Academic publishing has an AI transparency problem. Researchers are using generative AI tools to write, analyze, and structure their work, and most journals still don’t have a clear, consistent framework for how that use should be reported.
Some journals require blanket disclosure statements. Others ban AI use outright. Many have vague policies that leave authors guessing. And in the middle of all this confusion, a growing number of researchers are quietly using ChatGPT, Claude, or Copilot without saying a word about it.
Cleland, Driessen, Masters, Lingard, and Maggio (2025) address this head-on in AMEE Guide No. 192, a practical framework for when and how to disclose AI use in academic publishing. The guide is aimed at medical education, but the principles apply across every discipline that publishes research.
AI Disclosure in Academic Publishing Is About Integrity
The authors make a clear argument from the start: disclosing AI use in academic publishing is about research integrity. It has nothing to do with policing behavior or shaming researchers who use AI tools. Transparency matters because undisclosed AI use can mislead readers, obscure how knowledge was produced, and weaken accountability.
When readers evaluate a study, they need to understand how the work was done. If AI played a role in generating text, structuring arguments, running analyses, or producing figures, that information changes how the work should be read and evaluated. Cleland et al. argue that making AI’s role visible allows other researchers to “critically evaluate a manuscript and its specific claims” (p. 4).
This framing is important. The conversation around AI disclosure in academic publishing has too often been wrapped in anxiety and suspicion, as if using AI is something to be ashamed of. The guide pushes back on that tone entirely.
Human Responsibility Doesn’t Disappear
One of the strongest points in the guide is also the simplest: AI doesn’t take responsibility for anything. People do.
As Cleland et al. put it: “Human authors remain fully responsible for the accuracy, originality, and integrity of all journal submissions, whether their manuscript was created with or without AI assistance” (p. 2).
That means treating every AI-generated output as unverified material. Fabricated references, biased phrasing, analytic errors, hallucinated citations. All of these are well-documented tendencies in generative AI, and all of them become the author’s problem the moment they appear in a manuscript. Using AI to draft a literature review or clean up a methods section is fine, but the author needs to verify every claim, every source, and every number before submitting.
I covered the intellectual property side of this issue in a previous post on how copyright law is struggling to keep up with AI-generated creative works. The authorship question there mirrors what’s happening in academic publishing: when AI contributes to the creative or intellectual process, the boundaries of human responsibility need to be stated clearly, because the law and the norms haven’t caught up yet.
When Disclosure Is (and Isn’t) Needed
The guide makes a useful practical point: routine AI use probably doesn’t need disclosure. If you used Grammarly to fix a comma splice or asked ChatGPT to rephrase a sentence for clarity, that falls into the same category as using spell-check. It’s becoming standard scholarly practice.
Disclosure becomes necessary when AI materially shapes the work. That means when AI influenced the ideas, the arguments, the analysis, the figures, or the interpretations in a meaningful way. Cleland et al. are direct: “Disclosure is warranted whenever an AI tool materially shapes the research or the manuscript” (p. 4).
The word “materially” is doing a lot of heavy lifting there, and the authors acknowledge it. When you’re unsure whether your AI use crosses the threshold, they recommend disclosing anyway. Better to over-report than to leave readers guessing.
How to Disclose AI Use in Academic Publishing
This is where the guide gets most practical. The authors argue that AI disclosure should look like methodological reporting. The same way researchers report which software they used for statistical analysis or which coding framework guided their qualitative work, AI use should be documented with the same specificity and the same matter-of-fact tone.
A good disclosure names the tool and version, explains how it was used, describes the extent of its influence on the final product, and clarifies how outputs were verified. Vague statements like “AI was used in the preparation of this manuscript” don’t cut it. They tell the reader nothing useful.
Cleland et al. put it plainly: “Treat disclosure as routine methodological reporting, not confession” (p. 6). That line alone should shift how a lot of researchers think about this. The goal is transparency, not guilt.
Start the Conversation Early
One of the most practical recommendations in the guide is that AI disclosure in academic publishing should start within the research team, long before a manuscript reaches a journal editor.
When collaborators discuss AI use early, they set shared expectations, reduce stigma, and make authorship decisions more transparent. This is especially critical for graduate students and early-career researchers. Power dynamics in academic teams can make it difficult for junior members to admit they used AI, particularly if they’re unsure how senior colleagues will react. The guide argues that normalizing these conversations internally creates a healthier, more honest research culture.
A Framework That’s Overdue
The academic publishing world has been reactive on AI. Journals rushed out policies, many of which were vague, contradictory, or unenforceable. This guide offers something more thoughtful: a set of principles grounded in research integrity and designed to be practical across disciplines and contexts.
AI use in research and writing will only increase. The question was never whether researchers would use these tools. The question is whether the norms around transparency can keep pace. Cleland et al. have given us a solid starting point.
Reference
Cleland, J., Driessen, E., Masters, K., Lingard, L., & Maggio, L. A. (2025). When and how to disclose AI use in academic publishing: AMEE Guide No. 192. Medical Teacher. https://doi.org/10.1080/0142159X.2025.2607513