For a practice with a large Medicare Advantage panel, the annual note is where risk adjustment lives or dies. A chronic condition that gets assessed and treated but never makes it into a coded, documented problem for that calendar year does not count toward the patient’s risk profile, and the gap shows up later as an unsupported code or a missed one. An ambient AI medical scribe touches exactly one part of this: it drafts the clinical note from what you say during the visit, so the conditions you actually address have a written record. It does not code the HCC, it does not compute a risk score, and it does not send anything anywhere. Understanding where that line sits is the whole point of this piece.
Key takeaways
- The scribe drafts your spoken assessment into a note; risk-adjustment coding and submission stay with your coder and your billing system.
- HCC capture depends on you documenting the condition and its detail. A condition you never dictate does not document itself.
- ICD-10 suggestions are starting points that turn on specificity, so confirm every one against what the note supports.
From end of visit to drafted note
Assist tier, per clinician, annual billing
Codes the scribe submits to a plan or CMS
Source: Patient Square product specifications, 2026.
What risk adjustment actually rewards
Medicare Advantage pays plans a monthly amount per member that is adjusted for how sick that member is. The mechanism is the CMS Hierarchical Condition Category model, which groups thousands of ICD-10-CM diagnosis codes into condition categories that carry weight, and rolls those weights into a risk-adjustment factor for the patient (CMS, Risk Adjustment in Medicare Advantage). A patient with well-documented diabetes with complications, heart failure, and chronic kidney disease has a very different risk profile from a patient whose chart shows only “diabetes” with no further detail, even when the underlying medicine is the same.
Two features of the model drive everything downstream. First, it resets every calendar year: a chronic condition documented last year does not carry forward, so it has to be captured again this year to count. Second, CMS has been phasing in an updated version of the model, V28, which changes which conditions map to a category and how they are weighted (CMS, 2024 Rate Announcement). The details of the model are a coding and finance problem, not a bedside one. What reaches the exam room is simpler and more stubborn: the code is only as good as the note behind it.
The note is the raw material, and it is often thin
Coders cannot code what clinicians do not write. A condition has to be documented as assessed or addressed during a face-to-face encounter, with enough clinical detail to support the specific diagnosis, before it can support a code that feeds an HCC. The industry shorthand for that standard is that a condition should be monitored, evaluated, assessed, or treated in the note, and the diagnosis should be stated clearly enough that a coder does not have to guess.
This is where documentation habits quietly cost a practice. You manage the patient’s CKD and their diabetes competently, you adjust a medication, you plan a follow-up, and then the note that comes out says “diabetes, stable” and never mentions the kidney disease you spent five minutes on. The care happened. The record of it did not. When that pattern repeats across a panel, the practice’s coded risk drifts below the panel’s real clinical burden, and the shortfall is invisible until an audit or a year-end reconciliation surfaces it.
Where an ambient scribe fits
The scribe listens through the visit and drafts the note from what you say. If you state that the patient’s type 2 diabetes now shows diabetic chronic kidney disease and you describe the stage, that linkage and that detail land in the draft because you said them. If you assess the heart failure and note it is compensated on the current regimen, that goes in too. Two minutes after the patient leaves, you have a draft that reflects the assessment you actually delivered rather than the compressed version you would have typed at 9 p.m.
The mechanism matters here because it also bounds the benefit. The scribe records what you say and nothing you do not. It has no prior chart, no problem list it reasons over, and no ability to infer a condition you managed silently. If a chronic condition never enters your spoken assessment, it never enters the draft, and no amount of ambient capture fixes a diagnosis that was addressed but never articulated. The habit the scribe rewards is the same one good risk-adjustment documentation has always rewarded: say the condition, say its detail, say what you did about it.
| Note element | What you say out loud | What the scribe drafts |
|---|---|---|
| Linked conditions | ”Type 2 diabetes with diabetic CKD, stage 3b” | The diagnosis and linkage as stated |
| Status of a chronic problem | ”Heart failure, compensated on current dose” | The condition with your status assessment |
| What you did about it | ”Continued the ACE inhibitor, recheck labs in six weeks” | The management and plan |
| Specificity that drives coding | ”COPD with an acute exacerbation today” | The stated manifestation, for you to confirm |
The boundary, stated plainly
AI Medical Scribe is one module inside Practice Copilot, the platform Patient Square builds. It listens during the visit and hands back a structured SOAP note, ICD-10 suggestions, and a prescription draft, ready to review and sign about two minutes after the visit. Read what is not in that sentence for risk adjustment. It does not group codes into HCCs. It does not calculate a RAF score. It does not know which version of the CMS model your plan is under, and it does not submit a diagnosis to any health plan or to CMS. Those are coding, compliance, and billing functions that sit in your coder’s hands and your revenue systems, and they stay there.
The ICD-10 outputs are the clearest place to keep this straight. The scribe offers a suggested code from what you said, and for risk adjustment that suggestion is a first draft, not a decision. Risk-adjustment coding turns on specificity that a single spoken sentence rarely resolves on its own, so the CKD stage, the diabetes manifestation, or the exacerbation status all need you or your coder to confirm the exact code the documentation supports. Our explainer on how ICD-10 suggestions are generated walks through why the suggestion is deliberately a starting point, and how notes affect claim denials covers the same discipline on the billing side.
Annual recapture without the year-end scramble
Because the model resets each January, the practical work of risk adjustment is recapturing every active chronic condition during that year’s visits. Many practices do this in a compressed annual-wellness sweep, which is exactly when documentation quality tends to suffer under time pressure. If the note drafts itself from your spoken assessment of each active problem, the recapture record accumulates as you work through the patient rather than as a separate typing task afterward. The scribe does not tell you which conditions still need recapture, and it does not track a patient’s coded status across the year. It gives you a cleaner draft of the visit you just did, which is a better input to whatever recapture workflow your practice and coder already run.
Security and ownership
Risk-adjustment work concentrates a lot of diagnostic detail in one place, so how the record is handled matters. Visit audio is processed in memory and discarded the moment the note drafts, so no recording of the encounter exists afterward. Notes are encrypted in transit with TLS 1.2 or higher and at rest with AES-256, they belong to your practice, and you can export or delete them at any time. We offer a signed BAA to every customer, our safeguards map to the HIPAA Security Rule, and our SOC 2 Type II audit is underway. If compliance gates the decision, start with the security page and the detail on BAAs and consent.
Whether it earns a place
If your practice carries meaningful Medicare Advantage risk and your documentation gaps show up as chronic conditions that get treated but not captured, the trade is worth weighing. The scribe does not close that gap for you. It makes the note reflect what you said, which only helps if you are saying the conditions and their detail out loud in the first place. For a practice already disciplined about spoken assessment, that is a faster path from visit to a coder-ready draft. For a practice whose real problem is silent management, the fix is upstream of any tool.
Pricing starts from $79 per clinician per month on the annual Assist plan, then steps up to Copilot and Autopilot, which add a bundled AI EHR, messaging, and receptionist and follow-up automation. On the scribe-only plan it runs alongside your existing EHR and hands you an export rather than writing back into it, which is the posture that matters for risk-adjustment: nothing the scribe drafts reaches a plan or CMS on its own. Compare options in our roundup of AI medical scribes, see the primary care scribe overview since that is where most risk adjustment happens, and read how to evaluate an AI scribe before committing to any vendor. Want to see it on a real annual visit? Book a demo. Pricing is on the pricing page.