Beyond Generic Answers and Empty Praise: The Differential Impact of AI Companionship Roles on Social-Emotional Learning in University Student Intervention

Social-emotional learning (SEL) is vital for university students’ development, yet access to scalable support remains limited. While conversational AI presents a promising avenue, existing research often treats “AI support” as a monolithic intervention, overlooking critical distinctions between designs aimed at providing emotional comfort and those fostering competency growth. This study directly investigates how distinct AI companionship strategies-Affirmation, Empathic-Inquisitive, and General Support—differentially impact SEL outcomes. Through a 14-day longitudinal experiment with 134 university students, using systematically prompted Large Language Models (LLMs), we demonstrate a clear functional dissociation: the Affirmative strategy most effectively enhanced positive affect; the Empathic-Inquisitive strategy produced the greatest reduction in negative affect and the most substantial gains in social-emotional competencies; while the General Support strategy offered balanced, moderate benefits. These findings underscore that AI companionship is not pedagogically neutral—its effects are specialized and contingent on design. Our work shifts the focus from whether AI can support SEL to how its design directs learning toward specific affective or competency-based outcomes. This provides a principled framework for developing goal-aligned AI systems in higher education and calls for future research into adaptive, context-sensitive companions that can dynamically respond to learners’ evolving needs.