How do I write an AI disclosure statement for a medical journal? (examples)

What is an AI disclosure statement?

An AI disclosure statement (also called an AI use statement or AI declaration) is a short paragraph in your manuscript that says whether, how and where you used generative AI or AI-assisted tools: chatbots such as ChatGPT or Claude, large language models built into writing tools, and AI image generators. Every major medical publisher now asks for one when AI was used beyond basic spelling and grammar checks.

The rules start with the International Committee of Medical Journal Editors (ICMJE). Its recommendations ask journals to require authors to disclose AI-assisted technologies at submission, and to describe the use in both the cover letter and the manuscript. Publishers then add their own detail on placement, wording, exemptions and images. Because those details differ, the right statement depends on your target journal.

What do ICMJE, COPE and WAME require?

The three bodies that most medical journals follow agree on the principles:

• AI cannot be an author. ICMJE, COPE and WAME all say chatbots and AI tools cannot meet authorship criteria. They cannot take responsibility for the work, approve the final version, declare conflicts of interest or sign licence agreements.
• Use must be disclosed. ICMJE asks for a description in the cover letter and in the manuscript. COPE asks for disclosure in the Materials and Methods (or a similar section), naming the tool and how it was used. WAME (recommendations revised in May 2023) asks for drafting help to be noted in the Acknowledgments with the prompts used, and for AI used in analysis to be described in the Abstract and Methods with the full prompt, the date and time of the query and the tool and version.
• Authors remain responsible for accuracy, plagiarism and attribution, including in AI-generated text and images.
• AI is not a source. ICMJE states that citing AI-generated material as a primary source is not acceptable. Cite the original literature instead; see our guide to AMA citation style.
• Nondisclosure has consequences. ICMJE notes that failure to disclose may require corrective action.

What do the major publishers require? (as of October 2026)

We checked each policy page on 24 October 2026 (the NEJM page via an archived copy dated 31 July 2026, because the live site blocks automated access). Policies change, so open the link for your journal before you submit.

Three practical consequences follow from this table. First, "where" differs: a statement that is correct for an Elsevier journal (end-matter declaration) is in the wrong place for JAMA (Acknowledgment) or NEJM (cover letter and manuscript). Second, exemptions differ: grammar-checker use is exempt at Elsevier, The Lancet, JAMA and Wiley, but Springer Nature asks you to describe even low-risk language editing. Third, the level of detail differs: JAMA and Wiley want versions and dates, so record them while you work.

Where does the AI disclosure statement go?

Decide by what the AI did, then check the journal's rule:

When in doubt, put the full statement where the journal says, and a one-sentence mention in the cover letter. Our guide to the cover letter for journal submission shows where that sentence fits.

Can I use ChatGPT to write a research paper?

You can use ChatGPT or similar tools to help, within limits, as long as you disclose it. None of the policies above bans AI assistance outright for original research. They ban specific things:

• listing AI as an author or co-author;
• letting AI generate results, interpretations or conclusions that you present as your own scholarly judgement (Springer Nature classes this as not permitted);
• fabricated data, citations or images;
• AI-created or AI-altered clinical and research images outside a formal method;
• AI drafting of certain article types (JAMA opinion pieces and letters; Lancet World Report and Perspectives).

The practical risk is accuracy, not permission. Chatbots produce references that look real and are not, and plausible numbers that do not match your data. Elsevier's policy reminds authors that any AI help to select, collate, generate or edit references must be declared. JAMA tells authors not to use AI to generate or format references at all. Every reference must be checked against PubMed or Crossref, and every number against your own output. CliniPaper verifies each reference this way, but the responsibility stays with the authors.

Two more points: check the tool's terms so that it does not keep rights to train on your unpublished manuscript (Elsevier and The Lancet both ask authors to do this), and never put identifiable patient data into a public AI tool.

AI disclosure statement examples (templates)

Adapt these to your journal's required location and level of detail. Replace everything in square brackets, and record versions and dates as you work.

1. Language editing

During the preparation of this manuscript, the authors used [tool name, version, developer] on [date or month and year] to improve the grammar and readability of [sections]. The authors reviewed and edited all suggested changes and take full responsibility for the content of the manuscript.

2. Translation of the authors' own text

The authors wrote the first draft of [sections] in [language] and used [tool, version] to translate it into English. All authors reviewed the translation against the original, corrected it where needed, and take full responsibility for the content.

3. Drafting or restructuring

The authors used [tool, version] to suggest an outline and to draft an initial version of [section] from the authors' notes and results. The authors substantially revised the text, checked every statement against the study data and the cited sources, and take full responsibility for the content. No AI tool was used to generate results, interpretations or references.

4. Literature search or screening (Methods)

We used [tool, version, developer] on [dates] to [screen titles and abstracts / suggest search terms] for [purpose]. Two authors independently verified [all / a random sample of X] AI decisions against the full records. All included references were verified in PubMed.

5. Data analysis or code (Methods)

Statistical code in [R version / Python version] was drafted with the assistance of [tool, version] on [dates]. An author with statistical training reviewed, tested and ran all code; all results were produced by the authors' executed code, not by the AI tool. [Prompts and code are available in Supplementary Material X.]

6. Figure caption

[Figure description]. This schematic was drafted with [tool, version] and revised by the authors; it does not depict patient or research images.

7. No AI use (some submission systems ask you to state it)

The authors did not use generative AI or AI-assisted technologies in the preparation of this manuscript.

Worked example (synthetic): a team writing a retrospective cohort study used a chatbot to polish the English of the Discussion and an AI coding assistant to write R code for a logistic regression model. For an Elsevier journal they would put template 1 in the end-matter declaration and template 5 in Methods. For JAMA, template 1 moves to the Acknowledgment and must add the tool's version, manufacturer and dates of use. For NEJM, both uses would also be described in the cover letter.

The AI statement is one of several end-matter items journals check at submission; the reporting checklist is another, so confirm which reporting guideline your study needs at the same time.

Common mistakes reviewers flag

• No statement at all, while the text reads as machine-written. BMJ states it may screen submissions with detection tools.
• A vague statement such as "AI was used in writing this paper", with no tool, version, purpose or sections.
• The statement in the wrong place for the target journal, or only in the cover letter when the journal wants it in the manuscript.
• AI listed as an author, or a chatbot cited as the source of a fact.
• References generated by a chatbot that do not exist or do not support the claim.
• Images or graphical abstracts made with a general-purpose image generator where the journal forbids it.
• A statement that says the AI did less than it did. Editors treat incomplete disclosure as a problem in its own right: JAMA warns of rejection or post-publication corrective action.

Authors remain responsible for every part of a manuscript produced with AI help. When in doubt, disclose more, not less: Springer Nature states that declaring AI use does not count against a manuscript in editorial decisions.