Consensus on the application of generative artificial intelligence in medical manuscript writing

This is the authors' abstract. We don't add key points for this paper.

Introduction: Generative artificial intelligence (AI) has become integral to medical writing, offering significant benefits in linguistic refinement and efficiency, particularly for non-native English speakers. However, its adoption is challenged by risks such as reference fabrication, authorship disputes, and data confidentiality breaches. While major publishers have introduced AI-related policies, significant heterogeneity remains in their operational granularity and technical requirements. This lack of harmonization creates ''regulatory uncertainty'' for researchers, necessitating an expert-led consensus to provide consistent, practice-oriented guidance while safeguarding research integrity. Methods: A systematic review of publicly available AI policies from 15 leading publishing organizations was conducted to identify convergences and divergences in ethical principles and operational guidance. Based on this review, a modified Delphi process was undertaken involving a multidisciplinary expert panel to develop consensus recommendations for the responsible use of generative AI in medical manuscript writing. Results: The panel reached consensus on three deliverables: (1) overarching principles emphasizing data authenticity, transparent disclosure of AI use, and full human accountability with explicit prohibition of AI authorship; (2) section-specific guidance defining permissible and prohibited AI applications across all manuscript components, including strict restrictions on AI-generated references and any AI fabrication or alteration of data-derived images; and (3) a standardized, section-by-section disclosure checklist to facilitate consistent reporting and editorial review. Discussion: The consensus offers a structured and actionable framework for AI-assisted medical writing. Emphasizing transparency, data authenticity, and full human responsibility, the consensus advocates responsible use of AI to facilitate academic research and communication.