The Duplicate Review Checker screens peer reviews for suspicious content.Credit: Grace Cary/Getty
A first-of-its-kind artificial-intelligence tool detects duplicate or suspiciously similar peer-review reports.
The system — developed by Institute of Physics Publishing (IOPP), based in Bristol, UK — can help to uncover cases of plagiarized or template-like peer reviews, which individuals or organized groups use to push manuscripts through the publication process or boost citations to their own work.
The technology aims to “combat this pressing issue”, said Lauren Flintoft, the IOPP’s research integrity manager, who presented the findings of a pilot study testing the tool at the 9th World Conference on Research Integrity in Vancouver, Canada on 4 May. Most integrity checks focus on manuscripts, but “peer-review process fraud is equally as important and targeted probably more often than you think by bad actors”, Flintoft told the meeting.
In an analysis of around half a million peer-review reports for manuscripts submitted to the IOPP between 2020 and 2025, the AI tool identified nearly 2,500 reports that had at least 60% overlap with former reviewer reports, of which 785 reports had at least 80% overlap. Of the flagged cases, 89 were exact duplicates.
The IOPP is rolling out the tool across all its journals, the publisher announced on 5 May.
Review mills
Two studies, of open-peer-review reports1,2 — which involve reviews being posted alongside published papers — discovered hundreds of reviews for manuscripts across different publishers that seem to have been copied from a template. These reports, for example, contained similar wording or identical typos, and commonly included citations of the reviewers’ own papers. Research-integrity sleuth Maria Ángeles Oviedo-García, who studies marketing at the University of Seville in Spain and co-authored both analyses, called these cases “review mills”.
At the publisher PLOS, there have been “rising cases of peer-review integrity issues over the past few years”, says Renee Hoch, head of publication ethics in San Francisco, California.
Scientists hide messages in papers to game AI peer review
Between 2024 and 2026, PLOS investigated around 150 papers because of concerns about peer-review integrity. The publisher has so far retracted 40 articles and flagged 55 others as being under investigation. “A subset of these cases involved concerns about peer-review reuse or similarities”, says Hoch.
“I suspect that what we have seen and what has been reported publicly is only the tip of a very large iceberg,” she adds.
Two-stage screening
In response to this growing problem, the IOPP built the Duplicate Review Checker (DRC) to screen peer-review reports in two stages.
“The possibilities for the review mills are endless in the absence of plagiarism software for review reports,” said Flintoft.
For every report, an AI model filters out the 50 most semantically similar peer-review reports in the publisher’s corpus. Then, a text algorithm compares the characters between these reports to spot exact matches and flag what proportion of their text overlaps.
