Integration of Artificial Intelligence Into Extended Reality Debriefing in Healthcare Simulation: A Narrative Review

Extended reality (XR) is increasingly used in healthcare simulation; however, debriefing, the phase in which performance is translated into learning through guided reflection, faces distinct challenges in immersive, data-rich, and sometimes asynchronous environments. This focused narrative review examined how artificial intelligence (AI) can augment debriefing within XR/virtual reality (VR) simulation. Peer-reviewed literature published between 2015 and 2025 was identified through structured searches of PubMed/MEDLINE and Google Scholar, supplemented by reference screening. Search strategies combined terms related to XR (eg, “extended reality”, “virtual reality”, “augmented reality”, “mixed reality”), simulation and education, and AI-enabled debriefing (eg, “debriefing”, “feedback”, “learning analytics”, “natural language processing”, “large language models”, “conversational agents”). Studies were included when AI was directly applied to debriefing processes or generated debrief-relevant feedback within XR/VR healthcare simulations; studies focused solely on scenario generation, automated grading without reflective components, or non-XR contexts were excluded. Four recurring AI functions were identified: (1) automated performance analytics that convert XR telemetry (eg, timestamps, trajectories, error logs, and in some systems gaze or physiological data) into structured metrics to support procedural feedback; (2) natural language processing to enable transcription and discourse analysis that can surface communication patterns and candidate moments for team reflection; (3) conversational agents and large language model–enabled systems that scaffold reflective dialogue and summarize performance; and (4) multimodal fusion approaches that integrate action, speech, gaze, and physiological signals to deliver adaptive feedback. The evidence base remains dominated by feasibility studies, pilots, and prototypes, with limited controlled comparisons, psychometric validation, or evidence of sustained behavior change or clinical transfer. A pragmatic near-term approach is a hybrid model in which AI prepares transparent debriefing artefacts, while human facilitators retain responsibility for psychological safety, meaning-making, and high-stakes interpretation. Future research should prioritize validation, bias and privacy safeguards, faculty development, and longitudinal educational outcomes.