
Peer review is a cornerstone of scholarly publishing, but it is under strain. Editors struggle to find reviewers, review times are long, and the number of submissions keeps rising. Generative AI tools promise to help by summarising manuscripts, checking statistics, and even drafting review comments. At the same time, they raise serious concerns about confidentiality, accuracy, and accountability.
This article on AI in peer review explores how AI is being used, its potential benefits, the main risks, current policies from journals and funders, and best practices for editors and reviewers.
Table of Contents
How AI Is Being Used in Peer Review
AI can play different roles at different stages of the publishing process.
| Stage | Example AI Uses |
|---|---|
| Initial screening | Checking formatting, scope, plagiarism, and image problems |
| Reviewer selection | Suggesting experts based on publication records |
| Integrity checks | Flagging paper mill patterns or suspicious references |
| Statistical checks | Detecting reporting errors or inconsistencies |
| Review writing | Summarising manuscripts or drafting comments |
Publisher-run tools for screening and integrity checks are generally treated differently from individual reviewers using public chatbots.
Potential Benefits of AI in Peer Review
- Faster screening: automated checks can catch obvious problems early
- Better integrity detection: tools can identify duplicated images, tortured phrases, or fabricated references
- Reviewer matching: AI may help editors find suitable reviewers more quickly
- Consistency checks: spotting statistical or reporting errors that humans may miss
- Language support: helping reviewers who write in a second language express comments clearly
Key Risks and Concerns
Confidentiality
Manuscripts under review are confidential. Uploading them to public AI tools may expose unpublished research, data, or ideas to third parties, and may breach journal agreements. This is one of the main reasons many publishers restrict reviewer use of generative AI.
Accuracy and Hallucinations
AI tools can misunderstand methods, overlook key issues, or invent problems and references that do not exist. A review based on inaccurate AI output could unfairly harm authors.
Lack of Expert Judgement
Peer review requires deep subject knowledge, critical thinking, and an understanding of context. AI cannot fully assess novelty, significance, or ethical issues.
Accountability
Reviewers are responsible for their reports. If AI writes much of a review, it becomes unclear who is accountable for its content.
Bias
AI systems may reflect biases in their training data, for example favouring certain writing styles, institutions, or regions.
Manipulation
There have been reports of authors hiding instructions in manuscripts, such as invisible text asking AI tools to give positive reviews. This highlights how relying on AI could be exploited.
What Journals and Funders Say
Policies are evolving, but common themes include:
- Reviewers should not upload confidential manuscripts into generative AI tools that do not protect confidentiality
- Reviewers remain fully responsible for the content of their reviews
- Any permitted AI use should be disclosed to the editor
- Some funders prohibit reviewers from using generative AI to evaluate grant applications
- Publishers may use their own secure tools for screening and integrity checks
Because policies differ widely, reviewers should always read the specific journal’s or funder’s guidelines before using any AI tool.
Acceptable vs Risky Uses
Often Acceptable (Check Policy)
- Using AI to improve the language of your own review comments, without entering manuscript content
- Using secure, publisher-approved tools provided for reviewers
- Using AI to learn about a general method or concept unrelated to the confidential content
Usually Not Acceptable
- Uploading full manuscripts or confidential data into public AI tools
- Letting AI write the review without careful expert evaluation
- Submitting AI-generated reviews without disclosure
- Using AI to evaluate grant proposals where funders forbid it
Best Practices for Reviewers
- Read the journal’s AI policy before starting
- Protect manuscript confidentiality at all times
- Base your assessment on your own expertise and reading
- Check any AI-assisted text for accuracy and fairness
- Disclose AI use to the editor where it is permitted
- Take full responsibility for your review
Best Practices for Editors and Publishers
- Publish clear AI policies for authors, reviewers, and editors
- Provide secure, approved tools if AI support is offered
- Use AI screening as support, not a replacement for human judgement
- Monitor for hidden prompts and manipulation attempts
- Train editorial staff on AI risks and benefits
- Be transparent with authors about how AI is used in the process
Why Human Judgement Still Matters
Good peer review goes far beyond spotting errors. Experienced reviewers ask whether a study addresses an important question, whether its design is appropriate for its claims, whether alternative explanations have been considered, and whether the work raises ethical concerns. They also draw on knowledge of the field that may not appear in any published text. These judgements depend on expertise, context, and responsibility, which AI tools cannot fully provide.
A Simple Decision Checklist for Reviewers
- Does the journal allow AI use by reviewers at all?
- Will any confidential manuscript content be shared with the tool?
- Is the tool secure and approved by the publisher?
- Will I personally verify every point in my review?
- Have I disclosed the use to the editor where required?
If the answer to any of these questions raises doubts, it is safer not to use the tool.
AI and Reviewer Shortages
Some supporters hope that AI could ease the shortage of reviewers by handling routine checks, allowing human experts to focus on scientific judgement. Others worry that heavy reliance on AI could reduce the depth of review or encourage superficial reports. Most current approaches aim for a balance, using AI for technical screening while keeping human reviewers responsible for evaluation.
What Authors Should Know
Authors should expect that journals may use AI tools for screening, particularly for plagiarism, image integrity, and reference checks. They should also be aware that attempting to manipulate AI-assisted review, such as by hiding instructions in manuscripts, is a serious breach of publication ethics.
The Future of AI in Peer Review
AI is likely to play a growing role in supporting editorial processes, particularly in detecting fraud and checking technical details. However, most experts believe human reviewers will remain essential for judging the quality, originality, and significance of research. The challenge will be to use AI in ways that save time without sacrificing confidentiality, fairness, or trust.
Training and Awareness
Many reviewers are unsure what is allowed. Short training sessions, clear reviewer guidelines, and examples of acceptable and unacceptable AI use can help reviewers make informed choices and protect authors’ trust in the peer review process.
Questions Editors Should Ask
Editors considering AI tools can ask whether the tool keeps manuscripts confidential, how its outputs are checked, whether reviewers and authors are informed, and how decisions remain under human control. Clear answers to these questions help build trust in AI-assisted editorial processes.
Final Thoughts
AI in peer review offers real potential to speed up screening, detect misconduct, and support reviewers. But it also brings risks to confidentiality, accuracy, and accountability. By following journal policies, protecting confidential manuscripts, disclosing AI use, and keeping human expertise at the centre of evaluation, reviewers and editors can make the most of AI while preserving the integrity of peer review.





