How to Disclose AI Use in Your Research Paper: 2026 Journal Policies and Templates

#A Solid Paper Got Rejected for the Wrong Reason

A grad student we know had a manuscript bounce back desk-rejected last week. The science was solid. The reviewer notes praised the methodology. The rejection came because the cover letter mentioned ChatGPT had helped polish the discussion section, and the journal's policy required a Methods-section disclosure, not a cover-letter mention. Same information, wrong location, full rejection.

Knowing how to disclose AI use in a research paper is no longer optional knowledge. The disclosure rules are now precise enough to get you rejected on a technicality. AI disclosure sits in the same submission package as ORCID iDs, conflict-of-interest statements, and ethics approvals. This guide covers what every major journal requires in 2026, where to put the statement, how to phrase it, and which AI uses you do not need to disclose at all.

#Why AI Disclosure Matters Now

Three things changed between 2023 and 2026. First, every major publisher (Nature, Springer, Elsevier, Wiley, Taylor and Francis, Cambridge, Oxford) issued formal AI policies. Second, AI tools became deeply embedded in normal writing workflows, which made the line between "use" and "non-use" much fuzzier. Third, the International Committee of Medical Journal Editors (ICMJE) and the Committee on Publication Ethics (COPE) issued cross-publisher guidance that most journals now align to.

The result: editorial offices have a checklist. They look for the disclosure in a specific place, in a specific form. Missing it gets your paper bounced. Including it incorrectly invites questions. A single sentence in the right location is the difference between acceptance and a back-and-forth that delays your paper by weeks.

#The Two Universal Rules Every Journal Agrees On

Across every major journal policy reviewed for this guide, two rules are universal.

AI cannot be an author. Large language models cannot approve a manuscript, take responsibility for its claims, or sign a copyright agreement. ICMJE, COPE, ACM, and every individual journal policy we have read agree on this. Listing ChatGPT, Claude, or Gemini as a co-author triggers an immediate rejection.

AI use must be disclosed when it produces or shapes content. The threshold varies (more on this below) but the principle is consistent. The author remains fully accountable for everything in the paper, including any text the AI generated.

If you remember nothing else from this guide, remember those two rules. The rest is implementation detail.

#Where the AI Disclosure Goes in Your Paper

Every major journal asks for the disclosure in one of two places.

Methods section. Nature, Science, PNAS, JMIR, and most clinical journals require AI disclosures in Methods (or Materials and Methods, depending on field convention). Putting it in Methods signals that the AI use is part of the research process and is therefore subject to the same scrutiny as any other method.

Acknowledgments section. Used when AI was involved in writing-adjacent tasks (language polish, formatting, generating titles for figures) but not in producing scientific content. Some journals accept Acknowledgments-only disclosure; others (Nature, Science) explicitly require Methods even for writing assistance.

Cover letter is not enough. Several journals reject papers that disclose AI use only in the cover letter. The cover letter is read by the editor; reviewers and readers do not see it. Disclosure has to live in the published version of the paper.

The default rule when in doubt: put it in Methods. A Methods disclosure satisfies almost every policy. An Acknowledgments-only disclosure does not.

#What Major Journals Require

Here is the practical state of policies at the journals most researchers submit to in 2026.

Nature Portfolio requires Methods-section disclosure for any LLM use in research, manuscript preparation, or figure generation. Pure copy editing (grammar, spelling) does not require disclosure. AI-generated images are banned outright.

Science requires disclosure of any AI involvement in text or image generation in Methods, with prompts and outputs available on request.

Cell Press journals require disclosure in Methods or Acknowledgments depending on whether the AI shaped scientific content (Methods) or only language (Acknowledgments).

PNAS asks for AI involvement to be acknowledged in Materials and Methods, or in Acknowledgments if no Methods section exists.

JMIR requires disclosure in Methods or Acknowledgments and asks authors to retain full prompts, responses, and transcripts on file. Reviewers can request the full conversation as a Multimedia Appendix.

Elsevier journals require disclosure in Acknowledgments. AI use limited to copy editing does not need to be declared.

Springer Nature journals follow the Nature policy: Methods-section disclosure for substantive use, no disclosure required for pure copy editing.

ICMJE-affiliated medical journals require disclosure in the cover letter AND in the manuscript (Methods or Acknowledgments depending on use).

The pattern: substantive use goes in Methods, language polish goes in Acknowledgments, and pure grammar checking generally does not need disclosure. Always check the specific target journal because policies still vary.

#How to Write an AI Disclosure Statement

A complete disclosure has four elements: which tool was used (with vendor and version), where in the manuscript it was used, what kind of use it was, and a statement that the authors reviewed the output and take responsibility.

A template that satisfies almost every policy:

"During the preparation of this manuscript, the authors used [TOOL NAME, VENDOR, VERSION] to [SPECIFIC USE, e.g., improve language clarity in the Discussion section]. After using this tool, the authors reviewed and edited the content and take full responsibility for the content of the publication."

Concrete examples:

"During the preparation of this manuscript, the authors used ChatGPT (OpenAI, GPT-4) to improve language clarity in the Introduction and Discussion sections. After using this tool, the authors reviewed and edited the content and take full responsibility for the published text."

"We used Claude (Anthropic, Claude 3.5 Sonnet) to summarize 47 candidate papers during the literature review. All summaries were verified against the source papers by the first author before any content was incorporated into the manuscript."

"We used Microsoft Copilot to generate alt text for the figures. The authors reviewed and edited each alt text description before submission."

Notice the pattern: tool name, version, specific use, scope, and responsibility statement. Vague disclosures ("AI was used in the preparation of this manuscript") get flagged for follow-up by reviewers and editors.

#When You Do Not Need to Disclose AI Use

Most journals do not require disclosure for these uses:

• Spelling and grammar checks. Built-in checkers in Word, Google Docs, or Grammarly fall under standard copy editing.
• Reference management. Zotero, Mendeley, and EndNote use AI for metadata extraction, but this is not generative AI use.
• Search engine queries. Searching Google Scholar or Semantic Scholar does not require disclosure.
• AI-assisted code completion. GitHub Copilot for code is treated differently from LLMs for text in most policies, though some journals are starting to ask.
• Translation between languages when the translation is then reviewed by a fluent author.

You should disclose, even if it feels minor, in these cases:

• Generating any text that ends up in the manuscript, even a single paragraph.
• Summarizing papers, even if you then rewrote the summary.
• Generating outlines or section drafts.
• Producing figures, alt text, or captions.
• Brainstorming research questions or hypotheses with AI input.

The "disclose-when-in-doubt" rule from AMEE Guide No. 192 is the safest position. A small over-disclosure does not harm a paper. A missed disclosure that surfaces after publication can become a retraction risk.

#Keeping Records: Prompts, Transcripts, and Audit Trails

Several journals (JMIR, BMJ, some Elsevier titles) now ask authors to retain the full prompts and AI outputs on file in case reviewers want to see them. Some require the full transcript as supplementary material.

Practical record-keeping:

• Save every AI session that produced manuscript content. The web app of most LLMs lets you export conversations as JSON or text.
• Keep a single ai_log.md file in your manuscript folder noting tool, date, section, and a one-line summary of the use.
• For literature work, log which papers were summarized by AI and confirm each summary was verified against the source.

This is the same hygiene you already apply to lab notebooks and analysis code. Treat AI logs as part of the methodological audit trail.

#A Pre-Submission AI Disclosure Checklist

Before you submit:

1. Identify every AI tool used in the project, including writing assistants, literature search tools, and figure generators.
2. Decide which uses require disclosure under your target journal's policy. Use the disclose-when-in-doubt rule for borderline cases.
3. Write the disclosure paragraph and place it in Methods (default) or Acknowledgments (language-only use).
4. If the journal requires it, mention AI use in your cover letter as well.
5. Save full transcripts for any AI session that produced manuscript content.
6. Confirm no AI tool is listed in the author list.

#Where AI Disclosure Fits in the Broader Submission Package

AI disclosure now sits alongside conflict-of-interest, ethics approval, and data availability statements. Most journals have moved it into the standard required-statements block at the end of the paper. Treat it the same way: write it once, polish it for each journal's specific format, and include it in the submission template you reuse across papers.

If you use AI tools systematically across your work, especially for chatting with research papers or AI-assisted literature review, build the disclosure paragraph into your manuscript template now. Every paper gets the same paragraph format, updated for the specific tools and uses on that project. It saves time and makes you less likely to miss the disclosure when you submit under deadline pressure.

The shift from disclosure to mandatory reproduction (full prompt logs, version-locked AI outputs, sandboxed reproductions) is probably 18 to 24 months out. Building the habit of full disclosure and complete record-keeping now makes the next policy shift painless.