Do you need to declare AI use to a journal.
Waraq · July 25, 2026 · DESK REJECT FILES
Yes, in most cases you do. If you used a generative AI tool to draft text or build figures for your manuscript, the journal expects a disclosure statement inside the paper, and editorial staff check for it during the first screening read. A generative AI tool here means software that produces new text or images from a prompt, such as ChatGPT or Claude, not a spell-checker that underlines a typo. When the use was declarable and the statement is missing, the paper can be sent back before an editor ever weighs the science.
Why do journals now screen for an AI statement?
Because disclosure shifted from a polite footnote to a checklist item, and an editorial office can verify it in the time it takes to open the file. Screening staff already scan a new submission for the things that get papers bounced without review: scope, ethics approval, formatting, competing-interest statements. The AI statement joined that list. Many rejections happen at exactly this stage, before any peer reviewer is invited, which is why a compliance gap here costs you weeks. If the idea of a pre-review bounce is new to you, desk rejection without review explains what the term covers and what it does not.
The check is quick by design. A reader looking for one paragraph in a fixed section either finds it or does not; there is no judgment call to make. That is what makes the AI statement one of the cheapest things for staff to verify, and one of the easiest for you to get right.
What counts as declarable AI assistance?
Text and figures generated by AI count; routine language correction usually does not. The line most journals draw sits between producing content and cleaning up content you already wrote.
Frontiers in Neurology states the requirement directly in its author guidelines: "If generative AI technology was used to produce written or visual content, this must be acknowledged in the acknowledgements section (and methods section if applicable), listing the name, version, model, and source of the AI technology." Produced is the operative word. Drafting a paragraph, summarizing sources into prose, generating a diagram, or creating an image from a prompt all fall inside that description. Frontiers also makes you responsible for checking the factual accuracy of anything the tool produced, including any quotes, citations, or references it generated. The disclosure does not shift that burden onto the model; you still own every sentence under your name.
Basic grammar and spelling help sits on the other side of the line at many publishers. Elsevier, for one, treats AI-assisted copy editing, meaning improvements to the readability, grammar, spelling, and punctuation of text you authored, as work that does not require a declaration. The distinction turns on who owns the ideas: the tool polished your sentences rather than supplying the substance. When you cannot tell which side a given use falls on, declare it. A disclosed line about a grammar checker costs you nothing; an undisclosed draft written by a model can cost you the submission.
Where does the AI disclosure statement belong?
In the acknowledgements section, and in the methods section too when the tool touched your data or analysis. Frontiers asks for four specifics: the name, version, model, and source of the AI technology. A statement that reads only "AI was used for editing" fails that test, because it names none of the four.
Write it as a short, factual sentence while you assemble the manuscript, not as an afterthought at upload. Keep the tool name, the version number, and what you used it for in one place as you work, so the final statement becomes transcription rather than memory. This is the same habit that keeps reference formatting and author lists clean, and it removes one more thing a screener can flag. A pre-submission check against the target journal's guidelines, of the kind Waraq runs, will surface a missing or under-specified AI statement alongside the other first-read triggers before the editor sees the file.
What happens if the AI use is discovered later?
Caught at screening, undeclared AI use gets the paper returned, often within days. Caught after publication, it can lead to a correction or a retraction. Editorial offices now run their own AI-detection passes, and a mismatch between what the text reads like and what the disclosure claims is the kind of signal that ends a submission early. A fast return is the better outcome of the two; how long a desk rejection takes explains why the pre-review stage moves quickest.
The safer path is dull. Track your tool use as you go, declare anything that produced content, name the four specifics Frontiers asks for, and put the statement where the guidelines say it goes. None of that slows you down, and it removes a reason for the paper to stop before its argument is read. For the wider set of avoidable first-read rejections, the desk reject you can see coming collects the ones worth checking before you submit.