Global consensus sets rules for AI-assisted medical writing

By AI, Created 2:35 AM UTC, May 25, 2026, /AGP/ – An international research panel has published the first practical guidelines for using generative AI in medical writing, aiming to reduce fake citations, data leaks and undisclosed machine authorship. The framework tells researchers where AI can help, where it is banned, and how to disclose use to keep scientific work credible.

Why it matters: - Generative AI is now common in research writing, but inconsistent publisher rules have left scientists guessing about what is acceptable. - The new consensus gives researchers and journals a shared framework to reduce misconduct risks, protect patient data and preserve trust in published science. - The guidance is especially useful for non-native English speakers who want language help without crossing ethical lines.

What happened: - Researchers led by experts from The First Affiliated Hospital of Shenzhen University and an international panel published new guidance on March 10, 2026 in Regenesis Repair Rehabilitation. - The panel included contributors from universities, hospitals and editorial institutions in China, Italy, Japan, Canada, Australia and other countries. - The team audited AI policies from 15 major publishers, including Elsevier and the JAMA Network. - Using a structured Delphi process, the group created a section-by-section checklist for disclosing AI use in medical manuscripts. - The full source is available through the published DOI.

The details: - The guidance allows AI to improve grammar, readability and structure, but not to invent or alter scientific content. - The rules prohibit generative AI from creating or manipulating primary research images, including microscopy images, gels and flow cytometry plots. - The guidelines ban AI-generated references because large language models can fabricate authors, journal names and DOIs. - AI can be used for supportive tasks such as summarizing a discussion or refining an abstract, but a human must verify every claim. - In the Methods section, AI may improve readability but cannot invent missing procedures. - In the Results section, AI may polish text, but the numbers must come from real experiments. - AI is allowed for non-data visuals such as workflow diagrams, but that use must be fully disclosed. - The authors said they used Gemini and ChatGPT during their own writing process only for language and structure, not for core science.

Between the lines: - The consensus is a response to a practical problem: publishers have rules, but the rules are inconsistent and often hard to apply in real workflows. - The main concern is not simply AI use, but hidden AI use that can produce fake references, misleading text or undisclosed editing. - The framework tries to normalize transparency instead of forcing a blanket ban that many researchers would likely ignore.

What’s next: - The authors said they will update the rules regularly as AI tools evolve. - Researchers can use the checklist in supplementary files or share it with collaborators to document AI use. - Journals can use the framework as an audit trail during peer review.

The bottom line: - AI is treated as a permitted assistant, not a substitute author, and human researchers remain fully accountable for the science.