High Use, Low Preparation: A Needs Assessment to Inform Curriculum Development for AI in Residency and Fellowship Training

Generative artificial intelligence (Gen-AI) has rapidly entered clinical environments, posing a risk of "cognitive offloading" that may impede the development of independent clinical judgment. This study conducted a cross-sectional survey of residents, fellows, and faculty at a large academic medical center in January 2026 to quantify Gen-AI utilization and perceptions in physical medicine and rehabilitation (PM&R) to inform curriculum development. Thirty-six trainees (95% response rate) and 11 faculty members (30% response rate) completed the survey; 43 (91.5%) reported any Gen-AI use and formed the analytic sample. Among users, 72.1% used Gen-AI daily for clinical learning and 51.2% during patient encounters, yet 67.4% were self-taught and only 7.0% always requested citations. Trainees requested citations more often than faculty (P=.048), while faculty more often relied on clinical judgment without additional verification (62.5% vs. 25.7%, P=.089). Gen-AI-derived evidence was rarely surfaced in teaching: 60.5% discussed it during case presentations rarely or never, and only 16.2% always disclosed its origin. Residents are integrating Gen-AI into clinical workflows at rates that outpace faculty familiarity, while verification remains largely undiscussed and untaught. These gaps highlight an urgent need for structured Gen-AI curricula in PM&R residency.