Effect of Artificial Intelligence-Assisted Structured Guidance on the Quality of Narrative Literature Reviews Among Medical Postgraduate Students: A Randomized Controlled Educational Study

Generative artificial intelligence is increasingly used by medical students and postgraduate trainees for literature searching, summarization, outlining, and academic writing. However, free use of AI may be associated with inaccurate information, unsupported statements, fabricated references, and inadequate academic judgment. This study is designed to evaluate whether structured guidance can improve the quality and academic integrity of AI-assisted narrative review writing.

Participants will be medical postgraduate students enrolled in an academic writing course. After informed consent and baseline data collection, participants will be randomized in a 1:1 ratio to an AI-assisted structured guidance group or a free AI use group. Both groups will complete a narrative literature review of approximately 2,000 Chinese characters according to the same course requirements and may use AI tools under academic integrity requirements.

Participants in the AI-assisted structured guidance group will receive one structured guidance session approximately two weeks before final submission. The session will be based on the six SANRA dimensions: explanation of the importance of the article, statement of aims or questions, description of the literature search, reference support for key statements, scientific reasoning, and appropriate presentation of data and content. The instructor will provide directional and methodological feedback but will not directly write or edit the students' text line by line. Participants in the free AI use group will complete the review according to the same course requirements without the structured guidance session.

After submission, all narrative reviews will be anonymized and assessed by independent outcome assessors blinded to group allocation. The primary outcome is the final SANRA total score. Secondary outcomes include SANRA item scores, multiple-source academic writing self-efficacy, and reference authenticity error rate.