An AI scribe can make your visit notes more complete. That is good news for clinicians who are tired of charting after hours. It also raises a quiet billing question: if the note says more, does the E/M level you bill still match it?
The goal is not to bill higher, and it is not to bill lower out of fear. The goal is a level that the note actually supports, every time. This guide gives owners, billing staff and clinicians a simple, practical way to get there. It is general information, not legal or coding advice, so please check your approach with a certified coder or compliance advisor.
Why fuller notes change billing conversations
In 2026, researchers started to document what happens to coding after AI scribes arrive. Trilliant Health data, reported by ACDIS in April 2026, looked at hospital systems and found outpatient E/M visits shifting toward 99204-99205 and 99214-99215. The share of high-level new-patient visits rose by 12 to 20 percentage points in the studied systems, and by as much as 80% at one of them (ACDIS summary).
Two things are worth keeping in mind:
- These were hospital and health-system studies. Small practices work differently, so treat the numbers as a signal, not a forecast for your clinic.
- The researchers did not call this fraud. A fuller note can support a level that was previously under-coded. Many clinicians document less than they actually did, and a scribe can capture more of it.
Payers are paying attention, though. A Blue Cross Blue Shield Association analysis estimated about $22 million in extra maternity spending from 2023 to 2024 and called for oversight (BCBSA report). We mention it only as context: commentary expects payers to look harder at high-level E/M codes and to ask for more documentation (Medwave). The practical takeaway is calm: make sure each level you bill can be defended by the note in front of you.
How E/M levels are chosen, and what “supported by the note” means
For office and outpatient visits, CMS explains that the E/M level is based on either medical decision making (MDM) or total time on the date of the visit. CMS also notes that add-on code G2211 needs documentation of the reason for billing it (CMS MLN006764, May 2026).
If you code by MDM
The note should show the pieces that drive the level: the problems addressed, the data reviewed or ordered, and the risk of the management options. A long note is not the same as a note that supports a high MDM. Length is not a coding element.
If you code by total time
The note should state the total time and what kind of work it covered, according to current guidance. Do not rely on a template phrase that filled in automatically. The time statement should be the clinician’s own, and it should be true for that visit.
“Supported” in one sentence
A level is supported when a reviewer who has never met the clinician can read the note and find the specific lines that justify the code. If they cannot, the level is at risk, whether it is too high or too low.
Under-coding counts too. If your notes now clearly support a 99214 and you have been billing 99213 out of habit, that is an accuracy gap as well. The target is a match, in both directions.
A simple monthly self-audit
You do not need an expensive audit program to start. A small, regular check takes under an hour once you have a routine.
- Pull a sample. Each month, choose a handful of 99214 and 99215 notes (and 99204-99205 if you see new patients). Include a mix of clinicians, and notes written with and without the AI scribe.
- Underline the support. For each note, mark the exact lines that support the level: the problems, the data, the risk, or the time statement. If you cannot underline anything, flag it.
- Compare level and note. Mark each one as supported, over-coded or under-coded.
- Look for patterns. One odd note is a coaching moment. The same gap in five notes is a template problem.
- Fix the template, not the person. If notes often miss the risk discussion or the time statement, adjust the note structure so the information has a clear place to go. Blaming a clinician rarely fixes anything.
- Write down what you found. A one-page log with the date, sample size, findings and changes shows that you check your own work.
Our guide to building a clinical template is a good companion when the audit points to a template gap.
Clinician review before signing
An AI scribe drafts. The clinician is the author of record. That one principle protects both your coding and your credibility.
A quick review checklist for every AI-drafted note
- Does the note match what actually happened in the visit?
- Is anything wrong, missing, or something the clinician did not do or say?
- Are the problems addressed clear, and does the plan reflect the clinician’s own thinking?
- Is any time statement accurate for that visit?
- Is there copied or boilerplate text that does not belong?
- Does the level you plan to bill match the lines in the note?
Put it in writing
A one-page written policy helps everyone act the same way. It can cover:
- That the clinician reviews and edits every AI draft before signing.
- Who selects the E/M level, and when a coder or biller double-checks it.
- How patients are told about the scribe, following your state’s rules and your own consent process.
- How errors are corrected and how often the policy is reviewed.
If you are still setting up your scribe, our AI scribe rollout checklist for the first 30 days covers consent, templates and training.
Preparing for payer requests and downcoding
Even well-documented practices get documentation requests. Some payers also adjust E/M levels automatically, often called downcoding, and an AMA reference committee document says this is rising. A little preparation makes these moments routine.
Documentation requests
- Assign one person to receive and track requests, and note the due date the day they arrive.
- Send exactly what was asked for, in the format requested, and keep a copy.
- Make sure the signed, final note goes out, not an unreviewed draft.
Appeals
- Before appealing a downcode, re-read the note and underline the lines that support your level. If they are not there, an honest correction may be the right answer.
- If the note does support the level, cite those lines and the MDM or time elements in a short, clear letter.
- Track outcomes by payer, so patterns guide your next self-audit.
Billing and front-desk handoffs
Good coding starts before the visit. Accurate eligibility, correct visit types and clean demographics all reduce friction later. Our front-desk checklist to cut claim denials walks through the handoffs that keep claims clean, and our piece on charge capture covers making sure every service is recorded. Agree on who tells whom when a note is late, a level is queried, or a payer letter arrives.
Keep the time savings in perspective
Interest is high: Doximity’s 2026 report found that 94% of physicians use or are interested in AI (Doximity). Time savings are real but modest. A JAMA 2026 study associated AI scribe use with roughly 16 fewer documentation minutes per eight scheduled patient hours in the studied academic settings, and a NEJM AI study called savings of that size “modest.” A scribe is a helpful tool, not a shortcut around review.
How MedTec helps
MedTec is a cloud EHR and practice-management platform for small and medium practices, with an AI scribe built into the chart. The scribe drafts the visit note, and the clinician reviews, edits and approves it. That keeps the clinician in charge of what goes into the record and what level it supports. Our medical billing tools live in the same system as the chart, so the people who write the notes and the people who bill them work from the same place.
No software can guarantee compliance, and we do not claim to. What a connected system can do is make the review step easy to follow and easy to show.
Your next step
This week, pull ten 99214 and 99215 notes, underline the line that supports each level, and note where the template could help. Then write your one-page review policy.
Want to see how the AI scribe and the clinician review step work together in the chart? Book a free 30-minute demo at medtec.ai.