A Polished Care Note Can Still Miss the Point

Imagine a visit note that reads very well. The sentences are clean, the headings are in order, and the visit appears complete. On a closer reading, the patient mentioned that a medication was making them dizzy, and the draft recorded the visit as having no new concerns. Fluent writing can make an error look settled.

Reviewing clinical documentation is a large part of my work, and this is a newer form of a problem I have seen for years. In the American Medical Association’s 2026 physician survey, 30 percent of respondents said they use health AI to create discharge instructions, care plans, or progress notes, and 28 percent said they use it for documenting billing codes, medical charts, or visit notes. Those figures describe different uses and do not tell us how many notes are drafted by software. They do suggest that reviewing AI-assisted documentation has become an everyday workflow question.

Meaning first, then style

A well-written draft draws attention to how well it is written. I make a point of asking first whether it captured what the patient meant and what the team actually decided. I look for places where the patient’s uncertainty was recorded as a firm statement, where a possibility was recorded as a diagnosis, or where a tentative plan was recorded as a completed action. I also check whether the concern that prompted the visit is still in the note.

These tools are good at structure and speed. They can also be fluent when they are wrong. AHRQ’s diagnostic safety work also cautions about the limitations of AI and the need for clinician oversight. That review is a different task from proofreading. It means comparing the draft against the encounter, the rest of the record, and the judgment of the responsible clinician.

Follow each action out of the note

Documentation matters because other people act on it. A care manager arranges a service based on it. A clinician decides whether to call the patient. The patient may read it later and understand the plan differently than the team intended. For each action in a note, I ask who is carrying it out, by when, and how anyone will know it is finished. A line that says to follow up with specialty care reads smoothly, yet it is a weak instruction if the referral was never placed, the patient does not know whom to call, and no one is watching for the result. Anyone who has worked in care coordination has learned that the note and the operational plan need to say the same thing.

When reviewers find themselves correcting the same kind of omission repeatedly, that pattern is information about the tool, the workflow, or both. I suggest logging it, without copying patient details into an informal tracker, and using it to adjust prompts, training, vendor decisions, or the point in the process where review takes place.

Respect the conversation behind the note

Some tools record audio, some summarize, and some draft replies to patient messages. Consent, privacy protections, and review steps need to fit the particular tool in use. Patients deserve a plain explanation of how a tool is being used in their care, and staff need approved ways to handle what it produces. The AMA’s 2026 guidance on ethical AI use keeps the physician’s judgment at the center. From an operations standpoint, honoring that means giving the reviewer enough time and context to review properly, because a signature alone does not show that a review took place.

I bring the same question to an AI-drafted note that I bring to any other. If the next person sees only this record, will they understand what mattered, what was decided, and what they are expected to do? When the answer is yes, the note is doing its job, whoever or whatever produced the first draft.

Sources

• American Medical Association: 2026 physician survey on health AI use
• AHRQ: Limitations of AI and the resulting risk
• American Medical Association: Guidance on ethical use of health AI (2026)

About the author
Sherry Sayari, MPH, is an Operational Standards Evaluator who has led population health, community outreach, and social services programs. More on the About page.