AI in Peer Review: Ethical Guidelines for Journal Editors
Reviewed by the NeucitePress Editorial Board — PhD academics, peer-reviewed journal editors and subject-matter specialists.
Last updated: October 2026 • Reviewed by the NeucitePress Editorial Board • Reading time: 7 minutes
Reviewers and editors now have easy access to generative AI tools, and many are already using them informally. Journals need a written position before a confidentiality breach forces one. This guide sets out what the main publication-ethics bodies, a major funder and a large publisher actually say about AI in peer review, then turns it into a short policy checklist for editors.
We have not included adoption percentages, accuracy rates for named tools or bias magnitudes. Published figures of that kind vary by tool, discipline and test set, and we could not trace the ones previously quoted here to a primary source.
What the main guidance says
- ICMJE. The Recommendations say reviewers should not upload a manuscript to software or other AI technologies where confidentiality cannot be assured, and that reviewers should ask the journal for permission before using AI to help with a review. Journals, for their part, should tell reviewers what is allowed and be open about any AI use in their own processes.
- COPE. Its position statement says AI tools cannot be listed as authors, because authorship carries responsibility that a tool cannot take. Authors must be transparent about AI use, and humans stay accountable for the work.
- NIH. For grant peer review, NIH prohibits reviewers from using generative AI to analyse or formulate critiques of applications, because uploading confidential material to such tools can breach confidentiality.
- Springer Nature. Its editorial policies say reviewers should not upload manuscripts into generative AI tools, on confidentiality and proprietary-rights grounds.
The common thread is simple. A manuscript under review is confidential, humans make and own the judgement, and any AI involvement must be permitted by the journal and disclosed.
A decision flow for editors
What AI can reasonably help with, and what it cannot
Many editorial-office tasks use software without exposing a manuscript to an external service: reference checking against Crossref or PubMed, similarity reports from a licensed service, and checks that a submission is complete. These are screening aids that give evidence to an editor.
Judging novelty, significance, methodological soundness and fit is a different kind of task. It requires expertise and accountability, and the guidance above keeps it with people. Treat any claim that a tool can “review” a paper with suspicion, and do not let an AI score stand in for an editorial decision.
Bias and fairness
Tools trained on historical text can encode its biases. Published research has shown, for example, that AI-text detectors can misclassify writing by non-native English speakers as machine-generated (Liang et al., Patterns, 2023). If your journal uses any automated flag, treat it as a prompt for a human to look, never as grounds for rejection, and give authors a way to query the outcome.
A policy checklist for your journal
- Reviewers. State whether reviewers may use AI at all. If they may, require them to ask permission first and to disclose, and forbid uploading manuscripts to tools that do not guarantee confidentiality.
- Editors and staff. Apply the same confidentiality rule to your own office. Keep a list of the tools in use and what each one is allowed to touch.
- Vendors. Before you sign, ask in writing whether submitted content is stored, used to train models or shared, and how it is deleted.
- Decisions. Record that a named editor makes every decision, and that tool output is advisory.
- Disclosure. Publish your AI policy for authors and reviewers, and say how you use tools in the editorial process.
- Appeals. Give authors a route to challenge any decision, including one in which a tool played a part.
- Training and review. Train editors and reviewers on the policy and revisit it at least yearly, because the guidance is changing quickly.
Frequently asked questions
Can reviewers use ChatGPT or similar tools on a manuscript?
Not by uploading it. ICMJE, NIH and Springer Nature all point to confidentiality as the reason. Where a journal permits any AI use by reviewers, it should say so in writing and require disclosure.
Can an AI tool be a peer reviewer?
No guidance we reviewed treats an AI tool as a reviewer. Responsibility for the review and the decision stays with people.
Do we have to tell authors we use software in screening?
Transparency is the safe course. Describe the checks you run in your public peer-review and editorial policies, and say that decisions are made by editors.
Should an AI similarity or integrity flag lead to automatic rejection?
No. A flag is a reason to look, and a person should assess the evidence before any decision or any contact with the author.
NeucitePress services for journal editors
Three ways we support editorial policy and screening. Your editors keep authority over reviewer selection, assessment and acceptance.
A 4-hour live virtual workshop for up to 15 people on COPE Core Practices and real cases, with a reviewer segment on confidentiality, a 1-hour follow-up Q&A and 12 months of materials.
A twelve-month package: policies written for your journal and aligned with COPE and ICMJE, board renewal, integrity screening for an agreed volume of submissions, an ethics case retainer, training and an annual compliance review.
Each new submission is checked for completeness, scope, formatting, ethics and consent statements, similarity and your own checklists before it reaches the editor. It does not judge merit or make desk-rejection decisions.
Sources
- ICMJE — Artificial Intelligence-Assisted Technology
- COPE — Authorship and AI tools (position statement)
- NIH — The Use of Generative AI Technologies is Prohibited for the NIH Peer Review Process (NOT-OD-23-149)
- Springer Nature — Editorial policies, including AI
- Liang et al. — GPT detectors are biased against non-native English writers, Patterns (2023)
