The Friday Ethics Desk

Patients want to be told when AI is used in their care, and a study proposes three levels of telling

Interviews with 20 members of hospital patient advisory councils found that transparency comes first and that a clinician should stay the final decision maker. The other side: ambient AI scribes show measurable gains for the clinicians who use them.

By Matteo Favilli, 2026-11-06

In short

More and more clinics use AI, from tools that write up visit notes to diagnostic support. A new study asked patient advisers what they expect to be told, and they put being informed first and want their doctor to make the final call. The researchers propose telling patients at three levels, depending on the risk of the tool. Studies of AI note-writing tools show benefits for clinicians, which is the reason hospitals want to adopt them.¹²

What happened

On 4 November, PLOS Digital Health published a qualitative study from Hackensack Meridian Health, a large healthcare network, on what patients expect to be told about the AI used in their care. The authors note that, according to an AMA survey, AI use by physicians in clinical practice has doubled over the past two years, and that less attention has gone to whether and how patients want to be informed.¹

The team spoke with members of Patient and Family Advisory Councils, groups of patients, family members and community representatives who advise hospital leadership, at nine hospital sites. A total of 21 members completed an intake survey and 20 of them (95%) completed an interview of about 20 minutes. The sample was predominantly female (62%) and White non-Hispanic (95%), and 81% held at least a bachelor's degree.¹

In the survey, participants agreed that AI has a role in health care (average 4.29 on a scale where 5 is strongly agree) and that all patients should be informed about how AI is used (4.48). They were neutral on the need for an opt-out option (3.33) and on whether current safeguards are adequate (3.14). They strongly rejected letting AI make treatment decisions autonomously (1.38).¹

The interviews produced 1,242 initial codes, which were consolidated into 16 and then four themes: ethical prerequisites for AI, preserving clinician oversight, systemic prerequisites for trust, and strategies for meaningful communication. The most common single code was the patient's right to disclosure and transparency. Participants described transparency as an ethical "floor" that makes it possible to keep control over one's own health. They want their own clinician to be the final decision maker and the main person explaining what the AI does, and they prefer to hear about it early, before a moment of stress.¹

From this the authors propose a framework with three tiers of communication: disclosure, assent and consent. For a low-risk tool a simple notice is sufficient, while high-stakes decisions call for a more detailed discussion and formal patient permission.¹

The authors list limitations. The group is not representative of the general public. The council members were consulted but did not influence the design, deployment criteria or governance of any AI model. The study also used Google's Gemini 2.5 Pro for the first coding of the interview transcripts, followed by human validation.¹

The other side

The case for adopting AI tools such as ambient clinical documentation, which records a visit and drafts the clinical note for the clinician to review and approve, rests on what it does for the people who use it. UChicago Medicine described two studies published in JAMA Network Open in October 2025. In the first, a survey of over 250 physicians and advanced practice providers across six health systems that ran pilots, self-reported burnout dropped from roughly 52% to 39%, and participants also reported lower cognitive burden and less after-hours documentation.²

The second study compared clinicians using the tool at UChicago Medicine with matched non-users. The users spent 8.5% less total time in the electronic health record, and the time spent composing notes fell by over 15%. The researchers argue that for a clinician seeing 20 patients a day, saving two or three minutes per patient adds up to multiple hours a week that can go to patient care.²

The same article is open about what is still missing. The patient view is the next step: early internal survey data show modest but steady improvements in patient-experience scores for clinicians who use the tool, and the team plans interviews with patients and clinicians to see how visits feel. The burnout figures are self-reported.²

Matteo's take

Both sides have a point, and they come from different people. The council members in the first study say yes to AI in health care and no to AI deciding alone, and they ask to be told. The clinicians in the second study report a lighter workload. Neither group is arguing against the other.

The distance between them is about who gets asked. The scribe studies measure clinicians, and the patient perspective is listed as a future step. The council study asks patients, from a group that is mostly White, highly educated and already involved in hospital advisory work, and it says so itself. Each source covers half of the picture, and each one points to the other half as work still to do.

The three tiers are a practical idea. A simple notice for a tool that drafts a note, a conversation and a signature for a tool that influences a treatment. The survey numbers suggest where patients stand: 4.48 for informing everyone, a neutral 3.33 on opting out. That leaves a gap that a health system has to fill with its own choice of who may say no, and to what.

Looking at these sources, I would say AI in the clinic is fair when a few conditions hold. Patients are told, early and in plain language, by someone they trust. A named clinician stays accountable for the decision. And a patient who is uncomfortable has a way to raise it and be heard by a person. The first two are what the council members asked for. The third is still open, since the participants were neutral about an opt-out.

Here is the question I am left with. If a scribe tool saves the clinician time and the patient is told about it with a simple notice, what should happen when a patient says no, and the clinic's workflow now depends on the tool?

Sources

Drafted with AI from the sources below.

All of AI Bits

Loading…