Emerging Frameworks: A Systematic Review of Publishers’ Policies for Generative AI in Medical Sciences Research

Abstract

Background and Aim: The rapid integration of generative AI into scientific research has transformed scholarly publishing, creating an urgent need for transparent, operational policies. This systematic review aimed to map current policies of major publishers regarding generative AI, identify points of consensus and implementation gaps, assess the quality of existing guidelines, and provide evidence-based recommendations for responsible technology integration.

Materials and Methods: This systematic review was conducted following PRISMA 2020 guidelines. A comprehensive search was performed in PubMed, Scopus, and Web of Science (January 2023 to December 2024). Additionally, a manual gray literature search was conducted on the websites of 10 major publishers (Elsevier, Springer Nature, Wiley, Taylor & Francis, SAGE, Oxford University Press, Cambridge University Press, BMJ, JAMA Network, and The Lancet) and key organizations (COPE, ICMJE, WAME). Two independent reviewers performed study selection using Rayyan software. Extracted data were synthesized using thematic analysis (Braun & Clarke’s framework) and quantitative content analysis. Policy document quality was assessed using a structured checklist based on three criteria—clarity, explicitness, and comprehensiveness—measured on a 3-point Likert scale by two independent evaluators.

Results: This systematic review examined 28 studies, including 17 policy documents from 10 major academic publishers and 11 analytical articles published between 2023 and 2024. Regarding policy documents from publishers, key findings revealed a global consensus (100%) prohibiting AI authorship. However, significant implementation gaps were observed: 9 out of 10 publishers require mandatory AI disclosure (Oxford University Press recommends voluntary disclosure only). All publishers permit AI use for language editing, but only 4 publishers (40%) allow it for ideation. Only 3 publishers (30%) provide specific guidelines for AI use in peer review. Additionally, only 3 publishers (30%) mention detection mechanisms, and 1 publisher (10%) addresses consequences for policy violations. Thematic analysis of policy documents revealed four main themes: “ethical consensus on foundational principles,” “operational diversity in implementation requirements,” “policy gap in key areas,” and “quality and clarity of policy documents.” Quality assessment showed that 64.7% of policies were rated high in clarity, but explicitness (47.1%) and comprehensiveness (41.2%) remained suboptimal. Regarding analytical articles, the reviewed studies consistently emphasized several critical themes: first, the urgent need for harmonized and operational AI policies across publishers; second, concerns about the lack of transparency and accountability in AI-assisted writing; third, the ethical challenges of AI authorship and the importance of maintaining human responsibility for research integrity.

Conclusion: While a strong ethical consensus exists on core principles (particularly AI authorship prohibition), a profound gap remains between stated principles and implementation mechanisms. The development of harmonized, evidence-based, and operationally detailed frameworks—including mandatory detailed disclosure, peer review guidance, and oversight mechanisms—is an urgent necessity to safeguard research integrity in the AI era.