GP Clinic Undercoding Revenue: How to Stop Invisible Losses with AI

Undercoding is the consistent pattern of billing a lower-complexity evaluation and management code than the clinical note supports, assigning an unspecified ICD-10 diagnosis when a more specific one exists, or omitting comorbidities and risk factors that would change the reimbursement level entirely. It does not trigger a denial. It does not generate an alert. The claim is submitted, the payment arrives, and the revenue loss is invisible in standard AR reporting because the claim was technically paid

This blog explains exactly what undercoding is, why GP clinics are structurally exposed to it, what the three types look like in real notes at real dollar differences, and how AI clinical documentation corrects the problem at the point of the visit rather than after the revenue is already lost.

What Is Undercoding and Why Is It Harder to Detect Than Overcoding?

Undercoding is the assignment of a billing code that does not reflect the full complexity or scope of the clinical service delivered. In a GP clinic context it takes three primary forms: billing a lower E/M level than the documentation supports, using a non-specific ICD-10 diagnosis code when a more precise one applies, and failing to document all conditions actively managed during a visit so they are not factored into the complexity assessment at all.

Overcoding is detected quickly because it triggers audits, recoupment demands, and compliance investigations. Undercoding is detected slowly if at all, because the claim is accepted, the payment posts, and the shortfall between what was earned and what was collected never appears as a variance in any standard report.

A 2023 study published in JAMA Health Forum, examining EHR documentation data from more than 18,000 physicians across family medicine, internal medicine, cardiology, and orthopaedics, found that despite CMS modifying E/M coding requirements specifically to reduce documentation burden, meaningful reductions in documentation time did not follow. Clinicians continued to face the same structural time pressure that drives incomplete documentation and, by extension, systematic undercoding.

That structural pressure has not changed. It is the environment every GP clinic operates in every day, and it is the environment that produces undercoding encounter by encounter.

How Much Revenue Is a GP Clinic Undercoding Revenue Lost?

The revenue impact of undercoding operates at three levels simultaneously.

Level 1: The per-encounter E/M differential. The Medicare Physician Fee Schedule differential between a 99213 (low complexity established patient) and a 99214 (moderate complexity established patient) is approximately $37 per encounter at national rates. Between a 99214 and a 99215 (high complexity) the differential is approximately $45. When a moderate complexity encounter is documented as low complexity, that $37 is gone permanently. There is no recovery mechanism because the claim was paid.

Level 2: The annual compounding across a clinic day. A GP clinic seeing 80 patients per day across two providers, at a 19% undercoding rate, is undercoding approximately 15 encounters per day. At $37 per undercoded encounter on E/M alone, that is $555 per day, $2,775 per week, and approximately $133,000 per year in E/M revenue that was earned and never collected.

Level 3: The ICD-10 specificity layer on top. Separate from E/M undercoding, non-specific ICD-10 codes suppress reimbursement, trigger denials at a higher rate, and in some cases reduce the documented severity of a patient’s conditions in ways that affect value-based care metrics. The FY2026 ICD-10-CM update introduced over 400 new codes explicitly to increase diagnostic specificity. Practices that have not adapted face claim denials averaging $283 per rejected encounter, according to revenue cycle data. For a three-provider GP clinic with a 15 to 20% initial denial rate, outdated coding alone can represent more than $85,000 in annual revenue impact.

2026 Medicare Advantage adds a fourth layer: algorithmic downcoding. 73% of multi-specialty practices billing Medicare Advantage plans experienced systematic E/M downcoding in 2025, according to MBC’s 2026 revenue cycle management analysis. MA plan algorithms automatically reduce submitted 99214 or 99215 codes to 99213 without individual clinical review, costing the average multi-specialty practice between $40,000 and $180,000 annually in suppressed reimbursement. The practices best positioned to resist algorithmic downcoding are those with documentation that explicitly and completely supports the submitted code. A note that clearly documents the number of problems addressed, the complexity of data reviewed, and the risk level of management decisions is significantly harder for an automated downcoding algorithm to downgrade without generating an appealable basis.

What Does Undercoding Actually Look Like in a GP Note?

The gap between a note that accurately supports reimbursement and one that does not is rarely dramatic. It is almost always the difference between a note that documents the minimum required to describe what happened and one that documents what actually happened.

The following three examples show the same patient visit at three different documentation levels, with the resulting E/M code and approximate Medicare reimbursement for each.

Documentation Level 1: Minimal note

“Patient presents for follow-up. BP controlled on current medication. Continue lisinopril. Review in 3 months.”

Resulting E/M code: 99212 (straightforward medical decision-making) Approximate Medicare reimbursement: $57

This note reflects a visit where a provider reviewed one stable chronic condition, confirmed current medication, and scheduled a follow-up. That is what the documentation says. That is what the coder is able to support.

Documentation Level 2: Partial note

“Established patient for follow-up of hypertension and type 2 diabetes. BP 128/82, within target range. HbA1c reviewed, stable at 7.1. Continue current medications including metformin and lisinopril. RTC 3 months.”

Resulting E/M code: 99213 (low complexity medical decision-making) Approximate Medicare reimbursement: $94

This note documents two chronic conditions, reviews relevant lab data, and confirms medication management. Better than Level 1. But still incomplete because it does not document the complexity of the clinical decision-making process, the risk factors assessed, or whether any additional problems were considered or discussed.

Documentation Level 3: Complete note

“Established patient presenting for follow-up management of type 2 diabetes with stage 3 diabetic CKD and hypertension. BP 128/82, within target range on current lisinopril dose. HbA1c 7.1 reviewed alongside recent eGFR trend showing mild decline from 48 to 43 over 6 months. Medication reviewed for renal dosing appropriateness; metformin continued with monitoring plan. Patient reports new right knee pain over 3 weeks, no trauma history, mild swelling noted on examination. Range of motion limited. Assessment: early osteoarthritis vs early inflammatory process. Conservative management initiated with low-dose NSAID, limited to 2-week course given CKD stage. Review in 4 weeks with imaging if not resolved. Decision to defer nephrology referral pending next eGFR. Total time with patient and chart review: 22 minutes.”

Resulting E/M code: 99215 (high complexity medical decision-making) Approximate Medicare reimbursement: $175

Both Level 1 and Level 3 describe a real clinical encounter. The difference in reimbursement between them is $118 on a single visit. Across a clinic day, across a year, across multiple providers, that difference is not theoretical. It is the revenue gap between what the clinic earned and what it collected.

The Level 3 note is not longer because the visit was more complex. It is more complete because it documented the complexity that was actually there.

Why Does Undercoding Happen? The Three Root Causes in GP Clinics

Understanding why undercoding happens is essential to understanding why retrospective fixes like coding audits do not fully solve it.

Root Cause 1: Documentation written under time pressure defaults to the minimum. Ambulatory physicians spend nearly 40 to 50% of their workday on EHR and administrative tasks, according to research on clinical documentation burden. Under that pressure, the parts of the note that drive coding accuracy, including the explicit documentation of all problems addressed, the reasoning behind management decisions, and the comorbidities actively considered, are exactly the parts that get compressed or dropped. The note defaults to what happened. It does not capture the full complexity of how the clinician thought about what happened.

Root Cause 2: Habit-based coding without reference to the note. Some clinicians code from habit, assigning the same level code to most established patient visits regardless of what the note actually documents. AAPC data shows that when clinicians code by habit rather than by documented MDM, a high proportion default to 99213 even when their documentation supports 99214 or 99215. The coder then has no basis to assign the higher code because the submitted level reflects the clinician’s entry.

Root Cause 3: ICD-10 specificity not maintained with annual updates. The ICD-10-CM code set is updated annually. The FY2026 update added over 400 new codes. Practices that have not updated their documentation templates and EHR code sets continue to assign codes that were valid under the previous year’s guidelines but are now non-specific under the updated standard. The clinical picture supports a higher-specificity code. The note as written only supports the outdated, less specific one.

What Coders Actually See in an Undercoded Note

When a billing or coding team receives a clinical note, they are working from the text alone. They cannot call the provider to ask what was intended. They cannot infer complexity from the provider’s known clinical expertise. They can only assign codes that the documentation explicitly supports.

Here is what the same clinical information looks like from a coder’s perspective at each documentation level.

What the coder sees in a minimal note: One problem addressed. No data reviewed or referenced. No complexity of decision-making documented. Risk level unclear. Single medication confirmed. Code supportable: 99212.

What the coder sees in a partial note: Two chronic conditions mentioned. One lab value referenced. No complexity of data interpretation documented. No additional problems addressed. No risk assessment documented. Code supportable: 99213.

What the coder sees in a complete note: Three problems actively managed including a new presenting complaint. Lab trend reviewed and interpreted. Medication adjusted based on clinical reasoning. Risk documented explicitly including renal dosing consideration. Referral decision documented with clinical rationale. Time documented. Code supportable: 99215.

The coder’s job is not to reconstruct the complexity of the visit. It is to code what is documented. AI clinical documentation generates notes where what is documented matches what happened.

How AI Clinical Documentation Fixes Undercoding at the Source

Retrospective fixes for undercoding, including coding audits, E/M distribution reports, and provider education sessions, identify the problem after the revenue is already lost. Even an AI medical scribe does not solve this, because it generates a note and stops without engaging with coding accuracy or MDM complexity capture.

AI clinical documentation fixes the problem at the point of the note, before any code is assigned and before any claim is submitted.

Step 1: Ambient listening captures the full clinical encounter. The AI system listens during the patient visit using ambient audio technology. Every problem discussed, every data point reviewed, every management decision made and its clinical rationale, and the time spent are captured in real time. No manual input is required from the clinician during the visit.

Step 2: Structured SOAP note generated with full complexity captured. The system generates a SOAP note configured to the clinic’s specific templates and clinical language. The note captures all problems addressed in the Subjective and Assessment sections, the data reviewed and interpreted in the Objective section, the complexity of the medical decision-making in the Assessment and Plan sections, and documented time where applicable. Every element that the 2021 CMS E/M guidelines and the 2026 MDM framework require to support a higher-complexity code is present because the system captured it from the encounter.

Step 3: ICD-10 specificity structured into the note output. The system generates notes with the diagnostic specificity that the current ICD-10-CM code set requires. Unspecified codes appear in notes because the documentation did not provide specificity. When the documentation is specific because it was captured from the full clinical conversation, the coding can be specific too.

Step 4: Provider review and approval. The clinician reviews, edits if needed, and approves. The note is not finalised automatically. Clinical judgement and provider sign-off remain the final step. The system is a documentation assistant, not an autonomous documenter. The difference is that what the clinician is reviewing is a complete note rather than a compressed one.

Step 5: Accurate coding from day one. The billing team receives a note that documents the full complexity of the encounter. The code they assign reflects what actually happened. The claim is submitted at the accurate level from the first encounter after go-live.

Without AI Documentation vs With AI Documentation: What Changes

Without AI documentation:

• Clinician sees a complex multi-problem patient
• Types a note between appointments capturing the minimum needed to describe what happened
• Comorbidities considered but not explicitly documented
• MDM complexity present but not stated
• Coder assigns 99213 because that is what the note supports
• Claim paid at $94
• $81 in earned revenue never collected on that single encounter
• No alert, no denial, no recovery mechanism

With AI documentation:

• Clinician sees the same complex multi-problem patient
• AI captures the full encounter in real time including all problems, data reviewed, MDM complexity, and time
• Complete note generated and reviewed at sign-off
• Coder assigns 99215 because that is what the note supports
• Claim paid at $175
• Full earned revenue collected on that encounter
• Repeats accurately across every subsequent encounter from day one

The compounding outcome: A two-provider GP clinic converting consistent 99213 coding to accurate 99214 and 99215 coding where the documentation supports it recovers between $50,000 and $133,000 annually in revenue that was already earned but not collected, based on AAPC undercoding rates and Medicare Physician Fee Schedule differentials.

What This Means for GP Clinics in 2026

The FY2026 ICD-10-CM update raised the specificity bar that undercoding analysis is measured against. Over 400 new diagnosis codes increasing specificity requirements mean that practices still using non-specific codes are not just leaving revenue on the table. They are generating denial exposure that did not exist before October 2025. The two problems, undercoding and outdated ICD-10 specificity, now compound each other.

Medicare Advantage algorithmic downcoding makes complete documentation the only viable defence. 73% of multi-specialty practices experienced systematic MA downcoding in 2025. The practices with the strongest position in MA downcoding appeals are those whose notes explicitly and completely document the MDM complexity that supports the submitted code. A note generated by AI documentation that captures all problems, all data reviewed, and all risk considerations is significantly harder to downgrade algorithmically without generating an appealable basis.

Coding audits identify past losses. AI documentation prevents future ones. Retrospective audits have value as a diagnostic tool. They reveal the pattern and quantify the damage. But the audit does not recover the revenue from encounters already billed. AI documentation corrects the documentation from the first approved note. Every encounter from go-live onwards is coded accurately because the note it came from is complete.

The return on investment is calculated from existing patient volume, not new revenue. The revenue opportunity from correcting undercoding does not require more patients, longer appointments, or additional services. It is revenue already earned from the existing patient panel, already delivered through existing clinical work, that is currently not being collected because the documentation does not reflect the complexity of the care.

Incomplete documentation that drives undercoding also increases exposure to prior authorization denials, since the same missing medical necessity language and unspecified ICD-10 codes that suppress reimbursement also fail payer review at the authorisation stage.

Clinics using AI documentation also benefit from a built-in HIPAA-aligned documentation framework that governs how patient data is captured and stored.

MedLaunch Documentation Intelligence is configured to GP and specialty clinic workflows before go-live, integrates natively into Epic and Athena Health, and generates notes that reflect the full complexity of the encounter from the first approved session. Most clinics are fully live within two to four weeks.

The full workflow is on the AI Clinical Documentation Intelligence solution page.

Frequently Asked Questions

What is undercoding in a GP clinic?

Undercoding is the practice of billing a lower-complexity evaluation and management code, a less specific ICD-10 diagnosis, or fewer services than the clinical encounter actually supports. It happens because the clinical note did not capture the full complexity of the visit, not because the care was simpler. The claim is paid, but at a lower rate than the care delivered warranted. According to AAPC audit data, up to 19% of office visit E/M levels are undercoded nationally.

How much revenue does undercoding cost a GP clinic annually?

The Medicare Physician Fee Schedule differential between a 99213 and a 99214 is approximately $37 per encounter. At a 19% undercoding rate for a GP clinic seeing 80 patients per day across two providers, the annual E/M revenue loss from undercoding alone exceeds $133,000. GP clinics lose between 8 and 14% of their total E/M revenue to undercoding annually, according to 2026 Medicare reimbursement analysis. This figure does not include additional revenue loss from non-specific ICD-10 coding on top of E/M undercoding.

Why do GP clinics undercode?

The most common cause is incomplete clinical documentation under time pressure. When clinicians document between appointments, the note defaults to the minimum required to describe what happened rather than capturing all the elements, including all problems addressed, data reviewed, comorbidities, and risk level, that accurate E/M coding requires. A 2023 study published in JAMA Health Forum examining over 18,000 physicians found that despite CMS modifying E/M requirements specifically to reduce documentation burden, meaningful reductions in documentation time did not follow.

What is the difference between undercoding and downcoding?

Undercoding is when the provider or coder assigns a lower code than the documentation supports, typically due to an incomplete note or conservative coding behaviour. Downcoding is when the payer reduces the submitted code on adjudication, either through automated review or audit. Both result in lower reimbursement than the care warranted, and both are worsened by incomplete documentation. A note that does not clearly document MDM complexity gives both the coder and the payer less to work with.

How does AI clinical documentation reduce undercoding?

AI clinical documentation systems generate structured notes from the clinical encounter in real time, capturing all problems discussed, the full complexity of the medical decision-making, comorbidities considered, and time spent. The note reflects the encounter as it happened, not what the clinician had time to type after it. When documentation accurately reflects clinical complexity from the start, coders can assign the accurate E/M level and the correct ICD-10 codes on every claim from go-live.

Does AI clinical documentation support the FY2026 ICD-10-CM updates?

Yes. AI documentation systems configured to current clinical coding requirements generate notes with the diagnostic specificity that the FY2026 ICD-10-CM updates require. The October 2025 update introduced over 400 new codes increasing the specificity needed for accurate claim submission. Practices using structured AI documentation that captures specific diagnostic language from the clinical encounter are better positioned to meet those requirements than practices relying on time-pressured manual notes written after hours.

Can undercoding be fixed through coding audits alone?

Coding audits are valuable as a diagnostic tool. They identify the pattern and quantify the revenue that has already been lost. But a coding audit does not recover the revenue from encounters already billed at the wrong level, and it does not prevent undercoding from continuing on future encounters unless the underlying documentation behaviour changes. AI documentation fixes the documentation at the point of the visit. Every encounter from go-live onwards generates a complete note that supports accurate coding, making undercoding from that point onwards a choice rather than a structural consequence of time pressure.

Conclusion: Closing the Gap Between Care and Collection

Undercoding is not a failure of clinical expertise; it is a structural byproduct of documentation fatigue. In the high-pressure environment of a GP clinic, the nuance required to justify a Level 4 or 5 visit is often the first thing sacrificed to save time. However, as the 2026 Medicare and ICD-10 updates prove, the cost of “brief” notes has never been higher.

When a clinic loses 8% to 14% of its E/M revenue to invisible undercoding, it isn’t just missing out on profit; it is losing the resources needed to scale, hire, and provide better patient care.

AI Clinical Documentation changes the math by removing the manual burden of proof. By capturing the full complexity of the encounter in real-time, AI ensures that your documentation finally matches the high level of care you provide. You no longer have to choose between a completed schedule and a correctly billed one.