That moment of misdirection, described by Dr. Danilo Rojas-Velasquez of Harvard Medical School and Beth Israel Deaconess Medical Center in a new viewpoint article published in Academic Psychiatry, sits at the heart of an argument that is gaining urgency across academic medicine: the language clinicians write into charts is not a neutral record but a durable force that shapes how every subsequent clinician perceives a patient. Psychiatry, a chart-heavy specialty in which practitioners learn to scan prior notes for buzzwords that key early hypotheses, is particularly exposed. Words travel with patients across years of records, and the language chosen by one team determines how the next team imagines the person in the bed.

Clinical medicine has already made a deliberate effort to soften the edges of its vocabulary. Undomiciled has largely replaced homeless, non-adherent has supplanted non-compliant, and declined is preferred to refused. These substitutions reflect a genuine ethical awareness that stigmatizing descriptors can follow patients for decades. But Rojas-Velasquez argues that such lexical swaps can function as a kind of hygiene that leaves the underlying clinical gaze unchanged. A patient called non-adherent by a frustrated team is not necessarily better understood than one called non-compliant; the person is simply described in the currently sanctioned vocabulary. The deeper issue is not which words are approved but what those words do once they are placed in a chart and begin to circulate.

The empirical evidence behind this concern is substantial. A growing body of research documents that stigmatizing language in the medical record measurably alters downstream clinical decisions, including pain management and the intensity of diagnostic workups. Studies have found that descriptors such as pleasant and cooperative are applied unevenly across patient race, meaning that the chart itself can become a vector for the transmission of bias. Physician use of stigmatizing language in patient records has been documented systematically in large-scale analyses, and researchers have shown that negative framing in one note propagates into the impressions formed by later readers. Yet documentation is still often treated as a matter of mechanics: billing compliance, problem-list hygiene, and the architecture of the assessment and plan. The ethics of clinical language, what a clinician is actually doing to a patient when writing about them, has remained a hidden curriculum, absorbed by osmosis from whichever notes a trainee happens to read.

Into this landscape, ambient scribes and large language model-assisted documentation are arriving faster than the educational infrastructure can adapt. Randomized trials of ambient AI scribes in clinical practice are already underway, and health systems are deploying the tools at scale. The technical mechanism of concern is subtle but consequential: AI tools inherit and polish the linguistic conventions of their training data, which means that stigmatizing patterns already present in charts are being laundered into outputs that feel more authoritative precisely because they are fluent. A trainee who might once have written that a patient seems frustrated may now sign a note describing him as hostile and uncooperative simply because that is what the model produced. Fluency, in this context, is not accuracy; it is a cosmetic upgrade applied to inherited bias.

There is also an educational cost. Historically, the note has been the place where clinical formulation happens. Writing was thinking, and the attending’s question of why a trainee described a patient in a particular way was a teaching moment built directly into the workflow. When the note arrives pre-formed by an algorithm, that moment is structurally removed unless educators deliberately rebuild it. The formulation step, in which a clinician weighs observations, considers alternative explanations, and commits to a reasoned characterization of the patient, is being quietly absorbed by the model. What remains for the trainee is verification, which is a fundamentally different and far less formative cognitive task.

In response, the article proposes three practical steps for teaching note writing in the age of AI. The first is to teach note-reading before note-writing. On a rotation, a resident might review prior documentation, name the impression those notes created, and then compare that impression with what actually emerges at the bedside. The point is not to grade the prior note or its author but to make the chart’s authorial power visible before the resident adds to it. This exercise converts an invisible influence into an object of explicit study, allowing trainees to feel firsthand how a one-line summary can pre-load an entire clinical encounter with assumptions.

The second step is to build a next-reader pause into supervision. Before signing a note, the trainee considers how it would read to a covering clinician at three in the morning, to the patient with portal access, and to a family member reviewing the chart. Open Notes initiatives and patient-facing documentation have already shifted the audience for clinical writing; patients can now read the words written about them, and imagining these readers helps residents notice attributions and tones of voice that would otherwise pass unexamined. Qualitative work on patient-centered documentation in the era of open notes suggests that this kind of deliberate audience-awareness changes how clinicians draft, encouraging precision and fairness over shorthand judgment.

The third step treats the editing of AI-drafted notes as genuine clinical work rather than clerical review. If formulation is being absorbed by the model, the trainee’s task is not merely to check grammar or fill gaps but to interrogate what the draft has already decided: where observations have hardened into character judgments, where hedge words have drifted toward unwarranted certainty, and where descriptors would not survive the next-reader test. The article suggests a weekly note rounds, in which a team reviews and revises an anonymized AI-drafted note together, turning a hidden judgment back into a shared teaching moment. In this arrangement, the AI draft becomes a specimen to be dissected rather than a product to be signed, restoring the pedagogical function that automation threatened to erase.

The chart remains one of the few places in medicine where a patient is constructed in their absence, by people with power over them, in language they have historically been unable to see. Open Notes is changing the last part of that sentence, and ambient AI is changing the second, but the construction itself endures, and arguably matters more now than it did when notes lived in locked basements. Consultation-liaison psychiatrists, who live in other clinicians’ notes before they ever meet their patients, are unusually positioned to notice this dynamic. They watch language do its quiet work, and they write notes that will in turn shape how the next team thinks. Whether the current generation of trainees inherits a thoughtful relationship to that responsibility, or simply a more polished vocabulary applied at greater speed, will depend in large part on how medical educators decide to teach it. The stakes are not stylistic. They concern who patients become on paper before anyone meets them in person, and whether the accelerating machinery of AI-assisted documentation amplifies clinical understanding or merely automates its failures.

Subject of Research: Teaching ethical clinical documentation and note-writing skills to psychiatry trainees in the era of AI-assisted medical records

Article Title: Teaching Note Writing in the Age of AI: Three Practical Steps

Article References: Rojas-Velasquez, D. (2026). Teaching Note Writing in the Age of AI: Three Practical Steps. Academic Psychiatry. https://doi.org/10.1007/s40596-026-02434-5

Image Credits: AI Generated

DOI: 10.1007/s40596-026-02434-5

Keywords: clinical documentation, artificial intelligence, psychiatry education, stigmatizing language, medical records, ambient AI scribes, large language models, Open Notes, clinical formulation, medical bias, consultation-liaison psychiatry, graduate medical education