Documentation workflow

3 Things Doctors Are Missing Out On If They Aren't Doing Structured Clinical Narration Right

Your AI scribe can only write what it hears. If your reasoning stays in your head, it stays out of the note: the coding, the accuracy, and the patient's understanding all go with it.

Side-by-side photo: in the silent visit a physician reads a chart while a patient with a wrist cast looks confused; with structured narration the physician faces the patient and explains while she smiles

Your AI scribe can only write what it hears

An ambient AI scribe is supposed to end the evenings spent closing charts. You talk with your patient, and a draft note is waiting when the visit ends. For many independent physicians, though, the draft does not match the visit. The thinking that drove your decisions never made it onto the page.

That gap is a workflow problem, not a software bug. The tool listens to the room. If you review a chart in silence, examine without speaking, or counsel in general terms, the note misses the reasoning that made the visit complex.

Structured clinical narration closes the gap. It means saying your clinical reasoning out loud, to the patient, as you work: the severity of each problem, the data you are reviewing, and the risks of the plan. Here are three things you lose when you skip it.

What is structured clinical narration?

Structured clinical narration is the habit of saying your medical decision making out loud during the visit: what you see on exam, what data you are reviewing, how serious each problem is, and why you chose the plan. You speak to the patient, not to a microphone, so the explanation does double duty as patient education.

Three-panel illustration of structured clinical narration: a physician narrates clinical context, the conversation is captured, and the physician reviews and signs the SOAP note draft
Narrate, capture, review: the physician stays the author of the note.

Where the habit comes from

The habit is older than the name. Physicians working with human scribes, especially in emergency departments, have long called out exam findings so the scribe could chart in real time. Ambient AI scribes made that habit matter in every exam room, because the software has no eyes. It cannot see the rash, the gait, the joint you are moving, or the outside report on your screen. If you do not say it, the tool cannot write it.

Why the E/M rules make it matter

The 2021 office visit changes, extended to other E/M settings in 2023, removed history and exam from code selection. Level is now set by MDM or by total time (AAFP, FPM). That has two consequences for practices using a listening tool:

  • Time-based coding rewards slower charting. Documentation time counts toward total time, so a faster note can mean a lower time-based level for the same work.
  • MDM rewards visible reasoning. The level depends on problems, data, and risk, and an auditor can only credit what the note records.

Narration is how your reasoning gets from your head into the note, in words the patient also hears.

Before and after: the same visit, two ways

A new patient presents with worsening right knee osteoarthritis despite six weeks of physical therapy and meloxicam. The physician switches the NSAID and orders labs.

Part of the visitSilent versionNarrated version
Data reviewReads the PT report and last month's outside X-ray on screen. The note never mentions them."I've reviewed the reports from your physical therapist and the X-ray report from last month."
Problem status"Is it worse?" "Yeah.""Your right knee arthritis is progressing despite six weeks of therapy and meloxicam, and it's now waking you at night."
ExamExamines in silence."Right knee: joint line tenderness, a small effusion, flexion limited to 110 degrees by pain, ligaments stable."
Plan and risk"Let's switch your medicine and check some blood work.""I'm stopping meloxicam and starting diclofenac 75 mg twice daily. Because of your kidney numbers, I'm ordering a metabolic panel today to check your kidney function."

The narrated note shows a chronic illness with progression, three data items (two outside reviews and one ordered test), and prescription drug management. That supports moderate MDM, or 99204. The silent note leaves an auditor guessing about everything but the medication change.

Narration will not turn this visit into a 99205. Under CPT definitions, a severe exacerbation carries significant risk of morbidity and may require escalation in level of care, and routine NSAID lab checks do not meet the definition of intensive monitoring for toxicity (URMC MDM definitions; Physicians Practice). Narration makes the real level visible; it does not raise it.

1. Accurate coding and fewer denials that hold up in an audit

Since January 1, 2021, office visit codes 99202 through 99215 are selected by medical decision making (MDM) or by total time on the date of service. History and exam are no longer part of code selection; they only need to be medically appropriate (AAFP, FPM). A longer review of systems will not move a 99204 to a 99205.

MDM has three elements, and the level is set by the highest level reached in at least two of the three (AMA MDM table):

MDM elementModerate (99204 / 99214)High (99205 / 99215)
Problems addressedChronic illness with exacerbation; 2+ stable chronic illnesses; or a new undiagnosed problem with uncertain prognosisChronic illness with severe exacerbation, or a problem posing a threat to life or bodily function
Data reviewed and analyzed1 of 3 categories: any 3 tests, external notes, or independent historian; independent interpretation of a test; or discussion with an external physician2 of the same 3 categories
Risk of patient managementPrescription drug managementDrug therapy requiring intensive monitoring for toxicity; decision about hospitalization or escalation of care

Paraphrased from the AMA MDM framework. Consult the full CPT definitions before coding.

An auditor can only credit what is in the note. AAFP coders put it plainly: a more explicit note could support a higher level by clarifying the physician's decision making (AAFP, FPM). When the reasoning happens silently, the listening tool has nothing to write down.

What narration sounds like for a 99204:

"Your blood pressure log shows your hypertension has stayed stable on this dose. These new daily headaches are a new problem, and the cause is uncertain until we test. I'm reviewing Dr. Smith's cardiology notes and last Tuesday's metabolic panel, and I'm ordering a CBC today."

That covers a moderate problem and three Category 1 data items (one external source, one reviewed test, one ordered test).

What narration sounds like for a 99205:

"I'm looking at the images from your lumbar MRI myself. My read shows severe central canal stenosis at L4-L5. That matches what your neurologist and I discussed by phone this morning."

Independent interpretation plus a discussion with an external physician meets two data categories, the bar for extensive data. Pair it with a high-severity problem or a high-risk plan, said aloud, to reach 99205.

There is also a time argument. If the tool makes documentation faster, time-based coding pays less, so MDM is the path that reflects the work. Confirm specific code choices with your coder or your Medicare Administrative Contractor. For a deeper look at the two pathways, see The Hidden Revenue in Your Complex Encounters: 99204 vs 99205.

Beyond E/M: fewer denials on procedures and add-on services

The same rule protects services billed alongside a visit: payers deny when the record does not show why the service was needed. Two common reasons on a denial are CARC 50 (not deemed medically necessary) and CARC 11 (diagnosis inconsistent with the procedure) (X12 CARC list). Saying the indication out loud puts it in the note.

ServiceWhat the record must showWhat to say aloud
12-lead EKG (93000)A symptom, sign, or known disease. Medicare does not cover an EKG done as a screening test or as part of a routine exam (CMS billing article A57326)."This chest pain is new and spreads to your left arm, so I want an EKG right now to check your heart."
Major joint injection (20610)The joint, the side, the indication, and what was injected. For hyaluronic acid, payers often require documented failure of conservative therapy (Highmark policy)."Your right knee arthritis hasn't improved after six weeks of anti-inflammatories and physical therapy, so we'll inject a steroid into the right knee today."
CGM interpretation (95251)At least 72 hours of data, your interpretation, and the resulting action; generally billable once per month (ACP; ADA)."I'm reviewing the last 14 days of your glucose monitor. You're in range 65% of the time with lows overnight, so we'll lower your evening insulin."
Psychotherapy, 60 minutes (90837)53 or more minutes of psychotherapy, with start and stop times or total time (CMS billing article A57520), plus the approach used."We're starting at 2:05." Then at the close: "It's 3:00. Today we worked on reframing your panic triggers."
Preventive visit plus a problem visit (modifier 25)A problem-oriented service that is significant, separately identifiable, and documented apart from the preventive exam (AAFP; AMA)."That finishes your annual exam. Now let's look at the numbness in your feet."

Three cautions. Speak reasons, not code numbers; your coder assigns codes, and patients only need the why. Narration documents medical necessity that exists; it never creates it. And because rules vary by payer and Medicare Administrative Contractor, check the applicable coverage policy before relying on any single requirement.

2. Notes that are accurate, specific, and yours

Ambient AI scribes make mistakes, and the published error types are consistent. Researchers describe four: hallucinations (content never said), omissions, misattribution of who said what, and misread context. Physical exams are especially prone to invented findings (Medical Economics interview on npj Digital Medicine commentary).

The first randomized trial of two commercial scribes found the same. Physicians reported occasional clinically significant inaccuracies, most often omissions and pronoun errors, and the authors called for active physician oversight (UCLA Health; NEJM AI).

Narration reduces the room for guessing:

  • Speak exam findings as you go. "Lungs clear bilaterally, no wheezing. No pedal edema in either leg." A spoken finding gives the tool a source; silence invites a template.
  • Name who is speaking. "Your daughter mentioned the falls started in June" keeps a caregiver's report from landing as the patient's own history.
  • Say the plan in one closing summary. A clear wrap-up gives the assessment section a single source of truth.

Specific notes also matter for payment. Medicare contractors have treated cloned documentation, entries worded the same as prior visits, as failing medical necessity because it lacks patient-specific information (ACEP EMR FAQ). A note built from your own words for this patient on this day is the opposite of a clone.

Narration does not replace review. Read every draft before you sign it; you remain responsible for the record.

3. Patients who understand their plan and trust your reasoning

Many physicians worry that talking "to the room" will feel stilted. Done well, it reads to the patient as a doctor thinking with them, not at a screen.

The evidence on attention is encouraging. In a multicenter study of physicians using an ambient AI scribe for 30 days, burnout fell from 51.9% to 38.8%. Physicians also reported significant gains in their ability to give undivided attention and in how well patients could understand their care plans from the notes (Olson et al., JAMA Network Open, as reported by Healio). In the UCLA trial, physicians said both tools enabled better engagement, and fewer than 10% of patients declined (UCLA Health).

Narration turns that attention into understanding. Compare a silent prescription with this:

"Because of your kidney function, I want to avoid anti-inflammatories like ibuprofen. Let's try this instead, and recheck your labs in two weeks."

The patient hears the why behind the plan. The note captures the same reasoning, including the risk you weighed and the option you set aside.

Start with consent. Ask before you record, every time. For telehealth, recording-consent rules follow the state where the patient is located, and some states require every party to agree (Healio).

How to start: narrate in three phases, then review

You do not need a longer visit. You need to say out loud what you are already thinking.

  1. Intake: set severity. After the patient consents, start recording and let them talk. Then summarize the trajectory aloud: "So the shortness of breath has gotten much worse this week, and it's worse lying flat."
  2. Exam: look and say. Speak each finding as you examine. "Thyroid symmetric, no nodules. Normal respiratory effort." Name any data you review: whose notes, which tests, what you interpreted yourself.
  3. Wrap-up: say the plan. Before you stop recording, summarize the assessment and plan to the patient, including risks and monitoring. "Methotrexate needs close monitoring for liver and kidney toxicity, so we'll draw baseline labs today and repeat them on a set schedule."
  4. Review before you sign. Read the draft against what happened. Remove anything you did not say or do, and add anything it missed.

Practice on two or three visits a day for a week. The phrasing becomes habit faster than most physicians expect.

How MediAttune approaches this. MediAttune drafts the note from what was said in the room, and you review, edit, and sign it. It never adds a finding you didn't state, never fills in an exam element you didn't perform, and never suggests a billing code: coding stays with you and your biller. If it wasn't said, it isn't in the draft. The scribe is in early access and built for practices on Office Ally EHR 24/7 and Practice Mate.

The bottom line: your voice carries your value

Be realistic about time. In the largest real-world rollout to date, Permanente Medical Group saw about 18 seconds saved per appointment compared with non-users (Ohde et al., npj Digital Medicine). In the UCLA trial, the better-performing tool cut note time by about 41 seconds versus 18 seconds for usual care (UCLA Health).

The bigger payoff is what the note contains. Structured clinical narration makes your decision making visible, keeps the note tied to what actually happened, and helps patients follow the plan. You are already doing the cognitive work. Say it out loud so it counts.

Frequently asked questions

What is structured clinical narration?

It is the habit of saying your clinical reasoning aloud during a visit: problem severity, data reviewed, and plan risks. An ambient AI scribe can then document your medical decision making accurately.

Does an AI scribe help me bill 99205?

Only if the note supports high MDM in at least two of three elements: problems, data, and risk. The tool documents what it hears, so unspoken reasoning cannot be credited.

Can narration help prevent denials for procedures like EKGs and joint injections?

Yes, when the service was truly needed. Payers commonly deny for missing medical necessity or a diagnosis that does not match the procedure. Stating the indication aloud gets that reason into the note.

Will talking through my reasoning lengthen visits?

It should not. You are voicing thinking you already do, and much of it doubles as patient education.

Do I still need to review AI-generated notes?

Yes. Trials and expert reviews report omissions, misattribution, and invented content. You are responsible for every note you sign.

Do I need patient consent to use an ambient AI scribe?

Ask every patient before recording. Recording-consent laws vary by state, and telehealth visits follow the patient's location.

Sources and further reading

This article describes documentation practice and general coding rules; it is not coding, legal, or billing advice. Confirm code selection with your coder, your MAC, and your payer contracts.

You already do the thinking. Let the note keep up.

MediAttune drafts your note from what you say during the visit, so your reasoning, your data review, and your plan make it onto the page while they're still fresh. You review, edit, and sign.

Join the early access list