Simulating Adolescents Using a Generative Text-Based AI Chatbot to Support Facilitator Training in a Motivational Interviewing-Based Prevention Program

Implementation fidelity is essential for preventive interventions to produce their intended effects, yet it is often difficult to achieve in real-world settings. Achieving fidelity depends in part on how well facilitators are trained; however, opportunities to practice program delivery skills are often limited by time, cost and staff turnover. Generative artificial intelligence offers a promising approach to simulating participant interactions for training purposes; however, evidence regarding the feasibility and performance of such simulations remains limited. In this formative feasibility study, we developed and iteratively refined a text-based generative AI chatbot designed to simulate adolescent participants in a brief motivational interviewing-based substance use prevention program. We evaluated the chatbot's feasibility, acceptability, and perceived utility for facilitator practice. Five program-certified facilitators each completed two text-based simulated sessions with the chatbot, resulting in a total of 374 distinct messages. Chatbot responses were rated independently by three trained coders on coherence, realism, and contextual appropriateness, and semi-structured interviews were conducted with facilitators following each session to assess their experiences interacting with the chatbot. Coder ratings indicated high mean message-quality scores across sessions, suggesting that the chatbot consistently generated coherent and context-appropriate adolescent-like responses. Qualitative findings suggested that facilitators perceived the chatbot as realistic, engaging, and useful for practicing motivational interviewing skills, particularly for navigating adolescent ambivalence and resistance. Facilitators also reported reduced performance anxiety relative to traditional role-play, while noting limitations related to the absence of nonverbal cues. These findings suggest that a text-based generative AI chatbot simulating adolescent participants could be a useful tool to enhance facilitator training in motivational interviewing-based prevention programs. Future research should examine whether such tools improve training effectiveness, facilitator skill acquisition, and implementation fidelity.