One of the most common questions we hear from physicians who are interested in ambient documentation but haven't tried it is: how does the system know what goes in which section of the note? A patient visit is a conversation. It doesn't come labeled. The physician and patient don't announce "we are now in the Assessment section." How does the system produce a structured note from unstructured speech?
This walkthrough takes a real outpatient visit type - a 15-minute follow-up for a patient with type 2 diabetes - and traces how the spoken conversation maps to each section of a structured SOAP note. The goal is to demystify the technical process and give physicians a concrete sense of what the system actually does with what it hears.
The visit scenario
Patient is a 58-year-old male with type 2 diabetes, hypertension, and hyperlipidemia. He is presenting for a three-month follow-up. His most recent hemoglobin A1c was 7.8. He has been on metformin 1000mg twice daily and lisinopril 10mg daily. He has no new complaints at check-in.
Here is an approximate transcript of the visit, with physician and patient dialogue represented separately.
The spoken visit
Physician: "How have things been going since I saw you in September?"
Patient: "Pretty good overall. I've been trying to watch what I'm eating. Still having a hard time with the blood sugar in the mornings though."
Physician: "What's your fasting reading typically running?"
Patient: "Usually around 140 to 155. Sometimes a little higher."
Physician: "Have you had any hypoglycemic episodes? Any readings below 70?"
Patient: "No, nothing like that. I haven't had that problem."
Physician: "How's the metformin tolerability? Any stomach issues?"
Patient: "No, I got used to it. No problems now."
Physician: "Good. Blood pressure's been OK at home?"
Patient: "Yeah, I've been checking. It's usually in the 130s over 80s."
Physician: "Alright. Let me check you over."
[Physical exam - physician narrates findings aloud]
Physician: "Blood pressure today 136 over 82, heart rate 76 and regular. Lungs clear bilaterally. Abdomen soft, nontender. No peripheral edema. Monofilament exam intact bilaterally."
Physician: "So the A1c came back at 7.8, which is improved from 8.2 but still above our target of 7.0. Fasting glucoses in the 140 to 155 range confirm he's not at goal. I think we need to intensify. I'm going to add glipizide 5mg once daily with breakfast. We'll recheck the A1c in three months."
Physician: "Blood pressure is controlled on current therapy. No changes to the lisinopril. Lipid panel due at annual visit in May."
Physician: "Any questions about the new medication? It can cause hypoglycemia, so if you notice shakiness, sweating, feeling faint, test your sugar and call us. I want you to check your fasting glucose daily and log it. We'll review the log at your next visit."
How this maps to the note structure
Subjective section - The model identifies patient-reported information: presenting concern ("blood sugar in the mornings"), fasting glucose range as reported by patient (140-155), absence of hypoglycemic episodes, metformin tolerability confirmed, blood pressure readings at home (130s/80s). This section contains what the patient said about their own experience.
Objective section - The model captures the physician's narrated physical exam findings: blood pressure 136/82, heart rate 76, lungs clear, abdomen findings, no peripheral edema, intact monofilament exam. It also captures lab context the physician references: A1c 7.8 (result on record). This section contains measurable findings.
Assessment section - The model extracts the physician's stated clinical conclusions: type 2 diabetes not at goal per A1c and fasting glucose pattern; hypertension controlled on current therapy; lipid management deferred to annual visit. The model must distinguish between descriptions of patient history (Subjective), measured findings (Objective), and the physician's interpretation and conclusions (Assessment). The signal is the physician's framing language: "I think we need to intensify," "confirmed he's not at goal," "blood pressure is controlled."
Plan section - The model captures the specific actions the physician stated: add glipizide 5mg once daily with breakfast; no changes to lisinopril; lipid panel at May annual visit; recheck A1c in three months; patient instructed to monitor fasting glucose daily and log it; hypoglycemia warning signs discussed with patient. This is action-oriented content - the things that are happening as a result of the visit.
What the system produces
The generated note draft would look substantially like a note a physician had typed, organized into the SOAP sections above. The physician's editing task is to verify that the clinical content is accurate - that glipizide 5mg is correctly captured, that the A1c value is correct, that no findings are misattributed - and to add any detail that was discussed but not said aloud during the visit.
In a visit like this, editing time is typically under two minutes. The physician is checking clinical accuracy, not rewriting. For more complex visits with multiple active problems and more complex clinical reasoning, editing takes longer - but the structure is still provided, and the physician is working from something rather than a blank screen.
The physician narration habit
One thing the walkthrough above illustrates is that the physician narrated physical exam findings aloud. This is not universal. Many physicians conduct physical exams silently and document findings afterward. When a physician does not narrate the exam, the model has less to work from in the Objective section and will either produce a thinner objective or rely on standard expected findings for the visit type.
Physicians who develop the habit of narrating findings aloud - "normal S1 S2, no murmur," "no tenderness to palpation in the right upper quadrant" - produce noticeably better objective sections. Most physicians who use ambient documentation tools develop this habit within a few weeks. It becomes a natural extension of the clinical exam rather than an additional task.