Attachment, Anthropomorphism and Usage Frequency in the context of Relational Dependence on Large Language Models

Abstract

Use of Large Language Models (LLMs) is fast becoming a routine part of modern existence, alongside increased loneliness and social anxiety. Relational bonding with LLMs poses a risk to vulnerable users. This study examined the predictive power of anxious attachment (AA), perceived anthropomorphism (PA) and usage frequency (UF) on relational dependence (RD) among general purpose LLM users. A cross-sectional quantitative survey was employed (N = 149) to collect data from a community sample of frequent users. Both AA (β = .24, p = .003) and PA (β = .33, p < .001) emerged as significant unique predictors. Weekly LLM users reported significantly lower dependence compared to the high-frequency baseline. These findings suggest that psychological mechanisms and perceptions of AI are critical drivers of relational bonds. As LLMs become more ubiquitous, this study highlights the need for informed AI ethics and increased clinical focus on the use of AI in vulnerable populations. Keywords: Anxious Attachment, Anthropomorphism, Relational Dependence, Large Language Models, Human-AI Interaction, AI Ethics