How to Disclose AI Use in a Research Paper

The AI disclosure statement was missed, and the editor at a mid-tier journal sent the paper back with one line in the decision letter: the AI disclosure statement is missing, please add it before resubmission. The methods were sound, the data checked out, and the delay cost the author six weeks anyway, because the paper had to re-enter the queue behind submissions that got the mechanics right the first time.

How to disclose AI use in a paper is a narrower question than whether AI use is allowed at all, and it is the one that actually determines whether an editor sends a manuscript back. Three things decide it every time: where the statement physically sits in the manuscript, what it has to name, and whether the use in question crosses the threshold that makes disclosure necessary in the first place. This piece covers journal articles specifically, not coursework and not a thesis, where the structure and the stakes are different.

TextPulse's wider map of university and journal AI policy covers what each publisher permits and forbids in full, publisher by publisher. What follows here assumes AI use is already allowed for your paper, and answers the mechanical question that comes right after: how do you actually write and place the statement so an editor accepts it on the first pass?

How to Disclose AI Use in a Paper: Where the Statement Goes

A disclosure statement only does its job if an editor can find it where the journal expects it. Four locations account for nearly every journal currently in print, and which one applies depends entirely on the publisher, not on the discipline or the country.

Location in the manuscript	Used by	What else typically sits there
A dedicated declaration section, placed immediately before the reference list	Elsevier	Funding statements, conflict of interest declarations
The Methods section, inside the normal description of how the work was done	Nature specifically	Study design, procedures, materials
The Acknowledgements, Introduction or Preface	Springer Nature's group-wide guidance; Wiley, as one of its accepted options	Thanks to colleagues, funding sources
The acknowledgments section, naming the AI system and the sections it affected	IEEE	Funding, technical support credits
A standalone disclosure statement, separate from every other section	Wiley, as its other accepted option	Nothing else, it exists only for this

Two placements do double duty, sitting inside a section that already exists for something else, and two stand alone. Neither pattern is more correct than the other; each is simply what a specific publisher's own instructions ask for, which is why checking the actual target journal's guide for authors matters more than assuming last year's journal used the same convention.

What Has to Be Named in an AI Disclosure Statement?

Location answers where; naming answers what. Every policy that requires disclosure at all asks for the same core information, regardless of which section it lives in: the specific tool and its version, not a general category like AI, what it was used for, and confirmation that a human reviewed the output before it reached the manuscript.

Granularity is where most disclosure statements fall short. IEEE's own instructions ask specifically for the sections of the article that used AI-generated content to be identified individually, with a brief explanation of the level of use for each one. A paper that used one tool to translate a non-native author's draft and a different tool to generate a code snippet for a supplementary analysis needs both named separately, not folded into one generic sentence about using AI somewhere during preparation.

Wiley's instructions go a step further and ask authors to state whether the AI output influenced key arguments or conclusions, not just whether a tool was used somewhere in the process. That distinction matters practically: a tool that reworded a paragraph sits in a different category from one that helped shape what the paragraph argues, and a statement that blurs the two under-discloses even when a statement is technically present.

If your target journal's reference list style expects the tool cited rather than just named in a sentence, a citation generator built for AI tools formats that entry the way the rest of your reference list already looks, version number included.

What Counts as AI Use You Must Disclose?

Not every pass through an AI tool crosses the line into disclosable use, and getting this threshold wrong in either direction causes real problems. Over-disclosing buries an editor in irrelevant statements, and under-disclosing is the mistake that gets a paper sent back or, worse, flagged after publication. The threshold itself is more consistent across publishers than the placement rules are.

Basic language checks sit below the line almost everywhere. Elsevier states that basic checks of grammar, spelling and punctuation do not need a declaration statement, and Wiley draws an equivalent line for tools used solely for spelling, grammar and general editing. A grammar checker used the way a human copyeditor has always been used, correcting mechanical errors without changing meaning or argument, stays inside that exemption under both publishers' own wording.

But above that line, you'll need to come clean quickly. All the specific uses, generating draft text, summarizing sources, suggesting structure, writing or debugging code, analyzing data, or producing a figure, are the kind of thing every major publisher asks you to name specifically, tied to the part of the paper the tool actually touched, rather than folded into one blanket mention that AI was involved somewhere in the process.

Writing the Statement Itself

The wording does not need to be original. Elsevier's own suggested phrasing is a reasonable starting point for any journal, not only Elsevier's: "During the preparation of this work the author(s) used [NAME OF TOOL/SERVICE] in order to [REASON]. After using this tool/service, the author(s) reviewed and edited the content as needed and take(s) full responsibility for the content of the publication." Replace the bracketed sections with the actual tool, version and reason, and the sentence covers most of what any publisher's disclosure requirement asks for.

Different situations call for different specifics inside that same shape. Language editing, literature search assistance, code generation, data analysis and image generation each name a different reason and often a different tool, and a statement that names the wrong one is barely more useful to an editor than no statement at all. Write the specific reason a reviewer would actually want to know, not a general description that could apply to any paper submitted that year.

Does the Statement Need Updating After a Revision?

A manuscript that started as a fully human draft and only picked up AI assistance during the response to reviewers still needs the statement added at that point, not left out because the original submission did not use AI. Editors read the disclosure as a statement about the version in front of them, the one being decided on, not a historical record of the very first draft.

The reverse also holds. A tool used heavily in an early draft but removed entirely from the version actually submitted does not need disclosing, because the statement describes the manuscript as submitted, not everything that happened during drafting. When in doubt, describe the paper in its current state, and update the statement every time that state changes materially.

What Happens If You Don't Disclose AI Use?

The consequence depends on when the omission surfaces. Caught during editorial screening or peer review, the usual outcome is a request to add the disclosure before the manuscript moves forward, the same six-week delay described at the start of this piece. Caught after acceptance but before publication, some journals still allow a correction at proof stage rather than restarting review from the beginning.

Caught after publication, the options narrow and get more serious. A journal can publish a correction if the underlying work is otherwise sound and the omission was the only problem, or pursue a retraction if the undisclosed AI use is bound up with other integrity concerns, such as content that turns out to be fabricated or plagiarized rather than simply AI-assisted and unreported. Editors generally treat undisclosed AI use as a transparency failure first, then investigate whether it is anything worse, which is also why disclosing early costs so much less than being caught.

TextPulse's own stance on ethical AI use in academic writing sets out the same distinction from the other side of the desk: editing your own argument and disclosing it is a different act from submitting an argument you never made in the first place.

A ready disclosure statement template saves drafting the wording from scratch, and disclosing AI use in a thesis follows different conventions again, so it has its own page.

Check your specific target journal's current guide for authors before you write the statement, not after. A publisher's own wording changes over time, and a statement copied from a paper published two years ago may already be missing something the same journal now asks for.