Profiles and candidate educational targets of large language model dependence among Chinese nursing students: a cross-sectional study

Abstract Background Large language model (LLM) are increasingly applied in nursing education, yet the manifestations of LLM dependence remain unclear. This study aimed to explore the status, response characteristics, and candidate educational targets of LLM dependence among nursing students. Methods A cross-sectional survey was conducted among 709 nursing students from two Chinese universities from May to June 2026 using a general information questionnaire and the Large Language Model Dependence Scale (LDS). Latent profile analysis identified LDS response profiles, Gaussian and Bayesian network analyses characterized LDS item networks, and computer-simulated perturbations explored candidate educational targets. Results Nursing students had a mean LDS score of 61.77 ± 9.77 and were classified into the Functional-Dominant (51.76%) and Global-Dependent (48.24%) profiles. In Gaussian networks, “Discomfort” was central overall and in the Functional-Dominant profile, whereas “Reliable results” was central in the Global-Dependent profile. Statistically upstream Bayesian nodes were “Work efficiency” and “Reluctant to start” overall; “Discomfort,” “Honesty,” and “Predictability” in the Functional-Dominant group; and “Excellent assistant” and “Discomfort” in the Global-Dependent group. The largest simulated increases and decreases were associated with “Reluctant to start” and “Discomfort” overall, “Spend more time” and “Boring tasks” in the Functional-Dominant group, and “Work efficiency” and “Excellent assistant” in the Global-Dependent group. Conclusions This study characterized how LLM dependence manifests among nursing students, identifying Functional-Dominant and Global-Dependent profiles. It further identified distinct LLM dependence structures and candidate educational targets.