AI Disclosure Statements: Per-Publisher Cheat Sheet

My colleague submitted a paper to Lancet last month and was asked to revise it because of non-compliant AI disclosure. We used the Elsevier template that we use for all other Elsevier journals, and placed it where Elsevier says to place it (a dedicated section prior to the reference list). It was picked up by the Lancet sub-editor on technical screen. Lancet needs the disclosure in the acknowledgments, not in a separate section. Lancet also needs disclosure of any use of AI, including copy editing, which is exempt in the Elsevier corporate template. Same publisher group. Different rules. Three weeks lost.

We track AI disclosure requirements across 14 publisher groups for our medical and life-sciences clients. In just the period between October 2024 and June 2026, we counted 38 changes to the policies of the seven publishers covered in this cheat sheet. Six were going in opposite directions, with the medical journals getting tighter while the multidisciplinary publishers were getting looser, and the corporate-level policies under the journal-level policies getting out of step. It's the fastest way to get one's paper bounced on technical screen if one is still using the "one harmonized template" approach one might've used in 2023.

We finish with the five disclosure mistakes that trigger rejection across all seven, and a short workflow we use internally to keep the disclosures right the first time. This post is the cheat sheet. We cover what each of the seven publishers requires as of mid-2026, the exact template language each one accepts, where the statement goes in your manuscript, what's exempt, and the rules on AI-generated images and authorship.

Why one template no longer works

In 2023, when ChatGPT was 12 months old, most journals had not yet written a policy. Authors who disclosed at all used variations of the same paragraph: "We used ChatGPT to improve readability and have reviewed the output." That was enough.

The policies came in 2025. In waves, from different committees, with different timing, with different focus areas. Medical journals were first and fastest, having patient safety and integrity standards to draw upon. Multidisciplinary publishers followed, more slowly and loosely, given their authors span multiple disciplines with differing drafting traditions. Sometimes there wasn't alignment between the publisher's top level corporate policy (on their main about page) and the journal's brand level policy underneath it. By mid-2026, the Lancet (an Elsevier journal) is stricter than Elsevier corporate; Nature (a Springer Nature journal) is more specific than Springer Nature corporate; and BMJ has built a structured disclosure field directly into its submission form that no other major publisher uses.

The net result is that an ai disclosure statement template publisher search returns a dozen results, most of which are out of date by months. The version below is current as of June 2026 and we update our own internal table monthly. If one finds a discrepancy with one's target journal's author guidance, the journal-level guidance wins.

The master comparison table

Read across the row for the publisher one is submitting to. Read down each column to see where the divergences hide.

The row that surprises authors most is the Lancet. It's an Elsevier-owned title that doesn't follow Elsevier's exemptions. The row that creates the most technical-screen rejections is BMJ, because authors who have written the disclosure correctly in the manuscript often miss the structured submission-form field and get flagged for inconsistency between the two.

Elsevier: the dedicated section, the exact template

Elsevier publishes the most prescriptive template of any major publisher, and it is the only one that mandates a specific section title. Use it verbatim.

Declaration of generative AI and AI-assisted technologies
in the manuscript preparation process

During the preparation of this work, the author(s) used
[NAME OF TOOL / SERVICE] in order to [REASON]. 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 published article.

Two operational notes. First, this section goes right away before the references list, not in the acknowledgments. The section is published in the final article, so it will be visible to every reader, not just to peer reviewers. Authors sometimes write disclosure language that they wouldn't want to appear in print. The Elsevier format does not give you that hiding place.

Basic grammar, spelling, and punctuation checking aren't included in Elsevier's disclosure. You do not need to disclose if you used Grammarly's free level or Microsoft Editor. But if you used Grammarly's GrammarlyGO (generative rewrites), Wordvice, Paperpal, or any other tools that restructure or paraphrase, then you'll need to disclose. The exemption is narrower than authors often assume.

Elsevier's policy was last updated June 2026. The biggest practical change in the most recent revision: AI-helped code writing or editing was added explicitly to the disclosure list, which had been ambiguous in earlier versions.

Springer Nature and Nature: the corporate vs brand split

Springer Nature publishes a corporate-level AI policy that covers BMC, Nature Portfolio, Springer, Palgrave, and Cureus. The corporate policy is short and absolute: AI cannot be an author, AI-generated images aren't allowed in any Springer Nature publication, and peer reviewers are asked not to upload manuscripts into generative AI tools. But the details about how this will work in practice (i.e., disclosure mechanics) are left up to each brand.

For Nature itself (the flagship journal and most Nature Portfolio titles), AI use must be disclosed in the methods section of the manuscript. Nature doesn't need a specific section title; the disclosure is built into the methods narrative. Copy editing for grammar and language is exempt from disclosure, which makes Nature's exemption narrower than Elsevier's: paraphrasing tools, summarizers, and any AI that touches structure or argument must be disclosed even when only used for "readability."

A workable Nature-compatible template:

Methods (excerpt):
We used [TOOL NAME, VERSION] to [SPECIFIC TASK, e.g., generate
draft summaries of source documents in the literature review].
All AI-generated text was reviewed, fact-checked, and substantively
revised by the authors, who take full responsibility for the
content. The model was not used for data analysis, figure
generation, or original scientific argumentation.

For BMC titles, Palgrave titles, and Cureus, check the journal's individual author guidelines: the templates and section locations vary. Springer (the imprint, distinct from Nature Portfolio) typically tracks Nature's pattern but accepts a separate "Disclosure" subsection within the methods rather than fully integrated text.

Where Springer Nature differs from everyone else is about the image rule. We don't accept AI generated images, illustrations, or figures in any of our titles. And this applies even if we understand it as "AI assisted" image improvement. If you used Midjourney, DALL-E, Stable Diffusion, or any generative image tool to create any kind of visual element in your manuscript, you need a different publisher.

The medical bloc: ICMJE, Lancet, BMJ, NEJM

The four medical references in this cheat sheet share a common framework (ICMJE recommendations) and then layer additional requirements on top. Understanding the ICMJE baseline first makes the journal-specific divergences easier to track.

ICMJE recommendations (2026). At submission, authors should report the use of AI. If AI was involved in writing, that information is placed in the acknowledgments section. Use of AI to collect data, analyze data, or generate figures is reported in the methods section. The cover letter should explain how the AI was used. AI can't be included as an author because it cannot take responsibility for accuracy, integrity, and originality. This is a minimum requirement of all ICMJE-compliant journals.

A workable ICMJE acknowledgments template:

Acknowledgments (excerpt):
During the preparation of this manuscript, the authors used
[TOOL NAME, VERSION] for [SPECIFIC TASK, e.g., language editing
of the discussion section]. The authors reviewed and edited the
output and take full responsibility for the content of the
publication. No patient data was input into any hosted AI tool.

The "no patient data" sentence is not in the ICMJE baseline but is increasingly expected by clinical-medicine journals; including it costs nothing and pre-empts a frequent reviewer question.

The Lancet. Stricter than its Elsevier parent. AI is allowed only for readability and language improvements; using a big language model to generate scientific arguments, draft method, write literature reviews, or produce any new content is prohibited. The disclosure goes in the acknowledgments. Every AI use, including copy editing, must be disclosed. The Lancet does not exempt grammar tools the way Elsevier corporate does. AI-generated images are prohibited.

BMJ. BMJ goes further than any other major publisher on disclosure mechanics. Authors must disclose in three places: in the methods section, in the cover letter, and in a structured field on the online submission form. The submission form creates a data point the editorial team can cross-reference against the manuscript text. The most common technical-screen rejection at BMJ is a mismatch: a disclosure in the manuscript that does not appear in the form, or vice versa. Name the tool (e.g., "ChatGPT-4o"), the version, and the exact use case. All authors must confirm they have reviewed any AI-helped text. AI-generated images are prohibited.

NEJM. ICMJE-compliant, with the addition that disclosure must appear in both the cover letter and the manuscript (the cover letter disclosure alerts editors before review; the manuscript disclosure ensures the published record is transparent). NEJM doesn't ban AI tools and does not restrict them to language editing only, which puts it slightly more permissive than the Lancet. Authors bear full responsibility; the accountability chain runs through the human authors, not the tool.

A BMJ-and-NEJM-compatible cover letter sentence:

Cover letter (excerpt):
We confirm that [TOOL NAME, VERSION] was used for [SPECIFIC TASK]
during the preparation of this manuscript. This use is disclosed
in the [methods section / acknowledgments] of the submitted
manuscript. No patient-identifying data was provided to the AI
tool. The authors take full responsibility for all content,
including any AI-assisted text.

If you're submitting to a clinical medicine journal not named here, default to the ICMJE baseline and add the cover-letter disclosure. You'll rarely be over-disclosing in this part of the field.

Wiley: the underspecified outlier

The least prescriptive is that of Wiley, which has published no mandatory sections, titles, or templates. Its corporate policy simply requests that authors state "all AI Technology used, including its purpose, whether it influenced key arguments or conclusions, and how they personally reviewed and verified any AI-generated content." It provides examples but does not need them.

Wiley exempts AI tools used solely for spelling, grammar, and general editing. The exemption is the broadest in the field; arguably broader than is wise, given that some Wiley journals have stricter individual guidelines than the corporate policy implies. As always, the journal-level guidance wins.

A Wiley-compatible declaration that works across most Wiley titles:

Generative AI Use Statement:
The authors used [TOOL NAME, VERSION] during the preparation of
this manuscript for [SPECIFIC TASK]. The use of this tool did
not influence the key arguments or conclusions of the work.
The authors reviewed and verified all AI-assisted content and
take full responsibility for the manuscript.

Put this statement in one's acknowledgments unless one's specific Wiley journal requires something different. The "did not influence key arguments or conclusions" part is important: Wiley's policy directly asks whether the AI influenced the major content, so an evasive disclosure would omit this point. At Wiley, one can use AI-generated images with disclosure. They're one of two publishers in this cheat sheet (alongside Elsevier) that don't absolutely forbid their use.

Last update: May 2026 Wiley's policy Last update: May 2026 This is the biggest change I've seen recently. It used to be ambiguous whether AI-helped translation would need to be disclosed. But now it's explicit that it should be.

The 5 disclosure mistakes that trigger rejection across publishers

In my editorial dataset, there're five mistakes that lead to about 70 percent of the AI disclosure-related technical screen rejections we see. And they're always these five mistakes.

1. Using a template from the wrong publisher. Note the Elsevier template in the Lancet submission at the top of this article. Note the Nature template in the BMJ submission that misses the cover letter and the form field. Start with what the target journal wants right now, then fill in with a cheat sheet like this one.

2. Disclosing in the wrong location. Putting an AI disclosure in a footnote when the publisher specifically says that the publisher wants it in a section, or putting it in the acknowledgments when the publisher wants it in the methods, that's also considered non-compliant, even if the information is technically correct.

3. Inconsistency between the manuscript, the cover letter, and the submission form. At BMJ, which has a structure for this in their forms; at NEJM and Lancet, which need both the cover letter and the manuscript. Inconsistencies will be noticed by editors and reported as an issue of research integrity.

4. Vague language. "AI tools were used during preparation" without specifying the tool, version, scope, or responsibility statement. Vague is evasive; specific is professional.

5. Listing AI as a co-author. Still happens, mostly in low-tier venues. All publishers in this cheat sheet ban it. All publishers in this cheat sheet treat it as instant desk rejection.

Our general AI disclosure guide covers the important what-to-disclose question in more detail. Our journal cover letter guide walks through the cover-letter consistency check that prevents mistake 3.

A workflow we use internally

This is a mechanical process, not an intellectual one. Almost no rejections happen if one gets into the habit of doing it shortly.

Step 1. Keep a running log as you draft: tool name, version, what you used it for, which sections of the manuscript. A plain text file works. Update it every time one uses AI.

Step 2. Before submission, identify one's target journal's current author guidance (not the publisher's corporate policy). Note the section, the cover-letter requirement (if any), and the form-field requirement (if any).

Step 3. Adapt the appropriate template from this cheat sheet, paste in the tool details from your log, and place it in the section the journal specifies.

Step 4. Cross check the disclosure in the manuscript against the disclosure in the cover letter and the disclosure in the submission form. Make sure they all have the same tool name, version, and scope.

Step 5. Get a senior co-author or your PI to read the disclosure before submission. They've usually seen what reviewers flag.

Our proofreader for academic writing keeps the AI-tool log automatically as you edit, including tool name and version, so the disclosure write-up is a copy-paste at the end. If you're also editing or rewriting around AI-touched passages, our note on paraphrasing that preserves citations covers the techniques that keep your reference list intact across the rewrites.

Disclosure-Ready Proofreader for Journal Submissions

Automatic edit-history log, tool-and-version capture, and one-click disclosure language for Elsevier, Springer Nature, Wiley, and ICMJE journals.

Frequently asked questions

Q: Do I have to disclose AI use if I only used it for grammar?

It depends on the publisher. Elsevier and Wiley exempt grammar and spelling tools from disclosure. Nature exempts copy editing for grammar and language. The Lancet doesn't exempt anything: every AI use, including copy editing, must be disclosed. BMJ in practice expects disclosure of anything beyond traditional spell-check. When in doubt, disclose; the cost of over-disclosure is zero, the cost of under-disclosure is potentially a desk rejection.

Q: Can I use the same disclosure statement across multiple publishers?

No, not safely. Section locations are different, mandated wording is different (Elsevier), mandatory cover letter (BMJ and NEJM, the medical bloc only) and image policies are very different. You can have one master list of your AI use and adapt the disclosure to each publisher's requirements, but if one copy paste a statement into all of them, one will hit technical screen at some.

Q: What happens if I forget to disclose?

Technical screen: Paper is sent back to authors for resubmission/revision. Peer Review: Editor identifies and requests formal correction or, in worst case scenarios, withdraws paper from review. Post-Publication: Papers using undisclosed AI may result in a correction notice, expression of concern from the editors or in the worst case, retractions. Reputational damage: Databases tracking journal integrity flags based on COPE standards.

Q: Can I list ChatGPT or Claude as a co-author?

No, at any publisher in this cheat sheet and at all major publishers worldwide. The classic statement comes from the ICMJE. Authorship suggests responsibility; therefore, AI can't be an author. Putting AI as a co-author is an immediate desk reject and a research integrity flag.

Q: I used AI to generate a figure. Which publishers will accept this?

Only two publishers in this cheat sheet allow AI-generated images, Elsevier and Wiley. All others, including Springer Nature, Nature, the Lancet, BMJ, and NEJM don't let AI-generated images. If one have a paper in the pipeline with a figure generated by Midjourney, DALL-E, Stable Diffusion, or similar, one've three choices: 1. Remove the figure 2. Replace with a non-AI figure 3.