Functional and existential dependence on large language models among clinical nurses and associated factors: a cross-sectional study

Abstract Background As generative artificial intelligence is rapidly adopted in health-care and nursing settings, large language models (LLMs) are emerging as tools that may help clinical nurses retrieve information, improve documentation, and support clinical work. However, inappropriate dependence on model outputs may develop during use. Such dependence could weaken nurses’ independent judgment and critical appraisal, thereby posing risks to care quality and patient safety. Objective To assess the current level of LLM dependence among clinical nurses and identify associated factors. Methods A cross-sectional survey was conducted from August to September 2026 in six hospitals across three provincial-level administrative regions in China using convenience sampling. Data were collected with a structured and validated questionnaire. Independent-samples t tests and one-way analysis of variance (ANOVA) were used to compare dependence scores across participant characteristics. Multiple linear regression was used to examine factors associated with functional and existential dependence. Results A total of 459 valid responses were analyzed. The mean functional and existential dependence scores were 35.42 ± 5.81 and 27.18 ± 7.06, respectively, corresponding to mean item scores of 3.94 ± 0.65 and 3.02 ± 0.78 on a 1–5 scale.Univariate analyses showed significant differences in functional and existential dependence scores according to age, educational attainment, frequency of LLM use, and receipt of related training ( P < 0.05). Multiple linear regression showed that frequency of LLM use and receipt of related training were significantly associated with both functional and existential dependence scores, whereas educational attainment showed a more complex adjusted association. More frequent LLM use was associated with higher scores on both dependence dimensions, whereas receipt of related training was associated with lower scores. Conclusion Functional items had a higher mean score than existential items in this sample. These scores do not establish validated severity levels or the prevalence of problematic dependence. Greater LLM-use frequency was associated with higher dependence scores, whereas receipt of related training was associated with lower scores. Future nursing management should address the potential risk of dependence in LLM use. Standardized training, institutional policies, and ethical oversight should be implemented to guide nurses toward appropriate, cautious, and responsible use.