Skip to main content
v2026.11,986 entries · CC-BY 4.0

AI Disclosure Statement Examples: Ready-to-Use Wording for Journals, Funders and Institutions

A publisher-by-publisher comparison table (ICMJE, JAMA, Nature Portfolio, Elsevier, PLOS, Wiley, Taylor & Francis, Science, BMJ, IEEE, NIH, NSF) plus ready-to-use AI disclosure statement wording for every task, the citing-vs-disclosing distinction, and the reviewer-side confidentiality rules most authors miss.

Written and maintained by CASRAI Editorial Board

Published · Last updated

On this page

An AI disclosure statement is a short passage in your manuscript or submission that names the generative-AI tool you used, says what it did, and confirms that named human authors checked the output and are responsible for it. ICMJE asks journals to require authors to disclose AI-assisted technologies at submission; the exact wording and location depend on the publisher. Copy-ready wording follows.

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.

Ask CASRAI · free to tryQuestion about this guide? Ask CASRAI answers with linked sources. 2 free questions a day, no account.

Ask your first question freeSubscribe

Publisher and funder AI disclosure requirements at a glance

This table summarizes what each policy requires, taken from each publisher’s or funder’s own author and reviewer guidance (sources and read dates are listed at the end of the page). A cell marked “Not covered here” means this page does not report a separate position for that organization; it does not mean none exists. Confirm against your target venue’s current instructions before submitting. A second, narrower table further down (the data table after this article) compares six bodies on exemptions, placement and review rules.

Publisher / funder Author disclosure required? Where it goes AI listed as author? AI use by reviewers/editors
ICMJE (baseline recommendations) Yes: journals should require authors to disclose at submission whether AI-assisted technologies were used Cover letter and the submitted work, in the appropriate section if applicable; most ICMJE-aligned journals use Acknowledgments or a dedicated AI-use declaration No Not covered here; check the venue’s reviewer guidance
JAMA Network / AMA Manual of Style Yes, substantive use (routine spelling/grammar checking exempt) Acknowledgment or Methods No Not covered here; check the venue’s reviewer guidance
Nature Portfolio (Springer Nature) Yes, for text generation; images/figures largely restricted regardless of disclosure Methods (or nearest equivalent section) No Not covered here; check the venue’s reviewer guidance
Elsevier Yes, substantive generative-AI use in the writing process; basic grammar/spelling checks exempt Standalone “Declaration of Generative AI and AI-assisted technologies in the writing process,” end of manuscript, before references No, explicitly Reviewers may not upload manuscript content into generative AI tools; narrow disclosed assistance (e.g. a background literature search via a private, non-data-retaining tool) may be permitted with disclosure to the editor
PLOS Yes: name each tool, how it was used, which aspects it affected, and how the output was validated In the article body: a dedicated Methods section, Acknowledgements, or relevant figure legends No AI tools cannot serve as peer reviewers or decision-issuing editors; reviewers and editors must not upload submissions or unpublished content to generative-AI tools; any permitted use is disclosed in the review form or decision letter
Wiley Yes, for substantial use: describe it transparently, state its purpose and how output was verified; tools used solely for spelling, grammar and general editing are excluded Integrated into the manuscript; the journal’s own author instructions govern the section No Reviewers and editors may not upload manuscripts into generative-AI tools; reviewers may use AI only to improve the clarity of their own written feedback and must disclose that to the handling editor
Taylor & Francis Yes; coding assistance and LLM-enhanced literature search are explicitly named as permitted-with-disclosure uses Methods or Acknowledgments for articles; for books, disclose to your editorial contact at proposal stage, then state in the preface/introduction if approved No “Editors and peer reviewers must not upload files, images, or information from unpublished manuscripts into Generative AI tools” (Taylor & Francis’s own wording)
Science / AAAS Yes Both the cover letter and the Methods/Acknowledgments section; full prompt, tool name, and version disclosed in Methods No — and an AI tool cannot be cited as a source either Not covered here; check the venue’s reviewer guidance
BMJ Yes Structured AI-use field in BMJ’s ScholarOne submission system, plus the manuscript text No Not covered here; check the venue’s reviewer guidance
IEEE Yes, general acknowledgment-style disclosure Varies by journal/conference — no single centrally branded IEEE-wide document; check the specific venue’s author instructions No Not covered here; check the venue’s reviewer guidance
NIH (grant applications & peer review) No disclosure-statement template; for applications submitted from the 25 September 2025 receipt date, NIH will not consider applications substantially developed by AI to be original ideas of applicants (NOT-OD-25-132), and NIH’s misconduct focus covers AI-generated data presented as empirical, undisclosed AI image alteration and nonexistent AI-generated citations N/A — not a manuscript-authorship context N/A Reviewers barred from uploading application, proposal, or critique content into online generative AI tools (NOT-OD-23-149); reviewers sign a modified confidentiality agreement naming this prohibition
NSF (grant proposals & merit review) Encouraged, not required: proposers are encouraged to indicate in the project description the extent to which, if any, generative AI was used, and remain responsible for the accuracy and authenticity of the proposal (NSF notice of 14 December 2023) The project description N/A Reviewers are prohibited from uploading proposal content, review information and related records to non-approved generative-AI tools; a violation breaches the confidentiality pledge

The pattern across every row: nobody permits AI authorship, everybody requires disclosure of substantive use somewhere in the submission, and wherever a reviewer-side policy has been published, it draws a harder confidentiality line than the author-side disclosure rule — disclosure is not enough on the reviewer side, because the concern is protecting unpublished content from a third-party AI provider, not documenting a legitimate use.

The six 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 six of these:

  1. The tool’s name and developer — “ChatGPT (OpenAI),” not just “AI” or “a chatbot.”
  2. 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.
  3. 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.”
  4. 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.
  5. How the output was checked — PLOS asks for “a description of how the user evaluated the validity of the tool’s outputs”, and Wiley asks authors to document the purpose and how output was verified. A sentence such as “Outputs were checked against [source or dataset] by [initials]” satisfies both.
  6. 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.

Citing AI vs. disclosing AI use: two different obligations

These get conflated constantly, and they are not interchangeable. Disclosing AI use is a statement, usually in Methods or Acknowledgments, that names the tool and describes what it did — it satisfies the publisher’s transparency requirement. Citing a source is an attribution inside your reference list or in-text citation apparatus, pointing a reader to a retrievable, checkable work. A generative AI tool’s output is neither: it is not a stable, retrievable source (the same prompt run twice can return different text), and under every policy compared above it cannot hold the accountability that authorship or citation implies. Science/AAAS makes this second point explicit — sources used in a Science paper may not themselves be AI-authored, meaning you cannot cite an AI chatbot’s output the way you would cite a journal article, even informally.

The practical rule: disclose AI use in the section your target journal specifies (see the table above); do not add an AI tool to your reference list or in-text citations as though it were a source. If you need to reference a specific AI-generated output for reproducibility (for example, showing exactly what a model produced), some venues accept this as a footnote, a supplementary file, or an explicitly labeled non-bibliographic reference — check your target journal’s instructions rather than treating an AI chatbot transcript as citable in the normal sense.

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.”

Elsevier

Elsevier’s policy requires a standalone declaration, placed at the end of the manuscript immediately before the reference list, and published as part of the final work:

“During the preparation of this work, the author(s) used [tool name] in order to [specific task]. 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.”

Elsevier’s own wording is explicit that this declaration should not include basic grammar, spelling, and punctuation checking — reserve it for substantive drafting, editing, or content-generation use.

PLOS

PLOS requires disclosure in a dedicated section of the Methods, or in Acknowledgements for article types that don’t include a Methods section:

“[Tool name] was used to [specific task, e.g. assist with copyediting the manuscript for clarity]. The authors reviewed and edited the output and take full responsibility for the content of this publication.”

Taylor & Francis

Taylor & Francis names coding assistance and LLM-enhanced literature searching explicitly as permitted, disclosed uses, distinct from prohibiting AI-generated research content outright:

“[Tool name] was used to [specific task, e.g. assist with an initial literature search using an LLM-enhanced search interface / generate exploratory code for data cleaning]. All outputs were independently verified by the author(s), who take full responsibility for the accuracy of the final content.”

For Taylor & Francis book projects specifically, disclose intent to use generative AI to your editorial contact at the proposal or early-writing stage rather than waiting until submission — this is a distinct workflow from journal-article disclosure.

Science / AAAS

Science requires disclosure in both the cover letter and the Methods or Acknowledgments section, with the full prompt, tool name, and version stated in Methods:

“[Tool name, version] was used with the prompt ‘[prompt text]’ to [specific task]. The output was reviewed and verified by the authors, who take full responsibility for its accuracy and take full responsibility for the content of the publication.”

Because Science requires the cover letter to also flag AI use, don’t treat the in-manuscript disclosure alone as sufficient for this venue — both are required.

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.”

No AI used

A growing number of venues (BMJ’s structured submission field and Science/AAAS among them) ask authors to make an affirmative statement either way, not just when AI was used. Where a venue’s submission system or author instructions call for this:

“The authors did not use generative AI tools or AI-assisted technologies in the preparation of this manuscript, its text, figures, or underlying data.”

Only use this if it’s literally true for the entire manuscript, including revisions — if a co-author used an AI tool for a section you didn’t write yourself, that falls back to the appropriate task-specific template above, not this one.

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.

AI use by reviewers and editors: a stricter, separate set of rules

Everything above governs what an author discloses about their own AI use. A materially different, more restrictive set of rules governs what a peer reviewer, editor, or grant reviewer may do with a manuscript or application they’ve been asked to evaluate — and this is the part authors and reviewers alike most often miss, because it isn’t phrased as a “disclosure statement” question at all. It’s a confidentiality question: uploading someone else’s unpublished manuscript, proposal, figures, or data into a third-party generative AI tool can expose that unpublished content to the tool provider, regardless of whether the reviewer discloses having done so.

Elsevier, Springer Nature, Wiley, and Taylor & Francis all prohibit reviewers from uploading manuscript content into generative AI tools for this reason, and NIH applies the identical logic to grant peer review under notice NOT-OD-23-149, requiring reviewers to sign a modified confidentiality agreement that names the prohibition explicitly. A narrow exception some of these publishers carve out is polishing the wording of a review the reviewer has already written themselves — not uploading the manuscript being reviewed — typically with disclosure to the editor. NSF prohibits merit reviewers from uploading proposal content, review information and related records to non-approved generative-AI tools; reviewers may share publicly available information with current tools.

Publisher-specific reviewer rules

  • PLOS: AI tools cannot serve as peer reviewers or decision-issuing editors, and AI should not be used to summarize or evaluate submissions. Reviewers and editors must not upload submissions, summaries or comments that reveal unpublished content to generative-AI tools. AI is permitted only where it does not risk confidentiality or misrepresent the reviewer’s contribution, for example copyediting one’s own comments that contain no manuscript summaries, and the use is disclosed in the review form or decision letter.
  • Wiley: reviewers may use AI only to improve the clarity of their own written feedback and must disclose that to the handling editor when submitting the review. Editors may disregard reviews that show inappropriate AI use, and responsibility for the review “must not be delegated”.

If you review manuscripts or grant applications, treat “may I disclose AI use as a reviewer the way I would as an author” as the wrong question — the real question is whether the material may go into the tool at all, and for unpublished third-party work under confidential review, the answer across the policies compared on this page is no. See CASRAI’s full AI in Peer Review: Publisher Policy Comparison for the complete reviewer-side breakdown, and the AI in peer review dictionary entry for the operational definition.

Grant applications and institutional submissions

Funder policy on AI disclosure is considerably less standardized than journal policy, and — as of November 2026, neither NIH nor NSF 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. The NSF and NIH positions differ and are set out in the section on NSF and NIH below.

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.

Which uses need no disclosure? Exemptions compared

Grammar and spelling checking is the most common edge case, and publishers answer it differently:

  • Elsevier: basic checks of grammar, spelling and punctuation need no declaration.
  • Wiley: tools used solely for spelling, grammar and general editing are excluded; substantial use must be described transparently and integrated into the manuscript, and a missing description may be requested later.
  • PLOS: disclosure is required for any use that generated or revised submission content or supported the research; the policy text does not set a grammar-check exemption.
  • ICMJE: journals should require authors to disclose at submission whether AI-assisted technologies were used, and to describe the use in both the cover letter and the submitted work.

Where policies disagree, follow the stricter reading for the venue you are submitting to. The data table below the article sets these positions out alongside NSF and NIH.

Funders: NSF and NIH are not the same

Neither funder prescribes a disclosure template, but their rules differ in kind.

NSF: disclosure encouraged, reviewer uploads restricted

NSF’s notice on generative AI in the merit review process (dated 14 December 2023; the page was last updated 21 May 2026) says proposers are “encouraged” to indicate in the project description the extent to which, if any, generative AI was used. It is not a requirement, and proposers remain responsible for the accuracy and authenticity of the proposal. Reviewers are prohibited from uploading proposal content, review information and related records to non-approved generative-AI tools; doing so breaches the confidentiality pledge, though reviewers may share publicly available information with current tools. A sentence a proposer can use: “Generative AI ([tool, developer]) was used to [task] in preparing this project description; the PI reviewed all output and is responsible for its accuracy.”

NIH: an originality rule and an application cap

NIH notice NOT-OD-25-132, released 17 July 2025, applies to applications submitted for the 25 September 2025 receipt date and later. NIH will not consider applications that are substantially developed by AI to be original ideas of applicants. NIH will accept only six new, renewal, resubmission or revision applications per principal investigator per calendar year (training and R13 applications are excluded), will use detection technology, and may refer post-award findings to the Office of Research Integrity. The notice does not set a disclosure template, so the practical safeguard is to keep AI use limited to editing and to be able to show the applicant’s own authorship of the ideas.

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. Competing interests are declared separately; see conflict of interest disclosure statement templates for manuscripts.

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).

Does NSF require an AI disclosure statement in grant proposals?

No. NSF’s notice on generative AI in the merit review process, dated 14 December 2023, says proposers are “encouraged” to indicate in the project description the extent to which, if any, generative AI was used. Proposers remain responsible for the accuracy and authenticity of their proposals, whether or not they disclose.

Can a peer reviewer upload a manuscript or grant proposal to ChatGPT to help write their review?

No, under the policies compared on this page. Elsevier, Springer Nature, Wiley, Taylor & Francis, PLOS, NIH, and NSF all restrict uploading unpublished manuscript or proposal content into a generative AI tool (NSF specifies non-approved tools), because doing so can expose confidential, unpublished material to the tool’s provider. This is a stricter rule than author-side disclosure, not a milder version of it — see AI in Peer Review: Publisher Policy Comparison for the full breakdown.

Where does the Elsevier AI statement go?

At the end of the manuscript, immediately above the references, under the heading “Declaration of Generative AI and AI-assisted technologies in the writing process”. Elsevier’s template reads: “During the preparation of this work the author(s) used [NAME TOOL / SERVICE] in order to [REASON].” Elsevier does not permit generative AI to create or alter images, and AI should not be listed or cited as an author.

Do I put the disclosure in the cover letter as well?

ICMJE says to describe AI use in both the cover letter and the submitted work, in the appropriate section if applicable. Science/AAAS also asks for both, per the table above. Check whether your journal’s submission system has a separate field.

Does NIH require disclosure of AI in grant applications?

NIH publishes no disclosure template. Under NOT-OD-25-132, applications substantially developed by AI are not considered the applicant’s original ideas, and each PI is limited to six applications per calendar year.

Can reviewers use AI to polish their review?

Only narrowly. Wiley allows AI to improve the clarity of the reviewer’s own written feedback, with disclosure to the handling editor. PLOS allows it only for the reviewer’s own comments without manuscript summaries, disclosed in the review form or decision letter. Uploading the manuscript itself is prohibited by both.

Sources (accessed 12 November 2026): ICMJE, Elsevier, PLOS, Wiley, NSF and NIH pages listed below. Other publisher rows were last checked 16 August 2026. Publisher and funder AI policy is revised frequently; re-check your target venue’s current instructions before submission.

Sources

Related reading: Taylor & Francis Generative AI Policy — What Taylor & Francis’s generative AI policy actually requires: no AI tool as author, a named-tool-and-version disclosure statement in Methods or Acknowledgments, and a separate disclose-and-approve process for book authors.

Related reading: Journal Checker Tool — How the Journal Checker Tool actually works: the three inputs authors enter, what Compliant/Non-Compliant/Unknown mean, the three Plan S compliance routes, a step-by-step walkthrough, and how it differs from institutional platforms like ChronosHub.

Related reading: disclosing AI Assistance in a Thesis — Many graduate schools now require a dedicated AI-use disclosure statement in a thesis or dissertation, distinct from journal AI-disclosure policy.

Data: Generative AI disclosure and reviewer rules: ICMJE, Elsevier, PLOS, Wiley, NSF and NIH

Side-by-side comparison of what six bodies say about authors' or applicants' disclosure of generative AI use, where the disclosure goes, whether basic grammar tools are exempt, and what they say about AI in review. Compiled by CASRAI from each body's published policy page (accessed 2026-11-12); 'Not addressed in the policy' means the policy does not address the point, not that no other rule exists.

Period covered: 2023/2026.

Generative AI disclosure and reviewer rules: ICMJE, Elsevier, PLOS, Wiley, NSF and NIH
BodyAuthor or applicant disclosureWhere the disclosure goesGrammar and spelling tools exemptAI listed as authorAI in review or merit reviewPolicy date or version noted on page
ICMJEJournals should require authors to disclose at submission whether AI-assisted technologies were used and describe howCover letter and the submitted work, in the appropriate section if applicableNot addressed in the policyNo: chatbots should not be listed as authorsNot addressed in the policyNot addressed in the policy
ElsevierDisclose use of AI tools for manuscript preparation in a separate AI declaration statementEnd of manuscript, immediately above the references, under the heading 'Declaration of Generative AI and AI-assisted technologies in the writing process'Yes: basic checks of grammar, spelling and punctuation need no declarationNo: AI tools must not be listed or cited as author or co-authorNot addressed in the policyNot addressed in the policy
PLOSDisclose any AI use that generated or revised submission content or supported the research: tools, how used, aspects affected, how output validity was evaluatedIn the article body: a dedicated Methods section, Acknowledgements, or relevant figure legendsNot addressed in the policyNo: AI tools cannot be listed as authorsAI cannot serve as peer reviewer or decision-issuing editor; reviewers must not upload submissions or unpublished content; any permitted use is disclosed in the review form or decision letterNot addressed in the policy
WileyDisclose use of AI technologies at submission; describe substantial use transparently, with purpose and how output was verifiedIntegrated into the manuscriptYes: tools used solely for spelling, grammar and general editing are excludedNo: AI cannot be listed as an authorReviewers and editors may not upload manuscripts; reviewers may use AI only to improve clarity of their own feedback and must disclose to the handling editorNot addressed in the policy
NSFEncouraged, not required: indicate in the project description the extent to which, if any, generative AI was used; proposers remain responsible for accuracyProject descriptionNot addressed in the policyNot applicable (grant proposals)Reviewers prohibited from uploading proposal content, review information and related records to non-approved generative AI toolsNotice published 2023-12-14; page last updated 2026-05-21
NIHNo disclosure template in NOT-OD-25-132; applications substantially developed by AI are not considered the applicant's original ideas; six applications per PI per calendar yearNot applicableNot addressed in the policyNot applicable (grant applications)Not addressed in NOT-OD-25-132NOT-OD-25-132 released 2025-07-17; effective for the 2025-09-25 receipt date and later

Sources

Compiled by CASRAI from the sources above. Free to reuse under CC BY 4.0 with attribution.

Follow CASRAI

Research-administration guidance, standards updates and independent tool reviews.

Ask CASRAI · free to try

Ask about AI Disclosure Statement Examples: Ready-to-Use Wording for Journals, Funders and Institutions

2 free questions a day, no account.

An AI assistant for research: funder rules, compliance, methods and publishing. It links the sources behind its answers and labels answers that come from a web search.

Answers draw on CASRAI's guides and dictionary plus the federal and funder documents we index: Federal Register, Grants.gov, Regulations.gov and UKRI.

Works on this site and inside Claude, Cursor and the AI tools you already use.

Everything CASRAI publishes — this page, the dictionary, the guides and the news — stays open to read, with no account and no card.

Referenced across the research world

University of Cambridge logoColumbia University logoCrossref logoUniversity of Edinburgh logoHarvard University logoUniversity of Oxford logoPrinceton University logoStanford School of Medicine logoUniversity College London logoORCID logoUniversity of Cambridge logoColumbia University logoCrossref logoUniversity of Edinburgh logoHarvard University logoUniversity of Oxford logoPrinceton University logoStanford School of Medicine logoUniversity College London logoORCID logo
  • University of Cambridge logo
  • Columbia University logo
  • Crossref logo
  • University of Edinburgh logo
  • Harvard University logo
  • University of Oxford logo
  • Princeton University logo
  • Stanford School of Medicine logo
  • University College London logo
  • ORCID logo

View CASRAI adoption →