About one in three medical interns surveyed at a Chinese teaching hospital said they had used DeepSeek for learning. Among those users, most reported using the AI tool at least weekly, while views of its advantages and disadvantages were generally balanced. The results offer a snapshot of adoption in one training setting, not evidence about medical interns across China or the effects of AI on learning.
In a cross-sectional study published in PLOS ONE, Tao Zhang and colleagues surveyed undergraduate interns at the Second Affiliated Hospital of Anhui Medical University between June and October 2025. The anonymous questionnaire covered interns in clinical medicine, preventive medicine and nursing. Of 657 returned questionnaires, 634 were included in the analysis; 228 respondents, or 36%, reported using DeepSeek for learning.
Use was regular among that subgroup. Of the 228 users, 123 (53.9%) said they used DeepSeek several times per week, and 28 (12.3%) reported daily use. Another 23.2% used it about once a week, while 10.5% described their use as occasional.
Interns were asked to select one main learning purpose. Completing assignments was the most common response, at 27.6%, followed by information searches (22.4%), translation (18.0%) and polishing written work (17.1%). Data analysis was the primary use for 14.9% of users. These answers describe what respondents said they used the tool for; the survey did not assess the quality or accuracy of its output.
Users’ overall views were mostly neutral or optimistic: 44.7% were neutral, 41.7% optimistic and 13.6% pessimistic. In the questionnaire, an optimistic response meant intending to maintain or increase use, while a neutral response meant using DeepSeek depending on future academic needs. When asked to weigh advantages against disadvantages, 64.5% said they were roughly balanced, 28.1% saw benefits as greater and 7.5% saw disadvantages as greater.
Reported use also varied across groups. After adjustment for other measured characteristics, urban interns had higher odds of reporting use than rural interns. Compared with clinical medicine interns, preventive medicine interns had lower odds, while nursing interns had higher odds. Age and sex were not significantly associated with use after adjustment. Among users, attitudes differed by major, with nursing interns more often neutral and less often optimistic; use frequency and perceived balance of advantages and disadvantages did not differ significantly by major. These are associations in this sample, and the survey cannot establish why the groups differed.
The authors say the findings could inform future institutional guidance on responsible use, verification of AI-generated information and academic integrity. The study did not measure learning outcomes, test DeepSeek’s accuracy or establish whether use helped interns with their coursework. It also did not ask why some interns had not used the tool. Because the survey covered one hospital at one point in time, its results may not generalize to other institutions or show how use and attitudes change. The authors also note that the questionnaire was not formally piloted and used brief categorical questions rather than validated scales, so the attitude results are descriptive.
The report therefore provides a practical baseline: in this cohort, adoption was not universal, but users often engaged with DeepSeek regularly for academic tasks. Whether such use improves learning, and what guidance best supports it, remains open. <a href="https://journals.plos.org/plosone/article?id=10.1371%2Fjournal.pone.0362171">Source: PLOS ONE study</a>