Use of large language models in German general practice: A cross-sectional survey

Abstract Background Large language models (LLMs) are increasingly being used in healthcare, but quantitative evidence from German general practice remains scarce. This study examined the current use of LLMs among general practitioners working in Germany, barriers to their use, their perceived impact on the physician-patient relationship and expectations regarding their future role in general practice. Methods A cross-sectional online survey was conducted from 5 February to 5 June 2026. Complete questionnaires were analyzed using descriptive statistics. Associations were explored using Spearman’s rank correlation and the Kruskal-Wallis test. Results A total of 206 completed questionnaires were analyzed. Approximately 80% reported professional LLM use, with ChatGPT being the most frequently used tool (65%). LLMs were used mainly for knowledge- and decision-support tasks, whereas administrative applications were less common. More frequent use was associated with younger age (ρ = -0.302), fewer years of professional experience (ρ = -0.336) and higher self-assessed knowledge (ρ = 0.350), while use frequency did not differ by gender (Kruskal-Wallis χ² = 0.0117). The main barriers included fear of inaccurate information, concerns about data protection and uncertainty regarding the legal framework and liability. 71% had never attended training on LLMs, although 64% expressed willingness to do so in the future. LLMs were used during patient encounters by 41% of respondents, including 26% not disclosing their use to patients. Most respondents expected LLMs to become an integral part of primary care and anticipated increased future use. There was strong agreement that LLMs could not replace GPs. Conclusions LLMs were already used among the surveyed GPs, although their use remained selective and focused mainly on knowledge- and decision-support tasks. The expected expansion of LLM use, the identified barriers and limited participation in training, highlight the need for secure applications, clear legal and liability-related guidelines, as well as practice-orientated education to support a safe integration of LLMs into general practice.