Writing an AI-use disclosure your journal will accept

Somewhere near the end of your next manuscript there is a paragraph — three or four sentences — that declares how, if at all, an AI tool touched the work. It carries more weight than its length suggests: a vague version invites an editor’s follow-up question, and an incomplete one can become an integrity problem after acceptance. The good news is that the paragraph is now largely a solved problem. The major publishers have converged on what they want said, and a team that knows the shape of the statement can write it in a couple of minutes. This is a practical guide to that paragraph: what to declare, where to put it, and the exact words that clear the bar.

What the publishers actually require

Strip away the house-style differences and the requirements reduce to three that nearly every major venue shares. Read them as the spine of any disclosure you write.

• No AI authorship. An AI tool cannot be listed as an author or co-author. Authorship carries accountability — the ability to approve the final version and answer for it — and software can do neither. Springer Nature, Elsevier, Science, the ICMJE recommendations, and COPE all state this in the same terms.
• Disclose generative use. If a tool produced substantive content — prose, a literature summary, an analysis, a figure — that use must be declared, and responsibility for the content stays entirely with the human authors. Elsevier is explicit here: authors who used a generative tool in the writing must include a statement in a dedicated section, and disclose the tool by name.
• Copy editing is exempt.The most useful precedent is Nature’s stated exemption: using AI to improve the grammar, style, and clarity of text the authors wrote themselves does not need to be declared. Elsevier draws the same line — tools that merely improve language and readability fall outside the disclosure requirement. The logic is consistent across publishers: copy editing does not change who generated the ideas, so it does not touch what disclosure exists to protect.

That third point is the one that governs everyday writing. The question a disclosure has to answer is not “did you use AI?” but “did a tool generate content, or assistwith content you generated?” Get that distinction right and the rest of the paragraph writes itself.

Checks, not ghostwriting

A large share of legitimate AI use in research writing is not generative at all. It is checking: does this citation resolve to a real paper, do these reported statistics agree with each other, does the manuscript claim something the cited source does not actually support, is this sentence grammatical in a language that is not the author’s first. None of that is ghostwriting. The intellectual content — the claims, the argument, the interpretation — existed before the tool ran, and the tool’s removal would cost accuracy and polish, not substance.

This matters for disclosure because verification and copy editing sit on the exempt side of every policy we have read. A tool that checks your references against Crossref, flags a statistic that does not add up, or tightens a paragraph you wrote is doing the kind of work journals have never asked authors to declare, whether it was done by a co-author, a professional editor, or software. Ghostwriting is the opposite: content that originates with the tool and enters the manuscript without a human having reasoned it through. The first is a quality practice; the second is the thing disclosure was invented to surface.

The practical test we find most durable: if a reviewer challenged this passage, would your defense rest on your own reading and reasoning, or on the tool’s output?If any part of the honest answer is “the tool’s output,” treat the use as generative and disclose it specifically. If the answer is entirely your own work, checked or polished by a tool, you are in exempt territory — though a brief voluntary note never hurts.

What to declare, concretely

When a use does cross into generative territory, vagueness is the enemy. “AI was used in the preparation of this manuscript” is the statement editors dislike most, precisely because it could mean anything from fixing commas to drafting the discussion. A disclosure that reads as care rather than confession names four things:

• Which tool — the product and, where it matters, the version, so the statement is specific and reproducible.
• For which part — the section or task, not the whole paper: language editing throughout, or drafting a first pass of the introduction, or generating a figure.
• What the tool did — assisted with language, summarized a body of literature, produced a draft from an outline, checked references.
• Who is accountable — a plain statement that the authors reviewed, verified, and take responsibility for the content. Every policy makes this the load- bearing clause.

Where the statement goes

Placement is not decoration — putting the disclosure in the wrong section is a common reason production teams send a manuscript back. Three locations cover almost every venue, and the target journal’s author guidelines will name the one it wants:

• A dedicated declaration section.Elsevier asks for a distinct “Declaration of Generative AI in the writing process” statement placed just before the references. Many journals now provide a named section like this; when they do, use it rather than tucking the disclosure elsewhere.
• Methods or acknowledgements. Where a tool contributed to the research or analysis, the methods section is the right home; where it assisted with the writing, acknowledgements is conventional. Nature-family journals point authors to methods or acknowledgements depending on the nature of the use.
• The cover letter. Not a substitute for the in-manuscript statement, but a good place to give editors a one-line summary and signal that you have handled disclosure deliberately. It reads well and costs a sentence.

A template you can adapt

Most teams need three versions of this paragraph, matched to how they actually worked. Each is a starting point — edit it to the truth of your manuscript and to your target journal’s named section.

Assistive use only (usually exempt, stated voluntarily):“During the preparation of this work the authors used [tool] to improve the language and readability of author-written text and to verify that references resolve to their cited sources. No content, analysis, or interpretation was generated by AI tools. The authors reviewed and edited the manuscript as needed and take full responsibility for its content.”

Generative use, disclosed specifically:“During the preparation of this work the authors used [tool, version] to draft a first version of the [section] from an author-prepared outline. Every generated passage was checked against primary sources, all citations were resolved and verified, and the authors revised the text and take full responsibility for the content of the publication.”

No AI use at all:“No generative AI tools were used in the research or in the preparation of this manuscript.” Only claim this if it is true for everyauthor — one co-author’s undisclosed drafting assistant makes the statement false for the whole list.

Write it from the record, not from memory

The hardest part of an honest disclosure is not the wording — it is remembering, months after the fact and across several co-authors, what each tool actually did in each section. Reconstructing that at submission is where statements go wrong: something gets over-claimed out of caution, or quietly omitted because nobody remembered it. The teams that handle this cleanly keep a running record as they write — which tool, which section, assistive or generative, who verified the output — so the disclosure paragraph is a summary of the log rather than a feat of collective recall.

This is where the tooling earns its place. In Dissertatio, every AI interaction on a manuscript — what was read, what was suggested, what a human accepted — is recorded automatically and exportable as a per-manuscript usage report, so the disclosure statement is written from evidence and an editor’s follow-up question becomes a lookup rather than an argument. The same design keeps the assistant on the exempt side of the line by default: it verifies references, checks claims against sources, and polishes author-written text; it does not write the paper for you. Whatever tools your team uses, the principle is the same — record at the moment of use, and disclose from the record.

A disclosure statement is easy to treat as a compliance chore, but that framing sells it short. Written well, it is a credibility signal: it tells an editor that your team knows exactly how its manuscript was made and is not hiding any part of it. In a literature turning up fabricated references at uncomfortable rates, that provable, specific knowledge is quietly becoming part of what “rigorous” means.