Uncovering the Nexus Between Attitudes Toward LLMs and Problematic Dependency on Them: A Latent Profile Analysis

As large language models (LLMs) become increasingly integrated into daily life, their potential to foster not only functional use but also psychological dependency is gaining attention. This study applies Latent Profile Analysis (LPA) to identify distinct user profiles based on LLMs dependency (instrumental and relational), and attitude towards LLM (acceptance and fear). Using a UK-based sample (N = 526), two profiles emerged: Dependence-Oriented Users (DOU) and Independence-Oriented Users (IOU). They were then validated through the external variables of psychological distress and contextual use, which demonstrated meaningful differences between the profiles and supported their distinctiveness. DOU reported significantly higher levels of acceptance, relationship dependency, and perceived helpfulness in personal contexts, but also elevated levels of depression and anxiety. In contrast, IOU showed minimal relational reliance and pragmatic engagement. These findings highlight the dual nature of LLM use as a cognitive tool and emotional companion and raise critical questions about autonomy, digital well-being, and responsible AI interaction. The study offers actionable insights for adaptive LLM design, mental health safeguards, and user-centred policy development.