Artificial Intelligence (“AI”) is increasingly moving beyond clinical decision support and into the administrative functions that directly impact healthcare reimbursement. In a July 16, 2026 Science & Tech Spotlight, the U.S. Government Accountability Office (“GAO”) examined the growing use of AI tools for clinical documentation and medical coding, and identified accuracy and reimbursement as key areas of concern.[1]
AI Is Becoming Part of the Documentation and Coding Process
AI “scribes” can now record conversations between clinicians and patients and generate draft clinical notes for clinician review. GAO refers to this as “ambient” listening, because the technology can operate in the background during a patient encounter, potentially without the patient’s knowledge. AI coding tools can then analyze medical records and suggest billing codes, with increasingly limited human involvement. GAO notes that use of AI for documentation and coding is growing. According to an American Medical Association survey cited by GAO, the percentage of surveyed clinicians using AI for these purposes increased from 21% in 2024 to 28% in 2026.
These tools offer practical advantages. GAO points to reduced administrative burden, more detailed clinical documentation, and greater operational efficiency as among the potential benefits.
Accuracy and Reimbursement Present Compliance Concerns
The same technology can also create potential compliance risks. GAO found that there are relatively few independent studies evaluating the accuracy of AI documentation and coding tools. According to the GAO, inaccurate outputs may impact patient care and can also result in either over- or under-reimbursement.
GAO also observed that AI tools could increase reimbursement by identifying additional diagnoses or services that might not otherwise be captured. While documentation may support additional reimbursement identified by automated tools, using these tools also raises the question of how providers are verifying that AI-generated documentation and codes accurately reflect the services actually provided.
Further, ensuring that AI-generated documentation and codes accurately reflect the services actually provided is particularly critical when that documentation or coding supports claims submitted to Medicare, Medicaid, or other payors, given the potential for overpayment liability and other penalties if the claim is later found unsupported. Regardless of how a code or clinical note is generated, providers remain responsible for ensuring that submitted claims are adequately supported and accurate.
What Should Providers Be Thinking About?
For providers, the growth of AI-assisted documentation and coding raises several compliance considerations, including:
- Preserving clinical judgment: AI-generated coding recommendations should not be a substitute for a clinician’s independent judgment regarding diagnosis, severity, level of service, or medical necessity.
- Coding and documentation accuracy: Providers should maintain review controls to ensure AI-generated documentation and codes accurately reflect the services furnished and are supported by the medical record.
- Payor-specific requirements: AI-assisted workflows should be evaluated against applicable Medicare Administrative Contractor and other payor-specific documentation, coding, coverage, and authentication requirements.
- False Claims Act and overpayment risk: Claims based on AI-generated documentation or codes that do not accurately reflect the services provided may give rise to overpayment liability and, depending on scienter and the circumstances, potential False Claims Act exposure.
- Privacy and security: Ambient-listening tools may increase the volume of patient information recorded, transmitted, or stored, creating additional data security risks. Providers should also consider applicable patient notice and consent requirements, which may vary on a state-by-state basis before recording patient encounters.
Responsible use of AI-assisted documentation and coding should include clear written policies, training, appropriate human oversight, validation procedures, and privacy and security safeguards designed to address the risks associated with these tools.
FOOTNOTES