Exploring the Design Space of AI Simulation-Based Communication Training: A Qualitative Study of Online Peer Counselor Preferences

Abstract

A growing body of work uses AI-based simulation to teach communication skills, but current simulation-based communication training (SBCT) systems make inconsistent choices on design dimensions that prior theory identifies as critical: how skills are modeled, when feedback is given, and how self-reflection is scaffolded. These choices are especially consequential in domains like peer counseling, where trainees navigate emotional ambiguity and must anticipate how their words will land. Through prototype-driven and semi-structured interviews with 16 online peer counselors, we explore this design space of SBCT systems by eliciting reactions to 10 prototypes varying across the three dimensions. We find that participants wanted support throughout the process of composing responses (interpreting support-seekers’ messages, formulating goals, and anticipating consequences), not solely evaluation of finished messages. Participants also saw feedback timing as instilling mindsets about real-world consequences in sensitive communication, such as signaling proactive prevention of harm through feedback before sending messages. Participants described and anticipated different needs for novice and experienced trainees, suggesting a progression of training environments that adjusts simulation realism and scaffolding to experience level. We conclude with design implications for AI SBCT and peer counseling training, including a tension between users’ desire for validation and the need for ambiguity tolerance and authentic communication.