How AI agents’ self-disclosure in initial interactions shapes user perceptions: evidence from dual cognitive pathways

Purpose As users increasingly engage with various unfamiliar artificial intelligence (AI) agents, initial impressions become crucial. AI agents’ self-disclosure can play a key role in shaping these impressions. This study aims to conceptualize AI agents’ self-disclosure, identify key user perceptual factors and examine how self-disclosure influences user perceptions through different cognitive pathways. Design/methodology/approach A two-stage questionnaire survey was conducted. Firstly, exploratory factor analysis was employed to identify key perceptual factors shaped by AI self-disclosure and to examine their effects on users’ willingness to use. Then, users’ willingness to use the AI agents adopting different self-disclosure styles across utilitarian and hedonic scenarios was investigated, and the underlying cognitive mechanisms were further analyzed. Findings Three core perceptual factors were identified: perceived social individuality, perceived reliability and perceived emotional support. In both utilitarian and hedonic scenarios, users consistently preferred an emotional (over factual) tone for relation-oriented self-disclosure, which incorporated richer social cues, and a detailed (over brief) style for transparency-oriented self-disclosure, which conveyed more machine-related characteristics. Perceived social individuality was found to play a more important role in users’ evaluation of AI agents’ relation-oriented self-disclosure. The results provide empirical support for a dual perspective of AI agents that can be shaped by self-disclosure: relation-oriented self-disclosure activates social responses, whereas transparency-oriented self-disclosure triggers machine-based evaluations. Originality/value This study proposes and empirically validates the concept of AI self-disclosure as a dual-dimensional construct. By uncovering the mechanisms through which it shapes initial perceptions, it offers actionable insights for designing more transparent and socially attuned AI agents.