A critical narrative synthesis of psychological correlates, measurement, and reported findings on conversational AI engagement and dependence-related constructs

Conversational AI systems are used for task support, cognitive offloading, companionship, self-disclosure, and mood regulation. Research on these uses is distributed across trust, reliance, attachment, companionship, problematic use, and dependence, although these labels refer to different processes. This article presents a critical narrative synthesis of 51 peer-reviewed records retained from a curated working bibliography of 56 items. The corpus was assembled iteratively through database searching and citation chasing during 2025 and early 2026. Complete search histories and duplicate-removal logs were not retained, so the review is not presented as a reproducible systematic review. The sole author completed selection, extraction, coding, appraisal, and synthesis. Conceptual and measurement papers were separated from 22 first-order empirical records. Sixteen of those empirical records provided lower-leverage evidence and six provided moderate-leverage evidence. Mainly cross-sectional studies identified concurrent associations between stronger AI engagement or dependence-oriented scores and loneliness, social anxiety, depressive symptoms, low self-esteem, attachment insecurity, academic stress, escapism, anthropomorphic tendency, fatigue, weaker critical thinking, procrastination, and lower well-being. These associations do not establish prediction or consequence. Qualitative studies described both relational support and distress linked to emotionally significant use. Quasi-experimental and multi-study research reported some short-term reductions in loneliness and social anxiety, but this evidence remains limited and context dependent. The synthesis develops a provisional framework that separates instrumental-cognitive from relational-emotional use and treats regulation as a distinct dimension. The framework is an interpretive organization of the corpus, not a validated factor structure, causal pathway, or clinical model. Current evidence supports careful construct separation and design-matched claims, while stronger longitudinal, cross-cultural, and platform-comparative research is needed.