what it says, at the moment it’s produced, not after the fact. “I signed it” or “I approved it,” in the case of EMRs, is not authorship. Authorship is knowing, before you sign, whether it’s true.
What that requires at each of the three moments
Before the visit, authorship belongs to the patient, and the clinician’s job is to help them reclaim it. A patient who arrives having run their symptoms through an AI tool hasn’t necessarily lost their own story, but they may be leading with the tool’s framing instead of their own words, already half-convinced before they’ve said a word. The clinician’s task isn’t to distrust the tool or defer to it. It’s a short set of questions that hands authorship back: What did this help you say? What part of this feels right and fits? What part doesn’t? That’s not skepticism about AI. It’s making sure the story in the room is still the patient’s.
During the visit, authorship belongs to the clinician, and it means something specific: the work of deciding which AI-surfaced possibility deserves weight, and which should be set aside, even when the tool is confident and the citation is real. A guideline summary is not the same as knowing whether the guideline fits this patient. Authorship here is the discipline of disregarding what doesn’t fit, not because the tool was wrong, but because fitting this patient is a judgment the tool was never positioned to make.
After the visit, authorship belongs to whoever’s name is on the note, and it means reading an ambient summary as an interpretation, not a transcript, before it becomes the version every future clinician inherits. A summary that quietly turns “still doesn’t feel right” into “reassured” isn’t a small error. It’s a rewritten story that the next clinician will treat as fact. Authorship at this moment means catching that before it enters the record, not after it’s already shaped someone else’s differential three referrals later.
One method, not three separate conversations
Treating these as three unrelated topics, a documentation-liability question, a decision-support question, a scribe-accuracy question, is exactly why the field keeps re-litigating the same point without building the connective principle that would actually help a working clinician. Authorship is that principle, asked three times, at three different moments of one encounter. But authorship needs something to stand on. It has to be authorship of something specific, or it stays a slogan.
That something is the Care-Full Clinical Sequence: listening, examination, differential diagnosis, shared and understandable reasoning, and treatment planning. Those five steps are not new. They reflect how clinicians have traditionally been trained to approach patient care, long before any of these tools existed. The sixth step is not traditional, and I added it deliberately: feasibility and treatment burden. Medicine has rarely taught clinicians to ask whether a plan can actually survive contact with a patient’s real life, the cost, the transportation, the caregiving, the burden of managing it all. A plan can be perfect on paper and still fail there.
What has changed, with or without that sixth step, is how easily any of this can now look clinically complete without anyone actually having performed the thinking behind it.
That is what authorship has to mean at each of the three moments. Before the visit, it means the patient’s listening step, their own account, wasn’t quietly replaced by a tool’s account. During the visit, it means the clinician’s differential and reasoning are still theirs, even when AI proposed half the list. After the visit, it means the treatment plan and its feasibility for this patient’s actual life were verified by a person, not merely generated and signed.
Go back to the patient from the opening. Authorship doesn’t mean rejecting the app she used, the tool’s ranking, or the note’s structure. It means a clinician who asked what she meant by “probably stress” instead of accepting the frame, who treated the decision-support ranking as one input rather than the answer, and who reread the note before it closed the encounter. Same tools, same visit length, a different outcome, because someone owned the story at every stage instead of only the last one.
The method doesn’t change because the tool is new. It’s the reason the tool can be trusted at all.
This article is discussed on The Podcast by KevinMD: A diagnosis no one authored can follow your patient from visit to visit.
This essay is cited in the KevinMD record on artificial intelligence.
Alan P. Feren is a retired surgeon, independent physician, health care consultant, and patient advocate with more than 50 years of experience in clinical practice, system leadership, and health care innovation. Formerly in academic and community surgical practice, he has worked across the evolving landscape of managed care and clinical governance.
In the 1990s, Dr. Feren co-authored clinical guidelines that evolved into what is now MCG Health, now used by more than 80 percent of U.S. health plans and over 3,100 hospitals. He has advised health technology startups, helped shape managed care policy, and served as a clinical content developer for health care technology platforms.
His work centers on restoring shared understanding between clinicians and patients in an era defined by speed, fragmentation, and technological mediation. Drawing on both professional experience and his own journey as a complex patient, he writes about transparency, accountability, and the disciplined methods that make medical care trustworthy. He is a contributor to KevinMD and a podcast guest. More information is available at mypersonaladvocate.net and on LinkedIn.
