Skip to main content

Free guide

Use AI Without Getting Rejected: what journals actually require you to disclose

6 min read · verified August 21, 2026

Most advice about AI in papers is somebody guessing. This is not. Every rule below was read off the publisher or committee page that sets it, on the date above, and every one of them is linked at the end so you can check it yourself.

The useful finding is that there is no universal rule. One thing is agreed everywhere, which is that AI cannot be an author. Everything else, and especially WHERE you are supposed to disclose, changes by publisher. Guessing is not punished with a stern note. PLOS calls a missing disclosure a misrepresentation of your methods.

Yours to keep, free

The whole guide is on this page. The PDF is the same thing, formatted to print or keep on your phone.

Download the PDF (PDF file)

01The one rule nobody disagrees about

AI cannot be an author. Not a co-author, not an acknowledged contributor with a byline, not "with assistance from". Every source we checked says the same thing and gives the same reason.

  • The reason is accountability, not snobbery. Authorship means answering for the work, and a tool cannot. ICMJE puts it as AI being unable to take responsibility for the accuracy and integrity of the work.
  • Elsevier states it as plainly: authorship implies responsibilities and tasks that can only be attributed to and performed by humans.
  • This is the one place where there is nothing to check journal by journal. If you were considering it, do not.

02Where you disclose depends on what you used it for

This is the part that catches people, because it is not intuitive and it is not the same everywhere. The same tool goes in a different section depending on the job it did.

  • ICMJE splits it by purpose: AI used for WRITING assistance is reported in the acknowledgements, while AI used for data collection, analysis, or figure generation is described in the METHODS.
  • PLOS wants a dedicated section of the Methods, or the Acknowledgements for article types that have no Methods section, and wants three things named: which tools, how you used and evaluated them, and which parts of the study or article were affected.
  • Elsevier wants something different again: a separate statement headed "Declaration of Generative AI and AI-assisted technologies in the writing process", placed immediately above the references.
  • So the honest instruction is to check your target journal before you write the statement, not after. The shape of the disclosure is journal-specific even when the duty to disclose is not.

The disclosure statement, filled in

  • Name the tool and version: "ChatGPT (GPT-5, OpenAI)" beats "an AI tool".
  • Name the job it did, in the journal’s own vocabulary: language editing, code generation, figure production, literature screening.
  • Name what you did to check it. This is the sentence that separates a disclosure from a confession.
  • Say plainly that the authors take responsibility for the content. Every policy is built around that idea, so say it in words.

This shape follows the Elsevier declaration, which is the most prescriptive of the ones we read, so it is the safest default. Move it or rename it to match your target journal, and move anything touching data or figures into the Methods.

03Grammar and spelling are usually exempt. Usually.

You do not have to declare a spellchecker, and the policies say so, but the line between "language editing" and "writing" is where people get this wrong.

  • Elsevier says basic grammar and spelling checks need no disclosure. That is an explicit carve-out, not an inference.
  • The carve-out is narrow. A tool that rewrites your discussion into fluent prose is not doing a spelling check, whatever the marketing calls it, and that is the use the disclosure rules are written for.
  • The practical test: if the tool changed what the sentence CLAIMS, or produced sentences you would not have written, that is beyond a spellcheck. Disclose it.

04Images are the hard line, and the one people break

Text policy is about disclosure. Image policy is frequently about prohibition, and that difference is the most expensive thing in this guide.

  • Elsevier does not permit generative AI to create or alter images in submitted manuscripts. Not "disclose it", not "with permission". Not permitted.
  • What remains allowed is ordinary processing: adjustments of brightness, contrast, or colour balance, provided they do not obscure or eliminate any information present in the original.
  • There is one exception and it is narrow: AI image tools used as part of the research METHODOLOGY itself, documented fully in the methods.
  • This is where a casual "I just cleaned up the figure" becomes an integrity case rather than a correction, because image manipulation is screened for and is treated as misconduct rather than as an oversight.

05What happens if you do not disclose

The consequence is not a note asking you to add a line. It is categorised as a misrepresentation, which puts it in the same family as misreporting your methods.

  • PLOS states that noncompliance will be considered misrepresentation of methods, contributions, or results, and lists rejection, retraction, or an editorial notice as outcomes.
  • That framing matters more than the penalty. A missing disclosure is not treated as a formatting slip you can fix at proof, it is treated as having described your work inaccurately.
  • The asymmetry is what should decide it for you: disclosing costs you four sentences and no reviewer has ever rejected a paper for being transparent about language editing. Not disclosing risks the paper and the record.

06There is no single Springer policy, and that is the point

The instinct is to look up "the publisher’s policy" and comply once. For the largest publishers that does not work, and assuming it does is how people end up compliant with the wrong document.

  • Springer Nature’s public overview sets principles rather than procedures: human accountability cannot be transferred to AI, AI may support but not replace scholarly judgment, transparency is expected. It then points to the individual policies for BMC, Nature Portfolio, Springer, Discover and Palgrave.
  • So "I followed Springer’s AI policy" is not a statement that means anything on its own. The imprint you are submitting to has its own, and that is the one you are held to.
  • Do the lookup on the journal’s own author guidelines page, at the point of submission, and screenshot it. Policies here have changed repeatedly since 2023 and yours will have changed again by your next paper.

Where to check

The databases this guide sends you to, all free and open without an account.

Where these numbers come from

  1. 01AI cannot be an author; writing assistance goes in the acknowledgements, data and figure work goes in the methods ICMJE Recommendations, Defining the Role of Authors and Contributors
  2. 02A separate declaration above the references; grammar and spelling checks exempt; generative AI may not create or alter images Elsevier, The use of generative AI and AI-assisted technologies in writing
  3. 03A dedicated Methods section for AI contributions, and noncompliance treated as misrepresentation PLOS, Ethical Publishing Practice
  4. 04Principles rather than procedures, with the operative policy left to each imprint Springer Nature, Editorial Policies

This guide will go out of date

I will tell you when it does

One email a week on what changed in academic publishing and whether it actually matters for your paper. Every number links its source.

One email a week, Mondays. Leave whenever you like.

The next step

A policy tells you the rule. It cannot tell you if your paper reads as AI.

You can follow every disclosure requirement here and still get a reviewer who says the writing feels machine-made, because that judgement is about the prose and not about the paperwork. That is a different problem and it is the one we see most in the manuscripts we review: hedged, even, frictionless sentences that say less than the author knows. The Academy works the fix on your own draft, one task at a time, and it starts with making your claims specific enough that no tool could have written them.

Get published, one task at a time.

See what the Academy is