From emotional comfort to relational investment in generative AI companionship: evidence from SEM and fsQCA

With the rapid diffusion of generative artificial intelligence (AI), chatbots are evolving from instrumental assistants into emotionally responsive companions. This study develops a low-friction relational process model to examine the relationships linking generative AI companionship with emotional relief and positive social investment. AI emotional responsiveness, non-judgemental interaction, and memory-based continuity were conceptualized as interaction conditions, while perceived being understood, deep self-disclosure, and synthetic attachment were specified as psychological mechanisms. Structural equation modeling (SEM), bootstrap mediation analysis, moderation analysis, and fuzzy-set qualitative comparative analysis (fsQCA) were employed to analyse 765 valid responses from generative AI chatbot users. The SEM results show that non-judgemental interaction exhibits the most consistent pattern of associations with perceived being understood, deep self-disclosure, synthetic attachment, emotional relief, and positive social investment. Perceived being understood and synthetic attachment show relatively stable mediating patterns, whereas deep self-disclosure is significantly associated with emotional relief but shows a less consistent association with positive social investment. Perceived AI safety and self-disclosure habit play differentiated moderating roles. The fsQCA findings reveal three configurations for high emotional relief and four configurations for high positive social investment, complementing the SEM results. This study provides a systematic account of generative AI companionship and offers implications for safer and relationship-oriented AI design.