Using AI tools responsibly in medical writing

Generative AI tools are now part of many researchers’ writing routines. They can suggest clearer phrasing, summarise long texts and help non-native speakers write in English. They can also invent references, introduce errors and expose confidential data. Using them well means knowing where the limits are.

What AI can reasonably help with

• Improving grammar, clarity and readability of text you wrote.
• Suggesting a structure for a section or a shorter version of an abstract.
• Checking consistency of terms and abbreviations.

Where the risks are

• Fabricated content: tools can produce plausible but false statements and references that do not exist. Every claim and citation must be checked against the source.
• Accountability: an AI tool cannot take responsibility for the work, so it cannot be an author. Responsibility stays with the people listed.
• Confidentiality: uploading unpublished data, patient information or a manuscript under review to a public tool may breach confidentiality and data protection rules.
• Bias: generated text can reflect biases in the data the tool was trained on.

Disclose what you used

Journals increasingly ask authors to state whether and how AI tools were used, usually in the methods or acknowledgements. Name the tool and describe the task, for example language editing of the discussion section. Using AI for language help is rarely a problem; hiding it is.

For reviewers and editors

Manuscripts under review are confidential. Reviewers should not paste them into public AI services, and editorial decisions should always be made by people who have read the work.

Use AI as an assistant, never as an author, and always check its work.

Our journals’ policy on AI follows these principles. See our approach to innovation for details.