AI writing in academic papers
Major journals and publishers have converged on clear rules for AI-written text in submissions. This page summarizes the COPE framework and the policies of Nature, JAMA, Elsevier, Springer Nature, and Wiley, and explains what honest disclosure looks like.The COPE framework
The Committee on Publication Ethics (COPE) sets the international standard for academic publishing ethics. Its position on AI authorship, published in early 2023 and updated since, is the baseline most journals follow. The current guidance is at publicationethics.org. The core points:- AI cannot be listed as an author. Authorship implies accountability, and a tool cannot be accountable.
- AI use must be disclosed, in methods, acknowledgments, or both, depending on the nature of the use.
- Authors are responsible for AI output. Fact-checking, verifying quotations, and confirming accuracy remain with the authors, regardless of how the text was drafted.
Nature's policy
Nature's AI policy is one of the most explicit. Key points:- LLMs do not meet authorship criteria and cannot be listed as authors.
- Use of LLMs must be documented in the methods section, or in the acknowledgments for incidental use.
- AI-generated images and figures are not permitted in submissions.
- Authors remain fully responsible for all content in their paper.
JAMA and medical journals
JAMA's instructions for authors specifically address AI use. Authors must:- Disclose AI use in the manuscript, including which sections involved AI.
- Take responsibility for the accuracy of AI-generated content.
- Not list AI as an author.
Elsevier, Springer Nature, and Wiley
The major commercial publishers have converged on similar positions. Elsevier's publishing ethics page is a representative policy. Key themes:- AI tools do not meet authorship criteria.
- Disclosure in methods or acknowledgments is mandatory.
- Data-generating AI use requires more detailed disclosure than writing-assistance AI.
- AI-generated images and data visualizations face extra scrutiny.
What disclosure should look like
A typical journal-compliant disclosure names the tool, describes what it was used for, and confirms that the authors reviewed and take responsibility for the final text.Adjust the specifics for what you actually did. If AI helped with initial drafting, say so. If it assisted with data analysis, say what specifically. Vagueness can raise more questions than it answers.What is generally prohibited
- Listing AI as an author. This is universal across major journals.
- Using AI to generate data or results that appear as original research. This is fabrication and is prohibited by every journal.
- AI-generated images and figures. Some journals ban them entirely, some require disclosure and specific review.
- AI-generated peer review. Using ChatGPT to write peer reviews has become a specific concern, and most journals now explicitly prohibit it.
The detection question for authors
Some journals run submissions through AI detection tools. If your disclosed AI-assisted paper scores high on a detector, the disclosure in your methods section answers the question. If you did not disclose and the detector flags you, that is a problem.The simplest approach: disclose honestly, do substantial human editing, and if you are concerned about how the final piece reads, run it through a detector yourself before submission. Our piece on citing AI use covers the citation formats across major style guides.Keep in mind that detectors are signals, not proof. They can be wrong in both directions, and a score alone should never be the basis for an accusation. If a detector flags text you wrote yourself, which is common for non-native English speakers, a careful editing pass to vary sentence length and remove stock AI phrasing can reduce the false-positive signal while preserving your authorship.A responsibility note
Academic publishing runs on trust. Undisclosed AI use, even if a detector does not catch it, undermines that trust. The right approach is always: assist with AI if useful, disclose honestly, and take responsibility for the final text. Detection is a backstop, not a first line of defense against dishonest practice.Frequently asked questions
No. Nature, JAMA, Science, and every major journal explicitly prohibit listing LLMs as authors. The reasoning is that authorship carries accountability, and a tool cannot be accountable for the work.
Yes, for any substantive AI use. COPE's guidance and most journal policies now require disclosure in methods or acknowledgments. The specifics of what to include vary by journal, so always check the submission guidelines.
Generating text, analyzing data, drafting sections, and literature summarization all count. Routine grammar checking, spell-check, and similar tools generally do not require disclosure. When uncertain, disclose.
Some journals run submissions through AI detectors. If your text flags high, the editorial process may raise questions. Detectors are imperfect signals and can be wrong, but disclosure in the methods section answers the question before it is asked and is the safest approach.
Some smaller journals still do not. In that case, default to COPE's guidelines as the international standard: treat AI as a tool, not an author, and disclose substantive use.