Artificial intelligence systems have become a dominant resource for healthcare information and administrative support, with over 40 million people globally accessing ChatGPT daily for health-related queries, according to analysis released by OpenAI. This scale of adoption reflects a fundamental shift in how patients navigate healthcare systems, particularly regarding insurance questions, after-hours guidance, and administrative tasks where traditional healthcare access remains constrained.
The research reveals that one in four users submits healthcare-related questions to the platform on a weekly basis, indicating sustained engagement rather than isolated incident-driven consultation. This widespread adoption raises significant implications for healthcare equity, clinical safety, and organizational responsibility around AI-generated health information accuracy and appropriateness.
OpenAI has simultaneously released HealthBench, a new evaluation framework designed to assess large language model capabilities in healthcare contexts and identify areas where AI systems require improvement before broader clinical adoption.
Healthcare Access Barriers Drive After-Hours AI Consultation
Approximately 70 percent of health-related interactions on ChatGPT occur outside standard clinic hours—evenings, nights, and weekends—indicating that patients are using AI systems when traditional healthcare access is unavailable. This timing pattern reflects acute constraints in healthcare availability and suggests that AI fills a genuine gap in how patients manage health concerns when providers are inaccessible.
Rural and underserved areas generate particularly elevated volumes of health-related AI interactions, with users in rural regions sending close to 600,000 healthcare messages weekly on average. In areas classified as “hospital deserts”—locations exceeding 30 minutes from the nearest general hospital—users submitted over 580,000 health-related messages weekly during a four-week observation period in late 2025.
States including Wyoming, Oregon, and Montana ranked highest by share of AI health-related interactions, correlating directly with geographic healthcare workforce shortages and limited provider availability. This geographic pattern indicates that AI serves an access-bridging function for populations facing systematic healthcare availability constraints, though this reliance also raises questions about healthcare equity and whether AI-mediated guidance represents adequate substitute for evidence-based clinical care.
Insurance Navigation and Administrative Support Drive Utilization
A substantial portion of health-related AI interactions focus on non-clinical administrative tasks rather than medical diagnosis. Organizations estimate that 1.6 to 1.9 million weekly messages address health insurance questions, including plan comparisons, billing disputes, claims processes, eligibility determination, and cost-sharing clarification.
Users predominantly seek assistance organizing insurance documentation, understanding terminology, and preparing required paperwork rather than requesting medical diagnosis. This administrative focus reflects widespread consumer confusion regarding healthcare insurance systems and suggests that clear, accessible information about plan structures and claims processes remains inadequate across traditional provider and insurer communication channels.
Healthcare professionals themselves have adopted AI tools for administrative purposes at substantial scale, with 66 percent of US physicians reporting AI usage in 2024 compared to 38 percent in 2023, and nearly half of US nurses reporting weekly AI usage—primarily for documentation, billing support, and workflow optimization rather than clinical decision-making.
Clinical Safety and Governance Framework Concerns
The expansion of AI health guidance beyond administrative support into clinical territory has prompted development of governance frameworks and accuracy evaluation. OpenAI’s HealthBench benchmark assesses AI system capabilities across seven key healthcare domains including emergency care, uncertainty management, and global health considerations. The framework acknowledges that medical information accuracy and appropriate guidance boundaries remain critical gaps requiring ongoing evaluation and improvement.
Clinicians and privacy experts have cautioned that generative AI systems can produce inaccurate responses, are not subject to HIPAA privacy protections, and should not be treated as clinical guidance substitutes, particularly when patients upload medical records or imaging for AI interpretation.
