AI is becoming part of healthcare documentation, coding, billing, risk adjustment, claims review, and revenue-cycle workflows.
The practical question isn’t whether AI is simply “good” or “bad.”
It’s this: Can your team explain, verify, and defend every AI-influenced coding or billing decision?
Federal guidance and oversight make that question especially relevant for Medicare Advantage organizations. But the answer still depends on the specific workflow, the applicable rules, the record, and the decision being made.
An AI suggestion isn’t evidence by itself. And a mistake isn’t automatically fraud.
What Oversight Means for AI-Assisted Work
On February 3, 2026, the U.S. Department of Health and Human Services Office of Inspector General (HHS-OIG) published its Medicare Advantage Industry Segment-Specific Compliance Program Guidance.
The guidance is voluntary and nonbinding. It does not create new legal requirements.
It does, however, describe risk areas that Medicare Advantage organizations should take seriously.
In the risk-adjustment section, OIG identifies federal investigations involving electronic medical record prompts, including AI-generated prompts, that encouraged providers to add diagnoses patients did not have or that were not relevant to their care, treatment, or management.
That is not a finding that all prompts are improper or that all AI-assisted coding is unsafe.
It is a warning about how a prompt is designed, what it encourages, and whether the resulting diagnosis is supported.
OIG also recommends that Medicare Advantage organizations oversee query and prompt use, review relevant systems and vendor tools, and monitor high-risk coding patterns.
A pattern that stands out is a reason to investigate. It is not automatically evidence of wrongdoing.
CMS’s Risk Adjustment Data Validation (RADV) program provides another reason to pay attention to documentation. Through RADV audits, CMS reviews whether diagnoses submitted for Medicare Advantage risk-adjustment payments are supported by medical records.
The applicable audit year and program instructions matter.
A coding error may be an honest mistake. A compliance concern may arise when an organization has a process that repeatedly produces unsupported diagnoses. Knowingly submitting inaccurate diagnoses is a different and more serious matter.
Those distinctions are important.
An AI Prompt Is Not Always a Provider Query
The AHIMA/ACDIS Guidelines for Achieving a Compliant Query Practice, updated August 20, 2026, address the growing use of technology in documentation clarification.
One important principle remains unchanged:
A query must meet appropriate compliance standards regardless of whether a person or technology generates it.
But that doesn’t mean every AI alert is a provider query.
Consider three different situations.
An automated documentation prompt
A system may flag a possible documentation gap or suggest information for further review. Whether the prompt functions as a provider query depends on what it communicates, who receives it, and whether it presents a diagnosis or documentation option for a particular encounter.
A provider query
A query asks a provider to clarify or expand the medical record. When a technology-generated prompt presents a provider with a specific diagnosis or documentation option for a specific encounter, it can function as a query and must meet applicable compliant-query standards.
A coder-facing AI suggestion
A coding application may suggest a diagnosis or code to the person reviewing the chart. That recommendation is not automatically a provider query because it may never ask the provider to clarify the documentation.
This distinction matters.
When an AI-generated prompt functions as a provider query, it should include relevant clinical indicators, use neutral and nonleading language, avoid references to reimbursement, and preserve the provider’s independent clinical judgment.
Organizations should also involve appropriate coding, clinical documentation integrity, and compliance professionals in evaluating these tools.
A vendor calling something a “smart suggestion” does not establish that the suggestion is compliant.
The CKD Stage 3 Question: What Would You Do?
The following example is fictional and intended for education.
Imagine you’re reviewing a chart and an AI coding tool suggests chronic kidney disease (CKD), stage 3.
The suggestion appears to come from information documented during an earlier encounter.
But when you review the current signed note, you don’t find clear documentation establishing CKD stage 3 for the encounter.
Would you accept the recommendation?
This is where professional judgment becomes important.
Start with the actual medical record, not the AI-generated summary or suggested code.
Review what the provider documented. Determine whether the diagnosis is supported under the coding guidelines applicable to that setting and date of service.
A historical diagnosis may provide useful context. Laboratory results may raise clinical questions. Neither automatically authorizes the coder to establish or report a diagnosis independently.
Under the ICD-10-CM Official Guidelines, code assignment generally relies on the responsible provider’s diagnostic statement, subject to applicable coding conventions and exceptions.
The current record must support the code assigned.
If the provider documents CKD but the stage is unclear or documentation conflicts, an appropriate, compliant query may be warranted.
If the diagnosis is absent, however, the AI recommendation does not create it.
The next action depends on the circumstances and organizational procedures. It may mean not accepting the suggestion, reviewing additional relevant documentation, seeking appropriate clarification, or escalating the issue.
A provider query should not be used simply to obtain the diagnosis a technology system suggested.
For Medicare Advantage risk adjustment, there is an additional consideration.
The diagnosis must also meet applicable program requirements, including requirements concerning acceptable data sources and qualifying encounters.
Certain interactive audio-video telehealth encounters can satisfy the face-to-face requirement. Audio-only encounters should not automatically be treated as qualifying risk-adjustment encounters.
The correct determination depends on the applicable program requirements and payment year.
These Medicare Advantage requirements should not be confused with universal rules governing every coding setting.
What if the same problem keeps appearing?
Imagine that several charts contain similar AI-generated CKD recommendations originating from historical information.
An individual coder might identify the unsupported recommendations and follow the established review process.
A supervisor, coding auditor, or compliance professional may then need to determine whether the pattern reflects a documentation issue, system configuration problem, insufficient training, or another cause.
One questionable suggestion does not establish fraud.
But repeated unsupported suggestions deserve appropriate investigation.
Where AI Can Assist—and What Still Needs Review
AI and other automation tools may help organize information, identify possible code options, flag documentation inconsistencies, support internal audits, or assist with professional education.
Those uses can be valuable.
But a generated summary is not the original medical record.
A suggested code is not a completed coding determination.
And an unusual billing pattern is not automatically proof of improper conduct.
Before accepting an AI-influenced recommendation, professionals should examine several questions:
The source record: What does the original documentation actually say?
The diagnosis: Did the appropriate provider document the condition?
The encounter: Was the service performed, and does the record support the applicable requirements?
The coding decision: Are the code, specificity, modifiers, and relevant payer rules correct for the setting and date of service?
The missing information: Are there contradictions, omissions, or uncertainties that require further review?
The reasoning: Could another qualified professional understand how the final decision was reached?
Coding requirements also change over time.
The FY 2026 ICD-10-CM guidelines apply to services from October 1, 2025, through September 30, 2026.
For services beginning October 1, 2026, the FY 2027 ICD-10-CM Official Guidelines apply.
Professionals must use the code set and guidance appropriate to the encounter date rather than assuming a system’s information is current.
Build Controls Around the Entire Workflow
The HHS-OIG Medicare Advantage guidance and its General Compliance Program Guidance offer voluntary recommendations for managing compliance risks.
Those recommendations should not be confused with binding regulations.
Medicare Advantage organizations do have regulatory obligations concerning effective compliance programs and the accuracy of risk-adjustment information.
For organizations using AI-assisted coding or billing technology, practical controls may include the following.
Know which tools are being used.
Identify applications that influence documentation, coding, billing, claims editing, risk adjustment, denials, or compliance monitoring.
Include general-purpose AI applications that staff may use outside formally approved workflows.
Define appropriate use.
Establish written guidance explaining what employees may enter into AI tools, which systems are approved, how sensitive information must be protected, and what requires independent verification.
Include AI-assisted work in quality reviews.
Use appropriate risk-based auditing and monitoring. Review the underlying documentation rather than relying exclusively on the final code or system-generated explanation.
Investigate the causes of errors.
An error could result from documentation, user misunderstanding, outdated references, system configuration, templates, or vendor performance.
Correcting an individual claim matters. Understanding why the error occurred may help prevent it from recurring.
Train people to make decisions.
Employees need more than familiarity with software.
They need to recognize unsupported recommendations, locate reliable evidence, identify missing information, explain their reasoning, and know when to involve another qualified professional.
Tool familiarity is not the same as workplace readiness.
A Safe Process Is Built, Not Assumed
AI in medical coding is not automatically safe or unsafe.
Its impact depends on the documentation, the technology, the workflow, organizational controls, and the judgment applied to its output.
An AI recommendation may be correct.
It may be incomplete.
It may identify something that needs further investigation.
Or it may be wrong.
The professional responsibility is to understand the difference and take the action supported by the evidence.
That is part of what Queens Code Healthcare means by Think Like an Auditor™.
Review the information. Verify the evidence. Decide what the record supports. Escalate when appropriate. Explain the reasoning.
Preparing Coding and Billing Professionals for Today’s Workplace
Queens Code Healthcare focuses on the gap between learning coding and billing concepts and applying them in actual workplace situations.
Traditional coding knowledge remains essential.
But professionals also need opportunities to practice reviewing technology-generated recommendations, identifying documentation problems, handling exceptions, and explaining decisions.
Our workforce-readiness approach is designed to complement established medical coding and billing education—not replace it.
We want learners to understand not only what answer they selected, but why the evidence supports it and what could happen if the decision is wrong.
For community colleges, workforce development organizations, and healthcare employers interested in applied workforce-readiness training, visit our Medical Coding & Billing Workforce Readiness program.
AI assists. Evidence rules. Humans decide.
Educational disclaimer: This article is intended for professional education and general information. It is not legal advice or a substitute for organization-specific compliance review. Requirements vary by setting, payer, program, and date of service. Refer to current authoritative guidance and organizational policies when making actual coding, billing, or compliance decisions.
