Abstract for: Modeling trust dynamics in conversational AI

Trust plays a central role in how users adopt and rely on conversational AI systems. Because existing approaches often conceptualize trust as a stable outcome or direct response to specific system cues, they struggle to explain these longer-term patterns of change. We conceptualize trust in conversational AI as a dynamic, self-regulating process shaped by repeated interaction and feedback mechanisms. Building on a system dynamics perspective, we propose a conceptual model grounded in the Limits to Success archetype and implement it as a computer simulation. Trust follows characteristic rise-and-plateau trajectories, showing how early increases in trust can be driven by reinforcing feedback between perceived performance, affective responses, and usage, while longer-term stabilization or decline can emerge from balancing processes related to emotional habituation, relational distance, and bias-related constraints. These dynamics can emerge without assuming changes in system performance. By offering a process-oriented and mechanism-based account of trust development, this work advances trust research beyond static approaches and highlights trust calibration as a dynamic, ongoing process, with clear design implications.