Do AI scribes raise
coding intensity?
Yes, ambient AI scribes can raise coding intensity. Richer, auto populated notes make higher level E/M codes easier to justify, and recent multi system data shows high acuity billing climbed as practices adopted AI scribing. The risk runs both ways. Defensible coding ties every code to the documented medical decision making, not note length.
- 10.3%
- E/M improper payment rate (2024)
- $3.9B
- projected improper E/M payments
- 12 to 20
- point rise in high acuity new visits
What coding intensity actually means.
Coding intensity is the share of visits billed at higher acuity E/M levels, for example 99214 and 99215 instead of 99213, over time. When a practice or system shifts toward higher level codes, average reimbursement per visit rises. That shift is legitimate when the documented medical decision making supports it, and a compliance problem when it does not.
Under the AMA 2021 outpatient E/M guidelines that CMS adopted, level selection is driven by medical decision making complexity or total time. As CMS states plainly, the volume of documentation should not be the primary influence on the billed level. The note has to support the code, not inflate it.
Coding intensity rose as scribing rolled out.
A March 2026 Trillian Health study reviewed all payer claims at six large health systems that publicly adopted AI enabled scribing. From 2018 to 2024, both new and established outpatient visits shifted toward higher intensity E/M codes. The share of high acuity new patient visits rose roughly 12 to 20 percentage points, and established visit high acuity share rose roughly 7 to 12 percentage points.
The researchers were careful: the increase likely reflects more thorough, rules based documentation capturing work clinicians were already doing, not automatic fraud. Ambient scribes record the full encounter, so decision making that used to go undocumented now appears in the note. That can correct chronic undercoding. It can also tip routine visits upward when templates and auto population pad the record beyond what the visit supported.
- 6
- health systems analyzed
- 12 to 20
- point rise, high acuity new visits
- 7 to 12
- point rise, high acuity established visits
What payers and CMS are watching.
The scrutiny is not hypothetical. As ambient AI makes richer notes the default, payers expect the code to match the documented decision making, and they have the analytics to spot outliers. A few signals practices should know in 2026:
- CMS improper paymentsFor the 2024 reporting period, the E/M improper payment rate was 10.3%, a projected $3.9 billion. Incorrect coding drove 49.1% of those improper payments and insufficient documentation another 34.1%.
- OIG enforcementHHS OIG has flagged Medicare Part B improper payments exceeding $31 billion annually, with Recovery Audit Contractors recouping roughly $2 billion per year. A 2025 OIG audit of E/M services billed with modifier 25 recommended recovering up to $124 million.
- False Claims Act exposureThe FCA covers reckless disregard, so a systematic pattern of billing 99215 when documentation supports 99213 can create liability even with no intent to defraud.
- Risk adjustment precedentPayers point to settlements like Kaiser Permanente's $556 million agreement resolving allegations of coding that inflated risk scores.
A better note is not a defensible code.
Most practices are wrong in both directions at once: undercoding complex visits where the note never articulated full medical decision making, and overcoding routine visits where the EHR auto populated the record. Ambient scribes draft and suggest. Deterministic coding scores the encounter and writes the code the documentation supports.
| Capability | Ambient AI scribe (note only) | Deterministic coding (CarePilot) |
|---|---|---|
| Captures the visit | Yes | Yes |
| Selects ICD-10 / CPT code | Suggests, often in free text | Writes the code to discrete fields |
| Scores against AMA 2021 MDM | No | Yes |
| Rationale attached for audit | Rarely | Yes |
| Guards over and under coding | No | Yes, both directions |
| Coding accuracy measured | Not accountable | 98%, tracked |
How to keep coding defensible in 2026.
The fix is not less documentation. It is coding tied to the documented medical decision making, written to discrete fields, with a rationale you can defend in an audit.
- 01
Code to MDM, not note length
Tie each E/M level to the documented number and complexity of problems, data reviewed, and risk, the three MDM elements CMS uses.
- 02
Separate documentation from coding
A scribe that writes the note should not also be the thing that quietly inflates the level. Coding should be an explicit, reviewable step.
- 03
Write to discrete fields
Codes that land in structured EHR fields with a clear rationale are auditable. A suggestion in free text is not.
- 04
Audit prospectively
Catch misassignment before submission, not after a RAC demand with a three year lookback.
- 05
Measure accuracy
Track coding accuracy as a real metric, in both directions, so you can prove defensibility.
Coding that stays defensible.
Inside athenahealth and DrChrono, CarePilot turns the documented encounter into ICD-10 and CPT codes written back to discrete fields, at 98% coding accuracy, with the rationale attached. It is rules based and accountable, so the code reflects the medical decision making the note supports, not the length of the note. The ambient note captures the visit. CarePilot makes the coding defensible.
- 98%
- coding accuracy
- 78
- minutes back per day
- 1 to 2
- business days to go live
Coding intensity questions.
Do AI scribes cause upcoding?
Not by themselves, but they can contribute. Fuller, auto populated notes make higher level codes easier to justify, and multi system data shows coding intensity rose as ambient scribing spread. Upcoding happens when the billed level exceeds what the documented medical decision making supports.
What is coding intensity in medical billing?
It is the trend toward billing more visits at higher acuity E/M levels over time. Higher intensity raises reimbursement per visit and is legitimate only when documentation supports the level under AMA 2021 MDM rules.
Is rising coding intensity illegal?
No. It is legitimate when documentation supports the higher levels and can correct historical undercoding. It becomes a compliance problem when codes exceed documented medical decision making, which can trigger audits and False Claims Act exposure.
How does CarePilot keep coding defensible?
CarePilot codes inside athenahealth and DrChrono, scoring the documented encounter and writing ICD-10 and CPT codes to discrete fields with a rationale, at 98% coding accuracy, so the code matches the medical decision making rather than note length.
Will an AI scribe pass a payer audit?
A scribe documents the visit but does not make coding defensible on its own. Audits compare the submitted code to the documentation. Defensibility requires coding tied to MDM with an auditable rationale.
Make every code defensible.
Coding that matches the documented decision making. Written back into your EHR. Book a 30-minute demo.