Artificial Intelligence Use in Medical Writing: On the Meaning and Limits of Disclosure

Abstract

This opinion paper focuses on disclosure of artificial intelligence (AI) use in medical writing; AI use beyond manuscript writing is outside its scope. Almost all journals require full disclosure of AI use. However, several barriers may hinder such disclosure. First, there is a lack of clear statements by journals on how disclosure is evaluated. Authors may be concerned that a linguistically polished manuscript accompanied by AI disclosure might be scored lower. Conversely, a journal might question the nonuse of AI assistance when a manuscript presents highly significant data written in uncomfortable English. Authors may fear that either AI “use” or “nonuse” could result in disadvantage. Second, criteria for disclosure remain unclear. Many journals require disclosure when AI is used to “create, review, revise, or edit” but not when it is used for “checking grammar, spelling, and similar tasks.” However, the distinction between “edit” and “grammar check” is a delicate one; prompts such as “edit” and “check grammar” may produce almost identical revisions. Third, it may become difficult to distinguish between one's original expressions and AI-aided revisions after several rounds of linguistic consultation. Authors may regard the final text as their own authorship rather than AI intervention. These three factors, intentionally or unintentionally, may prevent honest and full disclosure. Journals may consider stating two points. First, that the use or nonuse of AI tools does not affect acceptance or rejection, provided that such use adheres to journal guidelines and is accurately disclosed. Second, that disclosure information is accumulated as a resource for future analysis, contributing to a more transparent publication environment. Such clarification could reduce hesitation surrounding disclosure. Transparency, from both authors and journals, would become an asset to the medical community.