Attachment towards generative AI influences self‐ AI agreement in personality reports

Abstract

Generative artificial intelligence (AI) is increasingly used as a conversational partner, yet little is known about when AI‐generated impressions of users converge with users' self‐views. Drawing on lens theory, attachment theory and the literature on belongingness and social connection, the current study examined whether users' emotional bonds with AI moderate self‐AI agreement in personality reports. A total of 738 Japanese ChatGPT users completed self‐report measures of the Big Five personality traits and AI attachment dimensions. Their personality was also inferred by ChatGPT based on their chat history. Self‐reported and AI‐inferred Big Five traits were positively correlated, although AI systematically rated users higher across all five domains. Moderation analyses showed that AI attachment anxiety, but not avoidance, strengthened self‐AI agreement for extraversion, agreeableness and negative emotionality. Supplemental sensitivity analyses, including all participants who provided AI‐inferred personality data ( N  = 789), produced the same conclusions regarding positive self‐AI agreement, AI overestimation and the absence of moderation by AI attachment avoidance, while indicating that the attachment anxiety‐based moderation was most robust for negative emotionality. These findings suggest that AI‐based personality inference is shaped not only by model capability but also by the relational cues users provide in human–AI interaction.