How to Be a Better Boyfriend: A Comparative Analysis of AI and Human Relationship Advice

Abstract

The growing use of LLMs in personal and social contexts has led to questions about how AI systems navigate subjective, emotionally-sensitive contexts. Reports have revealed that users have been increasingly turning towards AI systems for support in daily life, particularly in everyday advisory settings. AI’s ability to output relationship advice has largely been unexplored within these settings, especially from a strategic behavioral standpoint. In this study, we evaluate AI-generated advice and conduct a comparative analysis with human-authored advice. We collected real- world human relationship conflict scenarios including top-voted human comments from Reddit. Furthermore, we developed an evaluation framework based on the Thomas-Kilmann Conflict Mode Instrument (TKI), a validated psychological model which outlines five behavioral strategies in interpersonal conflict management. Our results reveal that AI advice is overwhelmingly concentrated in the Collaborating mode, whereas human advice is more widely distributed, with Competing and Avoiding as the dominant strategies. These findings suggest that alignment training systematically shapes AI conflict-handling behavior in ways that substantially diverge from naturally occurring human advice, with implications for users, designers, and researchers studying LLMs in advisory contexts.