Conversational Artificial Intelligence in Medical Education: A Scoping Review

Conversational artificial intelligence (AI), encompassing chatbots and large language models (LLMs), is rapidly emerging in the sphere of medical education as a dynamic tool for interactive learning. By generating realistic dialogue, simulating patient encounters, and providing adaptive feedback, these systems create new possibilities for learners to practise communication, clinical reasoning, and decision-making in a flexible and accessible way. Yet, despite increasing enthusiasm, evidence regarding their educational value remains fragmented. This scoping review examines the existing literature on conversational AI in undergraduate medical education, focusing on three key domains: educational utility, technology usability, and fidelity. A comprehensive search was conducted across PubMed, Scopus, and Web of Science in August 2025. After deduplication, 496 unique studies were screened, and 20 met the inclusion criteria. These studies employed diverse methodologies and evaluation approaches. Methodological rigour was assessed using a validated framework designed for medical education research.  Across the literature, conversational AI demonstrates considerable potential to enhance engagement, support self-directed learning, and expand access to experiential practice. Learners generally view these systems as intuitive and motivating, and many studies suggest benefits for clinical reasoning and communication training. However, limitations in reliability, realism, and technical accuracy persist, and outcome measures remain inconsistent. Few studies assess the impact of response latency or the long-term transfer of skills to clinical settings, and methodological rigour is often modest. Overall, conversational AI appears to be a promising adjunct to traditional medical teaching rather than a replacement. Its value lies in scalability, interactivity, and adaptability, but effective integration requires thoughtful design, validated evaluation frameworks, and ongoing human oversight. As technology advances, further research should focus on standardising assessment methods, exploring learning outcomes beyond user satisfaction, and addressing fidelity and responsiveness to ensure meaningful, safe, and sustainable implementation within modern medical curricula.