"Disclose your AI use" has become close to universal in academic publishing. What that sentence actually requires, in practice, differs enough between publishers that an author who is scrupulously honest can still get the format wrong, and a format mismatch on a disclosure obligation is not a formatting problem; several of these policies treat it as a misconduct problem.
Why the differences aren't just paperwork
Underneath the formatting differences, every one of these policies is asking for the same underlying thing: an accurate account of which words came from where, produced at submission time, days or weeks after the actual writing happened. Science's "the actual prompt used" and JAMA's "dates of use" are both details nobody writes down while drafting a paper, which means authors are reconstructing them from memory under a policy that treats an inaccurate reconstruction as potential misconduct, regardless of intent. Worth knowing exactly what each publisher wants, then, before you're the one doing the reconstructing:
The baseline: ICMJE
The International Committee of Medical Journal Editors, whose recommendations thousands of journals follow, sets the floor:
"Authors should not list or cite AI and AI-assisted technologies as an author. When AI is used, users should disclose which tool was used, and for what purpose. Humans are ultimately responsible for reviewing and ensuring the accuracy of any content generated with assistance of AI."
ICMJE Recommendations on Artificial Intelligence
Placement is typically the Acknowledgments section, though ICMJE leaves the specific location somewhat flexible (ICMJE recommendations on AI), which is exactly where the individual publishers below diverge from each other.
Nature Portfolio: Methods, with a copyediting exemption
Nature Portfolio requires disclosure in the Methods section (or the nearest equivalent) specifically when AI tools contributed to content generation, rewriting, or summarizing, or played any editorial role. Minor copyediting, grammar, spelling, punctuation, or tone correction, is explicitly exempt from disclosure, as long as the tool is not creating new content or making editorial decisions. Generative AI is banned outright for producing images or figures. The policy is blunt about where the line sits:
"Accountability for scholarly content, evaluation, and editorial decisions cannot be delegated to AI systems."
Nature Portfolio, AI editorial policy
(Nature Portfolio AI editorial policy)
Science / AAAS: the strictest reading, down to the prompt
Science journals take the hardest line of the group. AI-assisted technologies cannot be authors or co-authors, and disclosure has to appear in both the cover letter and the Acknowledgments section, but Science goes further than most: authors are expected to disclose the actual prompt used to produce AI-assisted content, along with the tool and its version, in the Methods section (Science, official policy update). This treats the interaction itself, not merely the fact that one occurred, as part of the disclosable record.
JAMA Network: a structured checklist, not a sentence
JAMA Network's requirement is the most granular of the five: tool name, version, manufacturer, dates of use, a description of the AI's specific role, and explicit confirmation that the author takes responsibility for the output. Disclosure goes in Acknowledgments or Methods, but is also captured through a structured question in the manuscript submission system itself, separate from the manuscript text. AI-generated images, illustrations, and figures are not permitted (JAMA Network, updated AI guidance; original JAMA guidance for authors, reviewers, and editors).
IEEE: name the system, name the sections
IEEE's policy reads like a checklist because it is meant to be used as one:
"The use of content generated by artificial intelligence (AI) in an article (including but not limited to text, figures, images, and code) shall be disclosed in the acknowledgments section of any article submitted to an IEEE publication. The AI system used shall be identified, and specific sections of the article that use AI-generated content shall be identified and accompanied by a brief explanation regarding the level at which the AI system was used."
IEEE, submission and peer review policies
Grammar and editing tools get a narrower carve-out than Nature's: "The use of AI systems for editing and grammar enhancement is common practice and... is generally outside the intent of the above policy... but recommended." IEEE reviewers are separately barred from running confidential manuscripts through public AI tools (IEEE Author Center, submission and peer review policies).
Five publishers, five different forms
| Publisher | Where it goes | Must include | Especially strict about |
|---|---|---|---|
| ICMJE (baseline) | Acknowledgments (typical) | Which tool, for what purpose | Any undisclosed use at any editorial stage |
| Nature Portfolio | Methods (or nearest equivalent) | Description of the AI's generative or editorial role | AI-generated images or figures (banned) |
| Science / AAAS | Methods + cover letter | Tool, version, and the actual prompt used | Text generation itself, treated as misconduct if undisclosed |
| JAMA Network | Structured submission field + Acknowledgments/Methods | Tool, version, manufacturer, dates, role, responsibility confirmation | AI-generated images or figures (barred) |
| IEEE | Acknowledgments | System name, affected sections, explanation of use | Reviewers using public AI tools on confidential manuscripts |
Filling in the checklist without reconstructing it
Write the paper in TeXposit and the reconstruction step mostly disappears. The per-character Source View keeps track of exactly which passages were typed directly and which came from an accepted AI suggestion, timestamped, so JAMA's "dates of use" field or Science's methods-section prompt disclosure stop being something an author has to dig up and become something the provenance report already has on record. The contributor and AI-share breakdown gives a specific figure to put in whichever section a given journal wants it, Methods for Nature, Acknowledgments for ICMJE, the structured field for JAMA, one a reviewer or editor can check rather than take on trust.
It doesn't decide what a given journal will accept. That's still an editorial call, and a report describes composition, not whether the composition clears a specific publisher's bar. It closes a narrower gap: between what actually happened while the paper was written and what an author can accurately recall about it weeks later at submission. The other half of this same trust problem is citations rather than authorship, and that one we've written about separately.