Researchers identified 275 potentially eligible people through departmental personnel and training records and the membership rosters of closed professional WeChat groups. They excluded 46 before invitation and directly invited the remaining 229 through WeChat. Of those invited, 185 submitted complete questionnaires and 44 did not respond, yielding a response rate of 80.8%.

In the full sample, 80 respondents, or 43.2%, described their understanding of AI as general. The AI-exposed subgroup comprised the 151 respondents who reported at least occasional exposure. Among this subgroup, 142, or 94.0%, selected independent online searching as a route of exposure.

Because the questionnaire allowed multiple routes, the finding identifies online searching as the most frequently selected one rather than showing that all exposure occurred outside formal systems. Institutional provision was a prominent expectation among the full sample. A total of 104 respondents, representing 56.2% with a 95% confidence interval of 49.0% to 63.2%, selected the strongest expectation category for their hospital or school to introduce dedicated AI tools.

The most frequently reported concern was an inaccurate AI diagnosis or suggestion, selected by 149 of 185 respondents, or 80.5% with a 95% confidence interval of 74.2% to 85.6%. Overreliance on AI was selected by 128 respondents, or 69.2%, while data-privacy breaches were selected by 115, or 62.2%. The survey was descriptive, so these frequencies document the concerns reported by participants but do not establish the incidence of errors, inappropriate reliance or privacy failures in clinical practice.

Respondents also endorsed several predefined strategies for managing perceived risks. Labeling information sources was the most frequently selected measure, supported by 140 of 185 respondents, or 75.7%. Clarifying that AI cannot replace clinical judgment was supported by 131 respondents, or 70.8%.

Regularly updating specialty guidelines and cases was selected by 119 respondents, or 64.3%, and strengthening data encryption and protection by 117, or 63.2%. These were reported preferences rather than tested interventions. The findings do not show whether any particular safeguard improves diagnostic accuracy, clinical decisions, training outcomes or patient safety.

Exploratory comparisons between trainees and endocrinologists produced small-to-moderate effect sizes and were interpreted descriptively. The investigators used two-sided Mann–Whitney U tests for ordered variables and Pearson’s chi-square or Fisher’s exact tests for nominal and multiple-response variables, reporting effect sizes without adjustment for multiple exploratory comparisons. The convenience sample, WeChat-based recruitment and reliance on self-report further limit generalizability and causal interpretation.

The researchers concluded that the findings identify priorities for the future development, evaluation and governance of AI tools, training and governance in endocrinology, particularly around accuracy, overreliance and privacy. The survey did not test AI systems or measure clinical benefit, and its reported results should not be interpreted as evidence that AI improves diagnosis or endocrinology training.

Subject of Research: Self-reported AI exposure, needs, preferences and concerns among Chinese endocrinologists and postgraduate trainees

Article Title: Artificial intelligence exposure, perceived needs, preferences, and concerns among endocrinologists and trainees in China: a convenience survey

Article References: Zhang, R., Zhao, Y., Duan, L., & Pan, H. (2026). Artificial intelligence exposure, perceived needs, preferences, and concerns among endocrinologists and trainees in China: a convenience survey. BMC Medical Education. https://doi.org/10.1186/s12909-026-10303-w

Image Credits: AI Generated

DOI: 10.1186/s12909-026-10303-w

Keywords: artificial intelligence, endocrinology, medical education, China, clinician survey, trainee survey, AI safety, data privacy, clinical decision support, Artificial, intelligence, exposure