Dynamic In-Group Personas Boost Human-AI Rapport

Establishing strong human-AI rapport is crucial for chatbots operating in sensitive interpersonal domains like counseling or peer support, yet it remains a significant challenge. This research introduces an innovative method that conditions large language models (LLMs) with dynamic in-group personas to improve this rapport. The approach involves two main steps: first, identifying a user's primary concern and a brief personal context, such as a student worried about career prospects. Second, the system generates a synthetic in-group persona that shares the user's core concern but has distinct background and narrative details, like a junior researcher facing similar career anxieties. A human-subject study compared this in-group persona agent against a conventional agent and one with minimal self-disclosure. Results from post-task questionnaires indicated that the in-group persona agent significantly boosted perceived rapport and personal relevance for users. Furthermore, it led to a more positive overall user experience, most notably by increasing user engagement.