Clinical Update — June 2026: This guide has been revised to reflect the CMS CY2026 Physician Fee Schedule final rule updates to E/M documentation requirements and G2211 add-on eligibility criteria, the AMA's 2026 Augmented Intelligence policy framework for physician oversight of clinical AI, and USCDI v3 conformance standards now enforced across certified EHR platforms. All FHIR endpoint specifications, LOINC references, and MDM logic have been updated accordingly.

What Is Scribing.io? The Clinical Operations Playbook for Context-Aware AI Documentation

TL;DR

Scribing.io is a context-aware clinical documentation assistant that merges ambient session audio with historical EHR data via secure FHIR endpoints—ensuring every note draft is patient-specific and historically accurate. Unlike generic ambient scribes, Scribing.io auto-surfaces longitudinal lab values (A1c, eGFR) through FHIR Observation/$lastn, performs multi-speaker diarization in noisy clinical environments, and drafts MDM language that explicitly documents data review and medication risk. The result: defensible E/M leveling, preserved G2211 add-on eligibility, and documentation that aligns with both AMA oversight principles and CMS billing requirements. Explore Scribing.io Pricing to see how this works for your organization.

Table of Contents

• What Is Scribing.io and Why Does It Matter for Clinical Documentation?

• Beyond Generic Ambient Scribing: The Documentation Gap Competitors Leave Open

• Clinical Logic Masterclass: How Context-Aware AI Prevents Downcoding

• FHIR Architecture: Observation/$lastn, DiagnosticReport Fallback, and LOINC Normalization

• Technical Reference: ICD-10 Documentation Standards

• G2211 Add-On Eligibility: Documentation Requirements and Scribing.io's Role

• AMA Augmented Intelligence Compliance and Physician Oversight

• Deployment Operations: Epic, Cerner, and Sandbox Validation

What Is Scribing.io and Why Does It Matter for Clinical Documentation?

Scribing.io is a context-aware clinical documentation assistant purpose-built for the complexity of modern patient encounters. It is not a transcription tool. It is not a generic ambient scribe. It is an AI system that fuses two data streams in real time:

1. Ambient session audio — captured from the clinical encounter, including multi-speaker environments with patients, caregivers, and clinicians.

2. Historical EHR data — retrieved through secure SMART on FHIR R4 endpoints using USCDI v3 standards, including lab results, problem lists, medication reconciliation, and prior visit context.

By merging these streams, Scribing.io produces note drafts that are not only accurate reflections of what was said in the room but also clinically enriched with the patient's longitudinal record. Every drafted note cites specific, timestamped Observations—lab values, diagnostic results, active medications—ensuring documentation is patient-specific, historically grounded, and defensible under payer scrutiny.

This distinction matters because the American Medical Association's 2026 Augmented Intelligence framework makes clear that AI in healthcare must operate under physician oversight, with transparency, accountability, and evidence-based grounding. Scribing.io was designed from its architecture forward to embody these principles: the physician always confirms the final note, the clinical logic is explainable, and every data point traces back to a verified source in the patient's record.

But the AMA's policy framework—while correctly establishing guardrails around AI autonomy—addresses the governance layer of AI in medicine. What it does not address is the operational documentation gap that causes real revenue loss and compliance risk every day in Family Medicine and Cardiology practices across the country. That is precisely the gap Scribing.io closes.

Beyond Generic Ambient Scribing: The Documentation Gap Competitors Leave Open

The ambient AI documentation market has matured, and virtually every competitor describes some version of "ambient scribing." The policy conversation has kept pace—the AMA, The Joint Commission, and CMS all now address AI-assisted documentation in their respective frameworks. But both the competitor landscape and the policy conversation miss a critical operational reality:

E/M leveling and CMS add-on code G2211 require explicit linkage to longitudinal problems and recent objective data. Ambient transcription alone cannot fulfill this requirement.

The CMS 2026 E/M documentation guidelines are unambiguous: Medical Decision Making at moderate or high complexity requires documented evidence of data reviewed, problems addressed, and risk of management. The clinician's cognitive work must appear in the note. If it doesn't, the work didn't happen—at least not for reimbursement purposes.

Here is what generic ambient scribes do versus what the documentation actually requires:

The core insight: a clinician's cognitive work—reviewing labs, assessing medication risk, weighing longitudinal complexity—often happens silently. A JAMA Internal Medicine study on documentation burden confirmed that physicians spend nearly two hours on documentation for every one hour of direct clinical care. The paradox is that much of the clinical reasoning driving that documentation occurs in the physician's mind during the encounter—not in dictated speech. Without explicit documentation of that reasoning, the note cannot support high-complexity MDM, and the G2211 add-on code is indefensible.

Scribing.io closes this gap by treating the EHR as a co-participant in documentation. The system doesn't wait for the clinician to say "the A1c is 8.4 percent." It already knows—because it queried the FHIR server before the encounter began—and it drafts MDM language that explicitly references the data, the clinical significance, and the continuity-of-care context.

Clinical Logic Masterclass: How Context-Aware AI Prevents Downcoding in Complex Chronic Care

Consider a scenario that plays out thousands of times daily across primary care:

A PCP in a bustling clinic treats a 68-year-old with Type 2 diabetes mellitus and CKD Stage 3b. The waiting room is full. Background noise is constant. The patient's adult daughter interjects frequently—asking about medication side effects, relaying symptoms her father doesn't mention, questioning the care plan. The clinician, operating efficiently, reviews the latest A1c and eGFR on screen, assesses the risk profile of the active SGLT2 inhibitor and ACE inhibitor combination, and adjusts the management plan. But amid the pace, noise, and interruptions, the clinician does not verbalize:

• The review of the A1c result (8.4%) or its trend

• The review of the eGFR value (41 mL/min/1.73m²) or its implications for medication dosing

• The medication-risk assessment that justifies the complexity of decision-making

• The continuity-of-care reasoning that supports G2211 eligibility

A generic ambient scribe captures the conversation—including the daughter's interjections mixed into the clinician's speech—and produces a note that reflects only what was said aloud. The result: the claim is downcoded from 99215 to 99214, and the G2211 add-on is denied, because the documentation lacks explicit evidence of data review, risk assessment, and longitudinal care complexity.

Step 1: Pre-Encounter Data Retrieval via FHIR

Before the encounter begins, Scribing.io queries the patient's EHR via SMART on FHIR R4 endpoints. Using the Observation/$lastn operation with LOINC-normalized parameters, the system retrieves:

• Most recent A1c: 8.4% (LOINC 4548-4), resulted 2026-04-12

• Most recent eGFR: 41 mL/min/1.73m² (LOINC 48642-3), resulted 2026-04-12

• Active medication list: empagliflozin 25 mg daily, lisinopril 20 mg daily, metformin 1000 mg BID (MedicationRequest resources, status: active)

• Active conditions: E11.65 (Type 2 DM with hyperglycemia), N18.3b (CKD Stage 3b), I10 (Essential hypertension)

When labs are not atomized as discrete Observations—a common reality in practices using older HL7 v2 interfaces or reference labs with limited FHIR maturity—Scribing.io executes a DiagnosticReport fallback, parsing structured report resources to extract relevant values. This dual-query architecture ensures data availability regardless of the sending laboratory's FHIR implementation granularity.

Step 2: Speaker-Aware Diarization in a Multi-Speaker Environment

During the encounter, Scribing.io's diarization engine separates speech into distinct attribution channels:

• Clinician — clinical reasoning, assessment language, plan directives

• Patient — symptom reporting, history responses, preferences

• Caregiver (daughter) — contextual information, questions, non-clinical interjections

The daughter's remarks about medication side effects are captured and attributed correctly—not conflated with the clinician's medical decision-making. This prevents the common ambient-scribe failure of attributing caregiver statements to the physician or muddying the clinical narrative with non-clinician language. Research published in the npj Digital Medicine journal has documented that speaker misattribution in clinical AI transcription is a leading cause of note inaccuracy in multi-party encounters.

Step 3: Non-Verbalized Reasoning Detection and Structured Prompting

Before note finalization, Scribing.io's context engine compares the ambient transcript against the pre-retrieved FHIR data and detects that the clinician did not verbally reference:

• The A1c value or its trend relative to prior results

• The eGFR value and its CKD staging implication

• The medication-risk assessment (SGLT2 inhibitor use in CKD 3b, ACE inhibitor with declining renal function)

• The continuity-of-care framing required for G2211

The system presents a structured confirmation screen:

"The following data points were retrieved from the patient record but not referenced during the encounter. Confirm to include in MDM documentation:"

• ☑ A1c 8.4% (2026-04-12) — above target, suboptimal glycemic control

• ☑ eGFR 41 (2026-04-12) — CKD 3b, relevant to medication dosing thresholds

• ☑ Medication risk: empagliflozin + lisinopril in setting of eGFR <45

• ☑ Continuity of care: established longitudinal relationship, ongoing management of 2+ chronic conditions

The clinician confirms with a single tap. No dictation. No after-hours note completion. No memory-dependent reconstruction of clinical reasoning that occurred hours earlier.

Step 4: MDM Draft Generation with E/M and G2211 Compliance Language

Scribing.io generates MDM language that explicitly satisfies CMS documentation requirements for high-complexity decision-making:

"Data reviewed: Most recent HbA1c 8.4% (2026-04-12) remains above goal of <7.0%, consistent with suboptimal glycemic control in the setting of Type 2 diabetes mellitus. eGFR 41 mL/min/1.73m² (2026-04-12) confirms CKD Stage 3b; renal function decline considered in medication management decisions. Risk assessment: Current regimen includes empagliflozin and lisinopril; medication risk is moderate-to-high given declining renal function and dual nephroactive therapy. Per KDIGO 2024 guidelines, SGLT2 inhibitor continuation is appropriate with eGFR monitoring. This encounter reflects ongoing management of a longitudinal patient relationship involving 2+ chronic conditions requiring continuous medical decision-making (G2211)."

This language supports:

• High-complexity MDM (99215) through documented data review, medication risk, and problem complexity

• G2211 add-on eligibility through explicit continuity-of-care language and longitudinal problem linkage

*Revenue differentials based on CMS CY2026 Physician Fee Schedule national rates. Actual reimbursement varies by payer, geographic locality, and facility/non-facility setting.

See a live run of our FHIR-aware E/M + G2211 eligibility engine against your Epic/Cerner sandbox—watch us pull last A1c via Observation/$lastn and auto-draft compliant MDM in under 30 seconds. Schedule a demo →

FHIR Architecture: Observation/$lastn, DiagnosticReport Fallback, and LOINC Normalization

Scribing.io's data retrieval layer is not a proprietary black box. It operates on standards-based FHIR R4 APIs conformant to the USCDI v3 data class requirements now mandated under the ONC 21st Century Cures Act final rule.

The Observation/$lastn Query Pattern

The FHIR Observation/$lastn operation retrieves the most recent n observations for a given patient, filtered by category and code. Scribing.io executes this at encounter launch with LOINC-normalized parameters:

• GET [base]/Observation/$lastn?patient=[id]&category=laboratory&code=4548-4,48642-3,2093-3&max=3

• This returns the three most recent values for A1c (4548-4), eGFR (48642-3), and total cholesterol (2093-3)—the high-frequency labs relevant to chronic disease management in primary care.

LOINC normalization is critical because different lab systems may report the same analyte under variant codes. Scribing.io maintains a curated LOINC synonym map that resolves common variants (e.g., 17856-6 for A1c panel vs. 4548-4 for A1c by HPLC) to ensure retrieval completeness.

DiagnosticReport Fallback

When a FHIR server returns labs only as narrative DiagnosticReport resources—common in health systems using legacy lab interfaces—Scribing.io parses the presentedForm and conclusion fields using structured extraction to surface discrete values. This fallback ensures that practices on older interfaces are not excluded from context-aware documentation.

MedicationRequest and Condition Resource Retrieval

Active medications are retrieved via MedicationRequest?patient=[id]&status=active, and active conditions via Condition?patient=[id]&clinical-status=active. These resources form the basis for medication risk assessment and problem list mapping that drive MDM complexity scoring.

Technical Reference: ICD-10 Documentation Standards

Accurate ICD-10 coding is inseparable from accurate clinical documentation. Two of the most frequently assigned codes in primary care illustrate how documentation specificity directly impacts reimbursement and compliance:

I10 — Essential (Primary) Hypertension and E11.9 — Type 2 Diabetes Mellitus Without Complications

I10 — Essential (primary) hypertension; E11.9 — Type 2 diabetes mellitus without complications

These codes are among the highest-volume diagnoses in Family Medicine, yet they are also the most frequently undercoded. The problem is specificity:

• E11.9 (T2DM without complications) is often assigned by default when the documentation does not explicitly name a complication—even when the patient clearly has one. A patient with A1c 8.4% and CKD 3b should not be coded E11.9. The correct code is E11.65 (T2DM with hyperglycemia) with a secondary code of N18.3b (CKD Stage 3b) or, if the diabetes-CKD causal link is documented, E11.22 (T2DM with diabetic chronic kidney disease). The difference affects CMS Hierarchical Condition Category (HCC) risk adjustment scores, RAF values, and downstream reimbursement for Medicare Advantage populations.

• I10 (Essential hypertension) requires no additional specificity under ICD-10-CM—it is the terminal code. However, documentation must distinguish essential hypertension from secondary causes (I15.x) and from hypertensive crisis (I16.x). In the context of CKD, the documentation must clarify whether hypertensive nephrosclerosis is present (I12.9 with N18.x) or whether the hypertension and CKD are coincident without causal linkage.

Scribing.io ensures maximum specificity through its context-aware architecture:

1. Active Condition cross-referencing: When the pre-encounter FHIR query returns both E11.x and N18.x conditions with concurrent lab evidence (A1c above target, eGFR below 60), Scribing.io flags the note draft if the documentation language defaults to non-specific codes.

2. Complication linkage prompting: The system prompts the clinician to confirm or deny the causal relationship between diabetes and CKD—a documentation decision that determines whether E11.22 vs. E11.9 + N18.3b is appropriate. Per ICD-10-CM Official Guidelines for Coding and Reporting Section I.C.4.a.1, if the provider documents a causal link, both the diabetes complication code and the CKD code must be assigned.

3. HCC impact alerting: For organizations managing Medicare Advantage populations, Scribing.io surfaces the RAF impact of code specificity changes, ensuring that the documentation accurately reflects clinical reality—not upcoding, but right-coding based on the evidence in the chart.

G2211 Add-On Eligibility: Documentation Requirements and Scribing.io's Role

CMS add-on code G2211 compensates for the inherent complexity of evaluation and management visits that involve ongoing care of a patient with a serious or complex condition. The code became effective January 1, 2024, and its documentation requirements have been further clarified in the CY2026 final rule.

G2211 requires documentation that the visit involves:

1. A longitudinal relationship with the patient (not a new patient or one-time consultation)

2. Ongoing medical decision-making related to a condition that is serious, complex, or requires continuous management

3. Evidence that the visit's complexity goes beyond what is captured in the base E/M code

Scribing.io's pre-encounter Condition resource retrieval and longitudinal encounter history automatically identify G2211-eligible visits. The MDM draft includes explicit language such as:

"This encounter represents an ongoing longitudinal relationship with the patient for the management of [specific chronic conditions]. The medical decision-making complexity of this visit reflects the continuous evaluation and adjustment of a multi-drug regimen in the context of interacting comorbidities."

This language is not templated boilerplate. It is dynamically generated from the patient's actual Condition list, medication profile, and encounter history—making it defensible under audit because every assertion maps to a verifiable FHIR resource.

AMA Augmented Intelligence Compliance and Physician Oversight

The AMA's principles for augmented intelligence require that AI systems in healthcare:

• Operate transparently, with explainable logic

• Preserve physician authority over final clinical decisions

• Avoid exacerbating bias or inequity

• Protect patient privacy

• Demonstrate evidence-based accuracy

Scribing.io's architecture satisfies each principle structurally:

A 2025 JAMA perspective on AI documentation tools noted that the primary risk of ambient AI in clinical settings is not inaccuracy of transcription but incompleteness of clinical reasoning capture. Scribing.io is the only platform architecturally designed to address that specific risk.

Deployment Operations: Epic, Cerner, and Sandbox Validation

Scribing.io deploys through certified SMART on FHIR app frameworks available in both Epic App Orchard and Oracle Health (Cerner) App Gallery. Deployment follows a standardized four-phase process:

Total time from contract to full deployment: 6–11 weeks, depending on EHR environment complexity and IT governance requirements.

See a live run of our FHIR-aware E/M + G2211 eligibility engine against your Epic/Cerner sandbox—watch us pull last A1c via Observation/$lastn and auto-draft compliant MDM in under 30 seconds. Request sandbox access →

For CMIOs evaluating AI documentation platforms: the question is no longer whether ambient AI can transcribe an encounter. Every vendor does that. The question is whether the platform can document the clinical reasoning that justifies the level of care delivered. Scribing.io is the answer to that question—not through marketing claims, but through verifiable, standards-based, physician-confirmed documentation that holds up under payer scrutiny, audit review, and the operational demands of modern medicine.