Companion agent offers growing emotional support over two weeks
From Momentary Emotion Inference to Sustained Emotion Support: Evaluating a Companion Agent in a Longitudinal Study
Human-Computer InteractionArtificial Intelligence
Summary
Sustained emotional support can help people feel better over time, but it’s unclear how AI tools can do this well across days. The authors tested a special AI companion called PAIR that tries to understand and guide emotions over 14 days with 19 people. They found the AI’s estimates of feelings matched users’ reports, and conversations with PAIR often improved mood and energy. People felt the AI understood them and gave useful advice that fit their situation, and they found it more helpful as time went on.
What this means in practice
- •For mental health app developers: Build apps that provide ongoing emotional guidance personalized through memory updates and learning from user feedback.$Commercial implications: Enables creation of adaptive emotional support companions that improve their help over time, appealing to wellness app markets.
- •For caregiving service teams: Incorporate AI companions to complement human support by tracking emotions and suggesting tailored coping strategies over time.
Authors
Kexin Quan, Zijian Ding, Jiaye Yong, Qinshi Zhang, Dong Wang, Jessie Chin
Abstract
Sustained emotional support is a long-horizon interaction task closely tied to human well-being. Recent research demonstrates generative agents' capacity for momentary emotional support, yet how these capabilities sustain support over time remains unclear. To examine this challenge, we deployed PAIR, a theory-based emotion-regulation companion, with 19 participants for 14 days. Across 1,093 sessions, we paired emotion estimates with self-reports before and after guidance and analyzed logs and interviews. Estimates corresponded more closely to self-reported valence and dominance than arousal. Guided conversations were followed by higher valence and state-dependent arousal changes. Participants felt understood through contextual exploration and emotional acknowledgment, acting on guidance suited to their needs and constraints. Perceived helpfulness of guided conversation significantly increased over time. Our findings link memory updates and retained corrections to cross-session personalization, informing future emotional support tools that adapt to evolving needs, learn from prior outcomes, and preserve user control over memory.