How to Write an AI-Use Disclosure Statement for Your Manuscript

A medical journal desk-rejected a paper from a colleague last month. The reason wasn't the science. It was a missing AI-use disclosure statement. The methods section said nothing about AI tools, but the cover letter mentioned that the team had used ChatGPT to draft parts of the introduction. The editor flagged the mismatch and bounced the paper back unread.

This is the new reality. Elsevier, Springer Nature, Wiley, the ICMJE (which governs most medical journals), and Horizon Europe have all formalized AI-use disclosure rules over the past two years, and most other publishers have followed. The penalty for not disclosing — or for disclosing wrong — is no longer a polite request for revision. It's desk rejection, sometimes retraction. This guide walks through what to disclose, what doesn't need disclosing, where the statement goes, template wording for the common scenarios, and the field-specific quirks that catch researchers out.

Why this matters now

The policy landscape changed faster than most authors noticed. The short version of where we are in mid-2026:

ICMJE (medical journals). Authors must declare any use of AI in writing or analysis. AI cannot be listed as an author. The statement goes in the methods section if AI was used in analysis, and in acknowledgments if AI was used only for writing assistance.

Elsevier. Disclosure required for any AI use beyond basic grammar correction. Goes in a dedicated "Declaration of generative AI and AI-assisted technologies in the writing process" section before the references.

Springer Nature. Required disclosure of large language model use in the methods section. Specific scope: anything beyond standard editing tools (Grammarly-class) must be disclosed.

Wiley. Required statement in the acknowledgments or methods, depending on the role AI played. Templates provided in author guidelines.

Horizon Europe (EU funding). Disclosure required in the proposal itself, with implications for how AI-edited grant text is evaluated. Failure to disclose is treated as a research integrity issue.

Major preprint servers. arXiv, bioRxiv, and medRxiv all now ask for AI-use declarations on submission. Not enforced as strictly as journals, but increasingly checked during moderation.

If your target venue isn't named above, check its instructions for authors — almost every English-language journal has updated guidance since late 2024. Assume disclosure is required and verify the format.

What you need to disclose

The line journals draw runs roughly between "tool that catches errors in what you wrote" and "tool that generated text you used." Grammar checkers and spell-check tools are usually fine without disclosure. Anything else, including the use of an AI editor on your own draft, generally needs to appear in the statement.

Always disclose. Use of any large language model (ChatGPT, Claude, Gemini, DeepSeek, Llama variants) to generate text that appears in the manuscript, in any quantity. This includes summaries, transitional sentences, abstract drafts, and any prose generated from prompts.

Always disclose. Use of AI for analysis, including statistical analysis, image processing, text mining of source documents, or screening for systematic reviews.

Usually disclose. Use of AI editing tools that paraphrase, restructure, or substantially rewrite text — even if you wrote the original. This includes humanizers, paraphrasers, and the deeper rewrite modes of academic editing tools.

Disclose if asked. Use of basic grammar tools (Grammarly's free tier, Microsoft Editor, the spell-check in your word processor). Most journals do not require disclosure for these, but Elsevier and Springer have started asking for it explicitly. If the author instructions mention "any AI tools," include them to be safe.

Don't need to disclose. Reference managers, citation generators (Zotero, Mendeley), or LaTeX equation editors. These are not generative AI.

The gray area in practice is academic editing tools that combine traditional grammar checking with generative rewriting. If your tool's output substantially changes the structure or wording of your sentences, treat it as generative and disclose. If it only flags errors and suggests punctuation fixes, traditional grammar-tool treatment applies. When in doubt, disclose — the cost of over-disclosing is zero; the cost of under-disclosing is potentially retraction.

Where the statement goes

Placement depends on what role AI played in your work.

AI used only for language editing or polishing. Acknowledgments section, or a dedicated "Declaration of generative AI and AI-assisted technologies in the writing process" before references (Elsevier's preferred format).

AI used to draft sections of the manuscript. Same placement as above, with more specificity about which sections.

AI used in analysis or methods. Methods section, in a subsection describing the AI tool, its parameters, and how it was used. The acknowledgments statement may also be needed.

AI used in systematic review screening or data extraction. Methods section, typically in a subsection on screening procedures. The PRISMA-trAIce checklist (the AI-specific extension of PRISMA 2020) requires explicit reporting of which AI tool, version, prompts used, and how human verification was performed.

AI used in grant proposal drafting. Depends on the funder. NIH and NSF currently accept disclosure in the cover letter or a dedicated declaration. Horizon Europe requires it in the proposal text itself, in a specific section. Always check the funder's current guidance, which changes faster than journal guidance does.

Templates for the common scenarios

These templates match the wording most major publishers accept. Adapt the tool name and version to your actual usage.

Scenario 1: AI used for grammar and language editing only.

Declaration of generative AI and AI-assisted technologies in the
writing process: During the preparation of this work, the authors used
[Tool Name, version] for language editing and grammar refinement.
After using this tool, the authors reviewed and edited the content as
needed and take full responsibility for the content of the publication.

Scenario 2: AI used to draft specific sections, then edited by authors.

Declaration of generative AI and AI-assisted technologies in the
writing process: During the preparation of this work, the authors used
[Tool Name, version] to draft initial versions of the [Introduction
/ Abstract / Discussion section]. The authors substantially revised
the generated text, verified all factual content against the source
material, and take full responsibility for the content of the
publication.

Scenario 3: AI used in methods (e.g., text mining, systematic review screening).

We used [Tool Name, version] to [screen abstracts for inclusion in
the systematic review / extract study characteristics from full-text
articles / perform topic modeling on the source corpus]. The tool was
accessed via [API / web interface] between [dates] using the following
prompt template: "[exact prompt]". Two reviewers independently verified
all AI-generated outputs against the source documents, with
disagreements resolved by discussion. No AI was used in final inclusion
decisions, quality assessment, or interpretation of findings.

Scenario 4: AI used for humanization or paraphrasing of own writing.

Declaration of generative AI and AI-assisted technologies in the
writing process: During the preparation of this work, the authors used
[Tool Name, version] to refine the language and clarity of text the
authors had originally drafted. The tool did not introduce new ideas,
arguments, or content. The authors reviewed all suggested changes and
take full responsibility for the content of the publication.

Scenario 5: No AI was used.

Some journals now require an explicit negative statement.

Declaration of generative AI and AI-assisted technologies in the
writing process: The authors did not use any generative AI or
AI-assisted technologies in the preparation of this work.

Field-specific quirks

The general rules above apply across most fields. A few areas have stricter or earlier-adopted requirements.

Clinical medicine. Specific journals (NEJM, JAMA, Lancet) sometimes require disclosure of AI tools used by reviewers as well as authors. Some require explicit confirmation that no patient data was input into a hosted AI tool. Check the specific journal — guidance varies even within ICMJE-compliant medicine.

Computer science. Conferences (NeurIPS, ICML, ACL) have published AI-use policies that often differ from each other. NeurIPS requires disclosure in the paper itself; ACL requires it in a separate field on submission. Some venues prohibit AI-generated text in specific sections (often the abstract or contributions). Read the call for papers carefully — these policies change yearly.

Humanities and qualitative social sciences. Disclosure norms are catching up. Several humanities journals now require disclosure even for editing-class AI use, on the grounds that voice and prose are part of scholarly contribution in these fields. Treat humanities disclosure rules as stricter than STEM defaults.

Law. Several law reviews now prohibit any AI-generated text in the body of the article (allowing AI editing of author-written prose). Citation accuracy is checked aggressively, in part because of high-profile cases of AI-fabricated citations appearing in legal briefs.

Common mistakes that trigger rejection

We've seen each of these cause delays or rejections.

Vague disclosure. "AI tools were used" without naming them, naming versions, or describing what they did. Editors increasingly want specificity, and vague disclosure looks evasive.

Disclosure that contradicts the cover letter or methods. The example at the start of this article. If your cover letter mentions ChatGPT and your declaration omits it, editors flag the mismatch as a research integrity concern.

Listing AI as an author. ICMJE, COPE, and major publishers all prohibit this. AI cannot meet authorship criteria because it cannot accept accountability. Listing ChatGPT as a co-author is now an instant desk-rejection trigger at most journals.

Disclosure in the wrong location. A disclosure buried in a footnote when the publisher requires a dedicated section is treated as non-compliance, even if the substance is correct. Match the format the publisher specifies.

Forgetting the responsibility statement. Most templates end with some form of "the authors take full responsibility for the content." Missing this language is treated as incomplete disclosure by some publishers.

Disclosing only one of several tools. If you used ChatGPT to draft, then ProofreaderPro to edit, then DeepL to translate a section — disclose all three. Selective disclosure is worse than no disclosure when discovered.

The workflow we recommend

A practical sequence that minimizes both work and risk.

Step 1: Keep a log as you write. A simple text file: tool name, version, what you used it for, which sections. Update it every time you use AI. This single habit makes the disclosure write itself at the end.

Step 2: Match the publisher's template. Check the author guidelines for the venue you're submitting to. Use their exact format if they provide one. Most publishers prefer their own wording.

Step 3: Coordinate with co-authors. Every co-author needs to know what AI you used and to confirm their own AI use was disclosed. A surprise revelation during peer review damages the team's credibility.

Step 4: Write the cover letter consistently. Your journal cover letter should reference the disclosure briefly if AI use was substantial. Mismatches between cover letter and declaration are a common rejection trigger.

Step 5: Pre-flight the disclosure with a senior author. If your PI or senior co-author hasn't read the disclosure language, get their sign-off before submission. They've usually seen what publishers want.

See the Full AI Proofreader

Tracked-changes editing, citation-aware corrections, and clear visibility into what changed. Free tier includes every feature.

Frequently asked questions

Q: Do I need to disclose using Grammarly or my word processor's spell-check?

For most journals, no. Traditional grammar checkers and spell-check tools are not classed as generative AI. The line some publishers (notably Elsevier and Springer) are starting to draw is whether the tool changes the structure or substance of your text. A spell-check fixing typos is fine. A "rewrite this sentence for clarity" feature triggers disclosure. When in doubt, check the publisher's instructions and disclose if the wording is ambiguous.

Q: What if I used an AI editor that I now realize substantially rewrote my text?

Disclose it. The cost of late disclosure is much lower than the cost of being caught not disclosing during peer review. If the manuscript is still in revision, add the declaration in your response letter along with the revised submission. If the paper has been submitted but not yet reviewed, contact the editor with a brief correction. Editors generally respect corrections; they don't respect omissions discovered later.

Q: Can I list ChatGPT as an author if it wrote substantial parts of the paper?

No. The ICMJE, COPE, and every major publisher prohibits AI authorship on the grounds that AI cannot meet authorship criteria — it cannot accept responsibility for the work, cannot agree to be accountable for accuracy, and cannot approve final versions. Listing AI as an author is now grounds for instant desk rejection at most journals and may trigger research integrity proceedings at your institution. Disclose AI use in the declaration; never in the author list.

Q: Does my response-to-reviewers letter also need an AI-use disclosure?

It should match what you put in the manuscript. If you used AI to help draft your response letter (which is common and acceptable), update your disclosure to reflect that. Many revised manuscripts now include a sentence like "AI tools were used in the preparation of this revision, including [tool] for editing the response letter, in addition to the uses disclosed in the original submission." Editors are increasingly aware of AI-assisted response letters and prefer disclosure over discovery.