AI Disclosure in Peer Review: What Reviewers and Editors Should Report

Peer review needs its own AI record

A reviewer can use AI without writing a single sentence of the paper.

That sounds obvious, but many disclosure workflows still focus on authors. They ask whether a manuscript contains AI-assisted text, code, images, or analysis. They ask far less often whether a reviewer used AI while judging the paper.

Peer review needs a separate record because the duties differ. Authors disclose how they made the manuscript. Reviewers disclose how they handled someone else's confidential work.

That distinction matters for editors too. If a journal team uses AI to draft decision letters, sort submissions, summarize reviewer reports, or check reporting items, that use can affect editorial judgment.

A simple sentence like "AI was used" will not help much. Reviewers and editors need to report the tool, the task, the inputs, and the role the output played.

If you need the basics first, start with What Are AI Usage Cards?. This article applies the same idea to peer review, where confidentiality sets the limit.

The rule most publisher policies share

Publisher rules differ, but many draw the same line: reviewers should not upload confidential manuscript material into public or non-approved AI tools.

ICMJE tells journals to have a policy for AI use in review and reminds editors and reviewers that submitted manuscripts are privileged communications. It also says uploading a manuscript to AI technologies where confidentiality cannot be assured is prohibited unless authors permit it. (icmje.org)

Elsevier says reviewers should not upload a submitted manuscript, or any part of it, into an AI tool because that may violate author confidentiality, proprietary rights, and data privacy rights. (elsevier.com)

Springer Nature asks peer reviewers not to upload manuscripts into generative AI tools while it explores safe AI tools for reviewers. It also asks reviewers to declare any AI support for evaluation of manuscript claims. (springer.com)

Wiley requires peer reviewers to keep manuscripts and peer review comments confidential and says reviewers must not upload manuscripts, figures, or tables into generative AI tools. (authors.wiley.com) Taylor & Francis gives similar reviewer guidance: reviewers must not enter unpublished manuscript files, images, or information into tools that cannot guarantee confidentiality or may store or reuse the content. (editorresources.taylorandfrancis.com)

You do not need to memorize every publisher policy. You need a working rule: do not paste confidential review material into an AI tool unless the journal policy allows that exact use in that exact setting.

Our [[[[AI Disclosure](/how-to-disclose-microsoft-copilot-use-in-academic-writing/)](/ai-disclosure-in-systematic-reviews-and-meta-analyses/)](/how-to-disclose-ai-use-for-neurips-icml-and-acl-submissions/) Policies by Major Journals](/ai-disclosure-policies-by-journal/) article can help you find publisher-specific rules before you accept a review.

Why peer review raises the stakes

A submitted manuscript is private.

It may contain unpublished data, methods, figures, software, interview excerpts, clinical details, field locations, or ideas that authors have not yet shared. A peer review report may add another layer of sensitive material: doubts about validity, comments on novelty, suspicions about missing citations, or advice to reject.

When you paste that material into an external AI tool, you may disclose more than text. You may disclose the authors' work before publication.

The second risk is judgment. If a reviewer asks an AI tool to decide whether a paper is novel, flawed, or publishable, the tool has entered the reviewer's role. Even if the reviewer agrees with the output, the editor cannot see how much the tool shaped the final report unless the reviewer says so.

That is why AI disclosure in peer review needs more detail than author disclosure. The editor needs to know what you shared and what you decided yourself.

What reviewers should report

A reviewer disclosure should answer six questions.

Which tool did you use? What task did it perform? What did you enter into the tool? What output did you use? Did confidential manuscript material enter the tool? Did the AI output affect your recommendation?

Short is fine. Vague is not.

This disclosure tells an editor almost nothing:

I used AI while preparing this review.

This version gives the editor something to assess:

AI use during peer review:
I used [tool name and version, if known] to improve grammar and organization in my draft review report.
I entered only my own draft comments.
I did not enter manuscript text, figures, tables, supplementary files, author identities, reviewer correspondence, or confidential editorial material.
I checked all suggested edits and made all scientific judgments myself.
The tool did not influence my recommendation.

That wording separates language help from review judgment. It also tells the editor that the reviewer did not upload the manuscript.

For more examples of structured wording, see AI Usage Cards Examples and Templates and How to Disclose ChatGPT Usage in Academic Papers. The author-facing language differs, but the habit is the same: name the tool, name the task, name the boundary.

What reviewers should not outsource

Some uses cross the line for most journals.

Do not upload the manuscript and ask a chatbot to "write a peer review." Do not ask an AI tool to recommend acceptance or rejection. Do not ask it to judge novelty, verify whether the work changes the field, or decide whether the methods support the claims.

Those tasks belong to the reviewer.

A tool may help you polish your own report if the journal allows that use and if you do not share confidential material. A tool should not become the hidden reviewer.

If you already shared confidential manuscript material with an AI tool, do not hide it. Tell the editor what happened.

AI use during peer review:
I uploaded [describe material, such as "two paragraphs from the methods section"] to [tool name] for [task].
This material came from a confidential manuscript under review.
I used the output to [describe use].
I made the final evaluation and recommendation myself.
I am reporting this so the editor can assess journal policy compliance.

That disclosure may reveal a breach. Still, it gives the editor a chance to respond, protect the authors, and decide whether the review can stand.

Editorial checks before publishing AI-focused content

Journal teams that publish AI-focused manuscripts need one extra layer of care.

AI papers often include model outputs, prompts, generated images, synthetic data, code, benchmarks, or claims about tool performance. Reviewers may feel tempted to use AI to test those claims. Editors may feel tempted to run manuscript checks through a chatbot before sending the paper out.

Before publication, the editorial team should check the process, not only the paper.

Ask whether any reviewer or editor used AI during assessment. Ask whether anyone entered confidential manuscript material into a tool. Ask whether the tool ran inside an approved editorial environment or outside it. Ask whether AI output shaped a decision letter, desk rejection, reviewer selection, or final recommendation.

Then check the manuscript itself. Does the author disclose AI use? Does the paper describe the tool or model with enough detail for readers to understand the work? Does the paper distinguish AI-assisted writing from AI used as part of the method? Does the paper report prompts, model versions, settings, data sources, or evaluation limits when those details affect the study?

For journal submissions, the author side of this process appears in AI Transparency Requirements for Journal Submissions and Do I Need to Disclose AI Usage in My Paper?. The reviewer side belongs in the confidential editorial record.

This is where an AI Usage Card helps. It gives the team a consistent format instead of scattered notes in emails, review forms, and decision drafts.

Editors should report their own AI use too

Editors also use AI tools.

They may summarize reviewer comments, draft decision letters, screen for missing declarations, classify article types, or compare a manuscript against reporting checklists. Some of those uses may fit journal policy. Others may require approval.

The same questions apply to editors:

AI use by editor:
I used [tool name] to [task].
I entered [describe inputs].
I did not enter confidential manuscript material into any tool outside the journal-approved workflow.
The AI output helped with [language editing, checklist comparison, summary drafting].
I made the editorial decision myself.

Editors should treat decision letters with care. A decision letter can contain confidential reviewer comments, editorial reasoning, and instructions to authors. If an editor copies that material into a tool that stores or reuses inputs, the editor may create the same confidentiality problem that reviewers face.

The role of the tool matters. Polishing grammar in a decision letter differs from ranking reviewers or predicting acceptance. The closer the tool gets to judgment, the stronger the need for a record.

For wider policy framing, see AI Ethics and Documentation in Academic Research and Why AI Transparency Matters in Research.

A policy reviewers can follow

Many reviewer invitations still say little about AI.

Editors can fix that with plain language in the invitation, reviewer form, and decision workflow. Reviewers should not need to guess.

A good reviewer instruction answers three questions:

AI use in peer review:
Reviewers may use AI tools only for language editing of their own draft review comments, unless the editor gives written permission for another use.
Reviewers must not upload manuscript files, figures, tables, supplements, author information, reviewer correspondence, or editorial communications into external AI tools.
Reviewers must disclose any AI use in the confidential comments to the editor, including tool name, task, inputs, and whether confidential material entered the tool.

That text gives reviewers a boundary before they begin. It also gives editors a standard when something goes wrong.

If your journal already asks authors for AI disclosure, add a matching field for reviewers and editors. Author disclosure alone does not cover the whole editorial process.

How an AI Usage Card fits peer review

An AI Usage Card is a record of tool use. It can include the tool name, purpose, input material, output type, human oversight, and limits.

For peer review, keep one caution in mind: do not make confidential manuscript details public. A reviewer can create a card for the editor or for personal records, but the card should not reveal author names, manuscript title, data, figures, or review content unless the journal permits that sharing.

Use the card as a private process record. Then copy the safe disclosure text into the reviewer form.

You can generate one at the AI Usage Card generator. If you write in LaTeX, the card can sit in a private review note or internal editorial file. Our LaTeX Tutorial for AI Usage Cards and How to Use AI Usage Cards in Overleaf show how to format the text.

A sample peer review AI Usage Card section

A reviewer or editor can adapt this structure.

\section*{AI Usage Card for Peer Review}
 
\textbf{Role:} Peer reviewer
 
\textbf{Tool:} [Tool name, model, and version if known]
 
\textbf{Purpose:} Grammar editing and organization of my draft review comments
 
\textbf{Inputs:} My own draft review text only. I did not upload manuscript text, figures, tables, supplementary files, author information, reviewer correspondence, or editorial communications.
 
\textbf{Outputs:} Suggested sentence edits and report organization
 
\textbf{Human oversight:} I checked every suggestion and accepted only wording changes that preserved my meaning.
 
\textbf{Limits:} The tool did not assess novelty, validity, methods, data interpretation, ethics approval, citation coverage, or publication recommendation.
 
\textbf{Disclosure:} I reported this use in the confidential comments to the editor.

This format works because it records the boundary. It tells the editor that the tool touched wording, not judgment.

For comparison with other documentation formats, see AI Documentation Frameworks Compared and AI Usage Cards vs Model Cards. Model Cards describe AI systems. AI Usage Cards describe how a researcher used an AI system in a specific scholarly task.

What to do when the policy is unclear

If the journal policy does not mention AI in peer review, ask the editor before you use the tool.

Write a short note:

Before I begin the review, I would like to ask whether the journal permits AI assistance for language editing of my own draft reviewer comments.
I will not upload the manuscript, figures, tables, supplements, author information, or confidential editorial material to any AI tool.
If permitted, I will disclose the tool and task in my confidential comments.

If you cannot get an answer, choose the safer route. Do not upload manuscript material. Do not ask AI to evaluate the paper. Write the review yourself.

That may take longer. It also protects you from a policy problem you did not need.

The boundary is the disclosure

Peer review runs on trust, but trust needs records.

Reviewers should disclose AI use when they use it. Editors should ask for that disclosure in a structured way. Journal teams should check AI use in the editorial process before they publish AI-focused work, not after a dispute starts.

The rule stays simple: protect the manuscript, keep judgment human, and write down what the tool did.

Create a private disclosure record now with the AI Usage Card generator, then adapt the text for your reviewer form, editor note, or journal policy.