Development and Validation of the German Version of the Large Language Model Dependency Scale

As large language models (LLMs) become increasingly embedded in everyday work, education, and decision-making, concerns have emerged regarding users’ reliance on these systems and the psychological mechanisms underlying such reliance. In German-speaking contexts, public discourse on generative AI is strongly shaped by issues of data protection, transparency, autonomy, and regulatory oversight, yet empirically validated instruments to assess dependency on LLMs are lacking. Addressing this gap, the present study translated and validated the German version of the Large Language Model Dependency Scale (LLM-D12-DE). A sample of 402 German-speaking active LLM users completed the German LLM-D12. Confirmatory factor analysis supported the hypothesized two-factor structure comprising Instrumental Dependency and Relationship Dependency. Both dimensions demonstrate excellent internal consistency, strong composite reliability, and satisfactory convergent validity. Scalar measurement invariance across gender was established, allowing meaningful latent mean comparisons. Network analysis revealed two coherent but distinct item clusters corresponding to instrumental reliance and relational attachment, with central items reflecting decision-related reliance and perceived reduction of loneliness. External validation analyses showed that both dependency dimensions were positively associated with AI acceptance attitudes, perceived trustworthiness of LLMs, and internet addiction severity. Relationship Dependency was additionally associated with lower need for cognition. Usage characteristics, particularly higher frequency of LLM use and multimodal interaction, emerged as robust predictors of dependency levels. Overall, the findings demonstrate that the LLM-D12-DE is a psychometrically sound and contextually appropriate instrument for assessing dependency on LLMs in German-speaking populations. The scale enables systematic research on cognitive and relational aspects of human-LLM interaction and provides an empirical basis for educational, clinical, and regulatory discussions on responsible AI use.