Psychological Outcomes of Sustained Engagement with Conversational AI Systems: A Narrative Review

Abstract

Background: Conversational Artificial Intelligence (AI) systems based on large language models have become part of everyday life, creating a new form of human–machine interaction with potential psychological consequences.

Objective: This narrative review synthesizes current literature on the psychological outcomes of sustained engagement with conversational AI, focusing on emotional attachment, cognitive offloading, dependency formation, identity development, and mental health risk, and weighs these against evidence of therapeutic benefit.

Methods: Peer-reviewed articles, preprints, clinical case reports, and institutional reports published up to mid-2026 were identified through targeted searches and synthesized thematically, with the type of evidence behind each claim reported explicitly.

Results: Sustained engagement was associated with emotional attachment, markers of cognitive offloading, and problematic use, particularly among emotionally vulnerable users, adolescents, and people with insecure attachment. In a four-week randomized controlled trial of 981 participants, voluntary usage volume, rather than assigned interaction condition, was associated with poorer psychosocial outcomes. A small set of case reports and one institutional case series link intensive use to acute psychiatric crises, but most involve pre-existing vulnerability, and none establishes incidence or causation. In contrast, randomized trials of structured therapeutic chatbots report small, mostly short-term reductions in depressive and anxiety symptoms (pooled Hedges’ g of roughly 0.2–0.3 in two meta-analyses), with considerable heterogeneity and limited follow-up.

Conclusions: Psychological outcomes appear to depend on system design, pattern of use, and user vulnerability rather than on conversational AI as such. Current evidence is largely cross-sectional, short-term, or anecdotal and cannot support definitive causal claims. Longitudinal cohort studies, clinical validation of general-purpose systems, and clear ethical and regulatory frameworks are priorities for the safe integration of conversational AI into emotional and clinical support contexts.