Most researchers assume an AI disclosure statement is just a sentence admitting “I used ChatGPT.” It isn’t. Every major publisher policy that has actually operationalized the ICMJE’s 2023 generative-AI position — Nature Portfolio, JAMA Network, BMJ, IEEE — asks for more: which tool, which version or approximate date of use, what specific task it performed, and confirmation that a named human author reviewed and takes responsibility for the output. A vague “AI assistance was used” line is the single most common reason these get flagged back to authors for revision.
This page is a bank of ready-to-use statement wording, organized two ways: by the publisher policy family it needs to satisfy, and by the specific task the AI tool performed. Copy the block that matches your situation, fill in the brackets, and check it against your target venue’s own author instructions before submitting — publisher wording changes, and the specific placement requirement (Methods vs. Acknowledgments vs. a dedicated AI-use field in the submission system) is usually non-negotiable even when the underlying principle is the same everywhere.
The five elements a compliant disclosure statement needs
Regardless of which publisher family you’re writing for, a disclosure statement that won’t get kicked back has all five of these:
- The tool’s name and developer — “ChatGPT (OpenAI),” not just “AI” or “a chatbot.”
- Version or date of use — models change fast enough that “GPT-4, accessed March 2026” is materially different information from “GPT-3.5.” If the interface doesn’t expose a version number, state the month and year of use instead.
- The specific task performed — “used to improve readability of the Discussion section” or “used to generate initial R code for the mixed-effects model, which was then reviewed and modified by the authors,” not “helped write the paper.”
- A statement of human accountability — that a named author reviewed the AI-assisted content/output and takes responsibility for its accuracy. No AI tool can be listed as an author under any major publisher’s current policy, precisely because it cannot hold this kind of accountability.
- Correct placement — the venue’s own instructions, not habit, decide whether this belongs in Methods, Acknowledgments, a cover letter, or a dedicated AI-use field at submission. Placing it in the wrong section is a common reason for an editorial query even when the wording itself is fine.
Templates by publisher policy family
Most journals fall into one of a small number of policy families. Match your target journal’s own instructions to the closest family below — if a journal explicitly names ICMJE, JAMA, or a specific publisher’s policy in its author guidelines, use that family’s template as the starting point.
Generic ICMJE-aligned journals (the most common default)
Most journals that simply state “we follow ICMJE recommendations” without further detail accept a statement along these lines, typically placed in the Acknowledgments or a dedicated AI-use declaration:
“During the preparation of this manuscript, the author(s) used [tool name, e.g. ChatGPT-4, OpenAI] to [specific task, e.g. improve the readability and language of the manuscript]. After using this tool, the author(s) reviewed and edited the content as needed and take(s) full responsibility for the content of the publication.”
JAMA Network / AMA Manual of Style
JAMA’s guidance (Flanagin et al., JAMA, July 2023) is unusually specific about what the disclosure must name, and it also draws a line most other policies leave implicit: routine grammar and spell-check tools are explicitly excluded from disclosure. For substantive generative-AI use, the required elements map onto a generative AI disclosure statement placed in the Acknowledgment or Methods section:
“[Tool name, version, e.g. ChatGPT, version 4o, OpenAI] was used to [describe role, e.g. draft an initial summary of the literature search results] on [month, year]. The authors reviewed, edited, and verified the accuracy and appropriateness of the output, and take full responsibility for the content of this publication.”
Nature Portfolio
Nature Portfolio’s policy requires generative-AI use in text generation to be documented in the Methods section (or a suitable alternative if no Methods section exists), and is notably stricter than most competitor policies on AI-generated images and video, which it currently states it is unable to permit outside narrow, editor-reviewed exceptions.
“Text generation: [tool name] was used to [specific task, e.g. generate a first draft of the Introduction based on author-provided notes and references]. All AI-generated text was reviewed, fact-checked, and edited by the authors prior to submission.”
Do not use this template for figures or images — check the current Nature Portfolio AI policy directly before including any AI-generated visual content, since that specific restriction is stricter than the text-disclosure norm and changes are more likely there than in the text-disclosure wording.
BMJ
Since April 2024, BMJ Group has required a structured AI-use disclosure step within its ScholarOne submission system, operationally distinct from (though consistent with) the general ICMJE position. Because the disclosure is captured through a structured intake field rather than free text, treat the wording below as what typically also needs to appear in the manuscript itself, not a substitute for completing BMJ’s own submission-system prompts:
“Generative AI tools were used in the preparation of this manuscript as follows: [tool name and version] was used for [task]. No content generated by AI was included without review and verification by the authors, who take full responsibility for the accuracy of the final manuscript.”
IEEE
IEEE has no single, centrally branded generative-AI policy document — guidance is distributed through the IEEE Author Center plus parallel guidance from individual technical societies, with substance consistent across venues but exact wording varying by journal or conference. The safe default is a general acknowledgment-style statement, checked against your specific IEEE venue’s own author instructions:
“The author(s) used [tool name] to assist with [task, e.g. checking grammar and improving clarity of language] in this manuscript. This tool was not used to generate original research content, data, or conclusions. The author(s) reviewed and edited all output and take full responsibility for the content of this publication.”
Templates by task
Some venues ask authors to describe AI use by task rather than by filling a single generic template. These variants slot into any of the publisher templates above in place of the bracketed “[task]” text.
Drafting or restructuring text
“[Tool name] was used to generate an initial draft of [section] based on author-supplied notes, data, and outline. The authors substantially revised this draft, verified all factual claims and citations independently, and take full responsibility for the final text.”
Language editing and readability
“[Tool name] was used to improve the readability and language of the manuscript. No new content, data, or scientific claims were introduced by this process; the authors reviewed all suggested edits before acceptance.”
Literature search or summarization assistance
“[Tool name] was used to assist in summarizing candidate references identified through conventional database searches. All cited sources were independently verified by the authors to confirm they exist and accurately support the statements attributed to them.”
This verification step is not optional wording — fabricated or nonexistent AI-generated references are treated as a research-integrity issue, not a formatting one, under current NIH guidance on AI-related conduct in funded research (see the funder section below).
Data analysis or code generation
“[Tool name] was used to generate initial code for [specific analysis, e.g. the mixed-effects regression model reported in Table 2]. The code was reviewed, tested, and modified by the authors, who verified the analytic approach and results independently of the AI-generated output.”
Image or figure generation
Check your target journal’s policy before using this template at all — several major publishers, including Nature Portfolio, restrict or prohibit AI-generated images and figures outright rather than simply requiring disclosure. Where a venue does permit disclosed AI-assisted image work (typically limited to non-generative enhancement of an existing figure, not fabricated imagery):
“[Tool name] was used to [specific task, e.g. adjust contrast and remove background artifacts from] Figure [X]. No pixels were added, removed, or generated by AI beyond this documented adjustment; the original unedited image is available from the corresponding author on request.”
Translation assistance
“[Tool name] was used to assist with translation of the manuscript from [source language] to English. The translated text was reviewed for accuracy by a fluent speaker of both languages among the author team, who take full responsibility for the final wording.”
What not to disclose
Over-disclosure dilutes the statement and can read as evasive. Per JAMA Network’s explicit guidance, routine spelling and basic grammar-check functionality — the kind built into standard word processors, not a distinct generative-AI drafting tool — does not need to be disclosed. The practical test most policies apply is whether the tool generated content (text, code, an image, a summary) versus merely flagging errors in content the author already wrote. If you’re unsure which side of that line a specific tool falls on, disclose — a redundant disclosure is a minor style note; an undisclosed one is a policy violation.
Grant applications and institutional submissions
Funder policy on AI disclosure is considerably less standardized than journal policy, and — as of August 2026 — no major funder has published a required disclosure-statement template comparable to the journal templates above. NIH’s guidance to date has focused on where AI use crosses into research misconduct (presenting AI-generated data as empirically obtained, undisclosed AI alteration of images, citing nonexistent AI-generated references, or an application “substantially developed by AI”) rather than prescribing specific disclosure wording for legitimate, disclosed use. See CASRAI’s coverage of NIH’s May 2026 notice for the misconduct-threshold distinction, and AI Disclosure Laws vs. Publisher Policy for how legally mandated disclosure (EU AI Act, US state transparency laws) differs from funder or publisher policy.
Absent a funder-specific template, a general-purpose institutional or thesis/dissertation statement — adapted from the ICMJE-aligned wording above — covers most cases:
“In preparing this [thesis/dissertation/application], the author used [tool name] to [specific task]. All AI-assisted content was reviewed and verified by the author, who takes full responsibility for the accuracy and originality of the submitted work, consistent with [institution]’s policy on the use of generative AI in academic work.”
Check your institution’s or funder’s current policy directly before submitting — this is one of the fastest-moving areas of research policy, and a growing number of institutions have published their own AI-use policies separate from any journal or funder requirement.
Common mistakes
- Listing the AI tool as a co-author or contributor. No major publisher policy permits this; AI tools cannot take responsibility for a work’s accuracy or hold accountability, which every current authorship definition requires.
- Vague task descriptions. “AI assistance was used to help with this paper” fails the specificity requirement in essentially every publisher family above.
- Omitting the tool version or date. Reviewers increasingly ask for this, since model capability differs meaningfully between versions of the same tool.
- Disclosing only in the cover letter, not the manuscript itself. Most policies require the disclosure to appear in the published record (Methods, Acknowledgments, or an equivalent section), not only in correspondence with the editor.
- Reusing identical wording across venues with different placement rules. The underlying content can often stay similar; where it goes in the manuscript usually cannot.
Frequently asked questions
Do I need to disclose AI use if I only used it for grammar or spell-checking?
Under JAMA Network’s guidance, and the general practice most other policies follow, routine grammar and spell-check functionality does not require disclosure. Tools that generate, draft, or substantially rewrite content do.
Where does an AI disclosure statement go in a manuscript?
It depends on the venue: most ICMJE-aligned journals place it in Acknowledgments; Nature Portfolio requires it in Methods (or an equivalent section); BMJ captures it through a structured field in its submission system in addition to the manuscript text. Always check the specific journal’s current author instructions rather than assuming one placement applies everywhere.
Can I list ChatGPT or another AI tool as a co-author?
No. Every major publisher policy — ICMJE, JAMA Network, Nature Portfolio, IEEE, BMJ — explicitly excludes AI tools from authorship, on the basis that authorship requires accountability an AI tool cannot hold. See why ICMJE rejects AI co-authorship for the underlying reasoning.
Is an AI disclosure statement legally required, or just a publisher policy?
For most academic manuscripts, it’s a publisher/journal editorial policy, not a legal requirement. A small but growing number of jurisdictions (the EU AI Act’s transparency provisions, some US state laws) impose separate legal AI-disclosure obligations that apply more broadly than academic publishing — see AI Disclosure Laws vs. Publisher Policy for how the two differ.
How is an AI disclosure statement different from noting AI use in a CRediT statement?
They serve different purposes and are often both required. The disclosure statement documents that a specific tool was used and how; a CRediT statement attributes specific author contributions to specific named authors using CRediT’s fourteen standard roles — AI tools cannot be assigned a CRediT role, but a CRediT statement can still note that AI assisted a named author within a role (for example, within Writing – Original Draft).
