AI SOAP Notes: Drafts Your Clinicians and Coders Can Trust
AI SOAP notes can lift charting work off your clinicians when every section of the draft holds up to coding and payer review. Learn how AI builds each section, where dictation fits, and what to check before you choose a tool.
A coding manager opens yesterday's clinic chart and finds a clean AI-drafted note. The Subjective holds a full history, and the Objective lists every vital sign.
Further down, though, the Assessment says "knee pain" with no side, and the Plan reads "follow up as needed." So the coder sends a query, and the claim waits.
Moments like this decide the value of AI SOAP notes, because saved charting time pays off when coders can work from every section.
In this article, we cover how AI drafts each section, what coders need, where dictation fits, and how to evaluate a tool.
TL;DR: AI SOAP Notes at a Glance
The short version, before the detail:
What they are: AI SOAP notes are draft clinical notes that AI builds from an encounter and sorts into Subjective, Objective, Assessment, and Plan for the clinician to review and sign.
Where the value is decided: The Assessment and Plan carry the diagnoses, specificity, and medical necessity that coding and payer review depend on.
Capture mode: Ambient capture suits conversational visits, and dictation covers notes with no single conversation behind them, such as many inpatient progress notes.
What to check: Test any tool on your own specialties for attribution, specificity, inference behavior, template fit, and EHR integration.
Who stays accountable: Your clinicians review and sign every note, and your coding team works from what they signed.
What Is an AI SOAP Note?
The SOAP note is a widely used four-part structure for clinical documentation: Subjective, Objective, Assessment, and Plan. NIH's StatPearls reference on SOAP notes credits physician Larry Weed with the format and describes it as both a record of the patient's status and a communication tool between clinicians.
An AI SOAP note keeps that structure and changes who writes the first draft. The AI turns what happened in the encounter into text, sorts it into the four sections, and returns the draft for review.
Most tools follow the same basic sequence:
Capture: The system records the visit conversation through ambient capture, or the clinician dictates the note directly.
Speech-to-text: Speech recognition converts the audio into text using medical vocabulary.
Sorting: Language models place each statement in a section and format the result to a template.
Review: The clinician edits the draft, approves any suggested codes, and signs before the note enters the record.
Placement deserves as much attention as speed. A sentence filed in the wrong section can change what a coder reads as the diagnosis.
What Each Section of an AI SOAP Note Needs
Each section serves a different reader downstream. Centers for Medicare and Medicaid Services (CMS) guidance on evaluation and management services lists what every encounter record should include: the reason for the visit with relevant history, exam findings, and prior test results; the assessment or diagnosis; and the plan of care.
Use this section-by-section check when you review sample drafts:
Section | What it holds | What to check in an AI draft |
|---|---|---|
Subjective | Chief complaint, history of present illness, relevant history | Statements attributed to the right speaker, negatives recorded accurately |
Objective | Vital signs, exam findings, results | Findings the clinician actually stated, dictated, or entered |
Assessment | Diagnoses and clinical impression | Specificity, laterality, and no diagnosis the clinician did not state |
Plan | Orders, medications, referrals, follow-up | The reasoning behind each order, written out or easy to infer |
The sections below take each part in turn:
Subjective
The Subjective section records the patient's account: the chief complaint, the history of the present illness, and relevant history. The same CMS booklet notes that the patient usually states the chief complaint in their own words.
Ambient capture performs well here, since the patient supplies most of this content in conversation. Check that the draft attributes each statement to the right speaker and records negatives accurately, so "denies chest pain" never turns into "chest pain."
Objective
The Objective section holds what the clinician measured and observed: vital signs, exam findings, and results. An AI system cannot see the exam, so these findings have to come from what the clinician said aloud, dictated, or entered.
Watch for templates that fill in a normal exam the clinician never performed or stated. Under Medicare's rules, the history and exam still need to be medically appropriate for the visit, even though they no longer set the level for most evaluation and management (E/M) visits.
Assessment
The Assessment is where diagnoses live, and diagnosis codes follow it closely. Under the current ICD-10-CM Official Guidelines, code assignment rests on the provider's diagnostic statement and is reported to the highest level of specificity the record documents.
Specificity is the first detail to check in any AI draft. The guidelines treat unspecified-side codes as a last resort for records that lack the detail and cannot be clarified, so a missing side turns into a query and a delay.
The AI should never fill that gap with a guess.
Our Director of Product Implementation, Garrett Dunne, describes the boundary we hold: "There's not a situation where the engine is being essentially, the model is being asked to infer a diagnosis. We wouldn't allow it to do that."
Chronic conditions need the same care. HCC medical coding for risk adjustment, built on Hierarchical Condition Categories, depends on what the Assessment supports for each condition.
Plan
The Plan records orders, medications, referrals, and follow-up, along with the reasoning behind them. The CMS E/M booklet says the rationale for ordering diagnostic and other services should be documented or easy to infer.
A thin Plan puts payment at risk. A CMS fact sheet on documentation requirements states that Medicare can deny payment for incomplete records, and its error reviews found insufficient documentation of medical necessity in the plan or intent to order diagnostic tests.
For most E/M visits, Medicare selects the level based on medical decision-making or time. The Assessment and Plan together carry most of the evidence a coder uses to support that level.
How to Keep AI SOAP Notes Concise and Complete
An AI draft can run longer than a note the clinician would type, because the model has the whole conversation to work from. Extra length adds reading time for everyone who opens the chart.
Length also does nothing for the visit level. The CMS E/M booklet states that the volume of documentation should not be used to decide the level of service billed.
The StatPearls reference cited earlier adds the clinical reason. A brief but complete SOAP note is easier for the next clinician to follow.
Ask vendors how their drafts handle small talk, repeated statements, and side conversations. A good draft keeps the clinical content and leaves everything else out of the chart.
When to Record the Visit and When to Dictate the SOAP Note
AI SOAP notes start in one of two capture modes, and each gathers the note's content differently.
The two modes work like this:
Ambient capture: Ambient clinical documentation records the natural conversation between your clinician and the patient, then drafts the note from that dialogue.
Dictation: The clinician speaks the note directly, section by section, using medical dictation software built for clinical vocabulary.
SOAP structure shows up in outpatient visits, inpatient progress notes, and postoperative follow-ups. The right mode depends on where each encounter's content comes from.
Use this view to match the mode to the encounter:
Encounter | Where the SOAP content comes from | Better fit |
|---|---|---|
Outpatient clinic visit | The patient-clinician conversation supplies most of the Subjective | Ambient capture |
Inpatient progress note after rounds | The clinician's interval history, exam, and plan, often with no single conversation to record | Dictation |
Postoperative follow-up | A short conversation plus findings and next steps the clinician summarizes | Ambient capture with a dictated Assessment and Plan |
Before you compare vendors, settle a few questions:
Care settings: List which of your encounters are conversational and which are dictated, across inpatient and outpatient care.
Mode switching: Ask whether a clinician can move between ambient capture and dictation within one encounter, for example, to dictate the Assessment and Plan after the patient leaves.
Shared templates: Confirm that recorded and dictated notes land in the same SOAP format, so coders read every note the same way.
Supporting both capture modes in one documentation workflow is the problem we built InstaNote to solve. DeliverHealth provides AI-powered clinical documentation and coding solutions for organizations ranging from small and mid-size practices to enterprise-level health systems.
InstaNote captures ambient conversation and dictation across inpatient and outpatient care and returns a near-final, structured note in about three minutes. Its speech recognition is designed not to add anything that wasn't said, and every note carries coding suggestions your providers review and approve before sign-off.
Book an InstaNote demo to watch a SOAP note take shape from a sample encounter in your specialty.
Connect the SOAP Note to Coding
The clinical note is where the revenue cycle starts. Every code, claim, and appeal downstream is a reading of what the Assessment and Plan say.
Our VP of Business Operations, Gautam Kumar, names what happens when that reading has gaps: "If the note is incomplete, or for whatever reason it is inconsistent, or if there is some, you know, clinical specificity missing, every downstream process becomes reactive."
Your coding team feels the effect first, in queries, rework, and delayed claims. Efforts to improve medical coding accuracy start with complete, specific documentation.
We extend that documentation into coding with InstaCode medical coding, which codes from InstaNote, traditional dictation, operative reports, or existing electronic health record (EHR) notes without changing how documentation happens. InstaCode delivers 95% coding accuracy with human-in-the-loop review, and AAPC- or AHIMA-credentialed DeliverHealth coders review every complex or flagged encounter.
Every coded encounter stays auditable, with the AI's output, each human change and the reasoning behind it, and any auditor edits on record.
How to Evaluate an AI SOAP Note Tool
A demo shows how a tool performs on sample notes aligned to your specialty. Benchmarking against your own encounters happens in a pilot, so plan for one before you commit.
Run the evaluation against these checks:
Attribution: Confirm the draft separates what the patient reported from what the clinician observed and concluded.
Specificity: Compare drafted Assessments against your coding team's expectations for laterality, acuity, and linked conditions.
Inference behavior: Ask what the system does when the clinician leaves a diagnosis open, and test it with encounters where the workup is still in progress.
Template fit: Check that formats follow your organization's standards and each provider's preferences, including specialty variations of the SOAP layout.
Care setting coverage: Run test encounters from each setting you serve, inpatient and outpatient, including notes your clinicians will dictate.
EHR integration: Ask whether the note writes to discrete EHR fields or arrives as a block of text.
Review workflow: Measure how long clinicians spend editing each draft, and track which sections they change most.
Security: Require a signed business associate agreement before sharing any patient data, and treat HITRUST certification as a separate, voluntary check on top of the Health Insurance Portability and Accountability Act (HIPAA).
Pilot results from your own specialty mix will tell you more than a headline accuracy figure. Track queries and edits by SOAP section during the pilot, so you can see where drafts hold up.
Frequently Asked Questions
Leaders evaluating AI SOAP notes tend to ask about these points:
Can Clinicians Sign AI SOAP Notes Without Editing Them?
Clinicians should read every draft before signing, since the signature makes them accountable for the record.
Tuning templates to each provider is meant to shrink edits over time, and the review step stays in place either way.
Can AI Write the Assessment and Plan?
AI can organize the Assessment and Plan from what the clinician stated during the encounter or dictated afterward.
The clinician provides the diagnoses and decisions, and a well-designed system will not infer a diagnosis the clinician didn't state.
Do AI SOAP Notes Work for Inpatient Progress Notes?
They do when the tool supports dictation and ambient capture.
Inpatient progress notes often follow rounds, so clinicians need a quick way to dictate updates in SOAP form.
Are AI SOAP Notes HIPAA Compliant?
Compliance depends on the vendor and how it handles protected health information (PHI).
Expect a signed business associate agreement, encryption in transit and at rest, and clear answers about where audio and text are stored.
How Long Does It Take to Roll Out AI SOAP Notes?
Timelines depend on your specialty mix, your EHR, and the integration scope.
For InstaNote, typical deployment runs two to three weeks, and discovery work often starts before the contract is signed.
How Is an AI SOAP Note Different From a SOAP Note Template?
A SOAP note template supplies the sections and formatting a clinician fills in by hand. An AI SOAP note uses that same template as its frame and fills the sections from the encounter, ready for review.
Start With the Section Your Coders Query Most
Pull a few weeks of coding queries and denials, then sort them by SOAP section. The section where they cluster is the first one to test in any AI draft.
A SOAP note that reads well and codes cleanly saves time twice, once for the clinician and again for everyone who works from the chart.
At DeliverHealth, we build AI-powered documentation and coding solutions around that idea, and InstaNote is where it starts. Trusted by more than 40,000 healthcare providers, InstaNote formats every note to your organization's and each provider's requested templates and works across more than 50 EHRs, including Epic, Cerner, MEDITECH, athenahealth, eClinicalWorks, and NextGen.
InstaNote runs on our HITRUST i1-certified eSOne platform and includes coding suggestions your providers approve before sign-off.
Schedule an InstaNote demo to see a specialty-aligned sample encounter move from conversation to a signed SOAP note.
Stay Updated
Subscribe to our newsletter for the latest healthcare AI insights and company updates.