A journal standard defines authorship to exclude any AI tool

The writer's problem

A researcher preparing a manuscript with help from a chatbot has to decide what to call that help, whether the tool can be credited the way a human contributor would be, and who answers for a wrong or invented claim the tool introduced. The International Committee of Medical Journal Editors' recommendations on AI use by authors is the document editors and authors across many journals point to first.

What the documents show

ICMJE's page states plainly that chatbots and other AI-assisted tools should not be listed as authors, because authorship requires the capacity to take responsibility for a work's accuracy, integrity and originality, a capacity a tool does not have. It requires disclosure of AI use in a cover letter and in the relevant part of the manuscript, and it tells authors to carefully review and edit any AI-generated content because such output can be incorrect, incomplete, or biased. AI-generated material cannot serve as a manuscript's primary source, and every quotation still needs a proper citation. Failing to disclose AI use, the page states, may require corrective action and may be treated as misconduct. The recommendations are a journal-editing standard about credit and responsibility, distinct from a copyright authority's separate test for whether AI-assisted text can be registered.

The editorial choice

Given this, an author's practical choice is to disclose specifically, in the section ICMJE names, rather than folding AI use into a general acknowledgment, and to independently verify every fact and citation a tool supplies rather than treating fluent output as already checked. This is an editorial reading of the recommendations, since ICMJE does not prescribe one universal disclosure sentence; individual journals built their own submission instructions on top of this baseline.

What stays with the author

Verifying accuracy, avoiding plagiarism in AI-generated passages, and answering for any lapse all remain the named human author's job under this standard; the recommendations make no exception for text a tool produced. What ICMJE's own site does not resolve is how an editor is meant to detect an undisclosed use in the first place, leaving enforcement dependent on an author's own disclosure and a journal's separate screening practices.

• Does a journal's own instructions to authors say exactly where AI use must be disclosed, or only that it must be?
• Has every AI-suggested citation been checked against the actual source, not just the AI's summary of it?
• If an AI-assisted claim later proves wrong, whose name is on the correction?

ICMJE's answer to the authorship question is narrow by design: it says what a chatbot cannot be, and puts the resulting responsibility back on a named person.