Prior Authorization AI: Evidence, Adoption, and Regulatory Landscape

The Prior Authorization Burden: Why AI Entered the Picture

Prior authorization (PA) has long been one of the most frictional points in the US healthcare system. The administrative overhead it imposes on clinicians is staggering. According to the American Medical Association's 2024 survey of 1,000 practicing physicians, physicians and their staff spend an average of 13 hours per week on PA-related tasks. Forty percent of physicians employ staff whose primary job is prior authorization. The clinical consequences are equally severe: 94% of physicians reported that prior authorization had a negative impact on clinical outcomes, 93% said it delays care, and 29% reported that it led to a serious adverse event for a patient. Twenty-three percent said prior authorization led to hospitalization, and 8% reported it led to disability, permanent damage, or death.

The scale of the problem has made it a prime target for automation. Health insurers process millions of PA requests annually, and the manual review of clinical documentation is both labor-intensive and slow. AI—particularly natural language processing (NLP) and machine learning models trained on historical claims and clinical data—promises to ingest documents, extract relevant clinical information, and render or recommend coverage decisions in minutes rather than days. But the entry of AI into this domain has created a polarized landscape: efficiency gains for some, algorithmic denials for others, and a growing regulatory scramble to govern a technology that is already deeply embedded in the utilization management infrastructure.

How AI Is Used in Prior Authorization: Payer vs. Provider Tools

AI in prior authorization operates on two sides of the same transaction, and the strategic incentives on each side are fundamentally different. On the payer side, AI tools are deployed to automate the review of incoming PA requests—approving those that meet established criteria and flagging or denying those that do not. On the provider side, AI tools help clinicians and billing staff prepare, submit, and appeal PA requests more efficiently. The result is an emerging technological arms race in which each side deploys AI to counter the other's moves.

The distinction matters because the same technology can produce very different outcomes depending on who deploys it. A payer using AI to auto-approve routine requests can reduce turnaround times and lower administrative costs. But a payer using AI to systematically flag and deny marginal requests—especially if the model is trained on historical denial patterns that may encode bias—can increase denial rates and create new administrative burdens for providers. The Health Affairs analysis notes that in the large-employer group market, 70% of insurers are using or exploring AI for prior authorization, and the NAIC 2025 survey of 93 insurers found that 68% use AI for PA approvals while 12% use it for PA denials.

For a detailed profile of one of the leading payer-side platforms, see our Cohere Health company profile, which covers its Clinical Intelligence tool for auto-approving routine musculoskeletal and cardiology requests.

Adoption Benchmarks: Insurers and Health Systems

AI adoption in prior authorization has moved from experimental to mainstream in a remarkably short period. The most frequently cited benchmark comes from the National Association of Insurance Commissioners (NAIC) 2025 AI/ML Survey Report, which surveyed 93 insurance companies across 16 states. The headline finding: 84% of insurers reported using AI/ML across product lines. For individual major medical plans specifically, the survey found that 71% of companies are currently using or exploring AI for utilization management, 68% for prior authorization approvals, and 12% for prior authorization denials.