Artificial intelligence (AI) is reshaping peer review from several directions at once. Authors use AI tools to prepare and polish manuscripts. Reviewers may use them to support parts of their assessment. Publishers are introducing automated checks before submissions reach reviewers, while AI-generated text, images, and data are appearing in manuscripts with varying levels of transparency.
Ethical AI peer review depends on knowing where human judgment remains essential, where AI can provide limited assistance, and where its use creates new responsibilities. The central question is simple: does the technology support the reviewer’s expertise, or does it begin to replace the intellectual assessment that the reviewer was invited to provide?
What can AI not replace in ethical peer review?
AI cannot replace the expert judgment at the center of peer review. Reviewers assess whether a research question is worth investigating, whether the methods are appropriate for the field and context, whether the evidence supports the conclusions, and whether the work contributes meaningfully to existing knowledge. These decisions require subject expertise, awareness of disciplinary standards, and the ability to recognize what a manuscript has overlooked or left unexplained.
A reviewer who delegates the substantive evaluation of a manuscript to an AI system is no longer providing the independent human assessment that the editor requested and this principle is central to AI peer review ethics. AI may support parts of the process, but responsibility for reading, reasoning, evaluating, and recommending must remain with the reviewer.
How can AI tools compromise reviewer confidentiality?
AI tools can compromise reviewer confidentiality when unpublished manuscript material is uploaded to a system that is not explicitly approved for confidential peer review. A manuscript under review is privileged, unpublished material. Reviewers receive access only because the authors and journal trust them to protect it. Uploading manuscript text, tables, figures, data, or supplementary information to an external AI service may expose that material beyond the authorized review process. The risk does not disappear because the reviewer submits only one section or asks a narrow question. Even a methods paragraph, abstract, figure caption, or statistical result may contain confidential information. Before using any tool during ethical AI peer review, ask:
- Does the journal explicitly permit this form of AI use in peer review?
- Does the tool’s privacy policy clearly explain how inputs are stored, processed, and reused?
- Can reviewer confidentiality be protected throughout the process?
- Could I explain this use openly to the editor and authors?
If you cannot answer these questions confidently, do not submit manuscript content to the tool.
Where can AI legitimately assist peer reviewers?
AI can legitimately assist when it supports the reviewer’s own work without processing confidential material or generating the substantive evaluation. The most defensible forms of AI use in peer review are limited, transparent, and consistent with the journal’s current policy.
Improving the reviewer’s own writing
A reviewer may use a grammar or language tool to improve the clarity of a report they have already written, provided the report does not contain confidential manuscript content and the journal permits this use. This is different from asking AI to decide what the manuscript’s weaknesses are or to generate the report from the submission. The analysis must already be the reviewer’s own.
Supporting literature discovery
AI-powered search tools can help reviewers discover published research they may have missed. This may be acceptable when the tool searches public, indexed literature without requiring the reviewer to upload the unpublished manuscript. The reviewer must still verify every suggested source. AI-generated references can be incomplete, irrelevant, or fabricated, so no citation should be included in a review without checking the original publication.
Using journal-approved tools
A journal may use AI for submission checks, reviewer matching, image screening, plagiarism detection, or other parts of its editorial workflow. Reviewers may also be given access to tools operating within a controlled journal environment. Where the journal has explicitly approved a tool and explained its permitted use, reviewers should follow those instructions. They should not assume that approval of one AI function permits all other forms of AI use in peer review. For a broader introduction to these boundaries, see ReviewerOne’s guide to AI in peer review: What reviewers need to know.
When does AI use in peer review require disclosure?
AI use should be disclosed whenever the journal requires it or when the technology has contributed substantively to the evaluation or review report. Journal requirements vary. Some policies focus on AI-generated text. Others require reviewers to disclose any use that helped assess the manuscript’s claims, methods, evidence, or conclusions. Before submitting a review, check the journal’s current reviewer instructions. Do not rely on the policy you remember from a previous review, even if you are reviewing for the same publisher. AI peer review ethics and disclosure requirements continue to change. If disclosure is required, explain briefly:
- Which tool was used
- What it was used for
- Whether manuscript content was processed
- How the output was checked
- Which parts of the assessment remained entirely your own
A concise disclosure might state: “I used an AI-assisted language tool to improve the clarity of my completed review report. I did not upload the manuscript or use the tool to generate the assessment or recommendation.” Disclosure does not remove responsibility. Reviewers remain accountable for every comment, citation, judgment, and recommendation in the report. When the journal’s policy is unclear, contact the editor before using the tool. Disclosure after the review cannot correct a confidentiality breach that has already occurred.
Key takeaways
Ethical AI peer review keeps human expertise, reviewer confidentiality, transparency, and accountability at the center of the assessment. AI can support limited parts of a reviewer’s work, such as improving the clarity of a completed report, searching published literature, or operating within an approved journal system. It should not replace close reading, methodological assessment, interpretation, or the reviewer’s final recommendation. Before using any AI tool, check the journal’s current policy and understand how the tool handles data. Never upload confidential manuscript material simply because the task appears minor.
ReviewerOne applies the same principle by using AI to support, rather than replace, reviewer judgment. Its AI-assisted tools help reduce mechanical review work so reviewers can focus more of their time on evaluating the methods, evidence, interpretation, and contribution of a manuscript. Sign up for ReviewerOne’s AI-assisted peer review platform.
