When Adam Day built a method to compare the language in peer-review reports across scientific journals, he expected to find the occasional lazy reviewer recycling old comments. What he found instead was a pattern so consistent it could only be explained by something more organized: a supply chain of fabricated reviews, assembled and sold by commercial operations known as review mills, quietly steering flawed research past the gatekeepers of science.
Now, a tool built on that insight is in active use. IOP Publishing, one of the world’s major scientific publishers, has deployed the Duplicate Review Checker, or DRC, an AI system designed to scan peer-review reports for suspicious text overlap. According to Nature’s reporting, it is the first such tool deployed by an academic publisher for this purpose, though no independent source has separately verified that claim. Integrated into IOP’s editorial workflow as of early 2026, the tool compares reviewer comments against one another and flags pairs with unusually high similarity. Under honest conditions, two independent reviewers evaluating the same manuscript will produce different critiques in different language. When large blocks of near-identical phrasing appear across unrelated papers, the DRC treats it as a red flag.
Nature’s reporting on the rollout states that the system has uncovered clusters of overlapping reviews pointing to at least one coordinated review mill operating across multiple journals. The reporting does not specify whether IOP itself confirmed that finding or whether Nature’s journalists inferred it from the data the publisher shared.
The research that made detection possible
The DRC did not emerge from a vacuum. Its intellectual foundation traces to a 2022 preprint on arXiv by Day, which demonstrated that high text duplication across independent peer-review reports is a reliable marker of organized fraud. The core logic is simple: identical reviewer language is vanishingly rare when reviews are conducted honestly, so systematic duplication points to coordination rather than coincidence.
A separate peer-reviewed study published in Scientometrics by M. Angeles Oviedo-Garcia examined 263 review reports and defined review mills as a distinct category of misconduct. The pattern she identified is specific and revealing: generic, vague boilerplate language paired with coercive demands that authors cite particular papers. That combination serves two purposes simultaneously. It pushes low-quality manuscripts through the review process, and it artificially inflates citation counts for papers the mill operators want to promote.
The most recent peer-reviewed investigation, by Oviedo-Garcia, Aquarius, and Bishop, was published in Accountability in Research in early 2026 (exact month and DOI not available at the time of this reporting). It documented how one putative review mill operated and why it escaped notice. The finding was sobering: without access to public peer-review comments, posted on platforms like PubPeer, these operations can run undetected for years. Most journals keep reviewer reports confidential, which means the very system designed to protect reviewer candor also shields fraud from scrutiny.
What the tool has revealed so far
IOP has confirmed that the DRC is now part of its standard editorial workflow, meaning new submissions are routinely scanned for duplicated reviewer language before final publication decisions. But the publisher has not released detailed data on how many manuscripts the tool has flagged, what proportion of flagged reviews led to retractions, or how many compromised papers may already sit in the published literature. No retractions, corrections, or other concrete editorial actions resulting from DRC findings have been publicly announced as of June 2026.
That opacity matters. IOP has a reputational interest in presenting the DRC as effective, and no independent audit of the tool’s accuracy or false-positive rate has been published. The supplementary dataset released alongside Oviedo-Garcia’s Scientometrics article offers a rare exception to this pattern: it enumerates the specific review reports analyzed and maps them to detailed records on PubPeer, allowing other researchers to verify the findings independently. That kind of open data is exactly what makes the review-mill research credible, and its scarcity elsewhere in the field is part of what makes the broader problem so difficult to measure.
The gaps that remain
No other major publisher has publicly announced adopting a comparable detection system. Whether Elsevier, Wiley, Springer Nature, or other large houses are developing similar tools, or whether they consider the problem urgent enough to warrant one, remains unknown. Their silence leaves open a basic question: is the review-mill problem concentrated in a few journals, or is it systemic across the industry?
The scale of the fraud is genuinely hard to estimate. Oviedo-Garcia’s sample of 263 reports was large enough to establish patterns but far too small to project the total volume of compromised reviews across all academic journals. The Accountability in Research investigation was designed to document detection methods, not to quantify the full scope of the problem.
Then there is the arms-race question. The DRC relies on text overlap as its primary signal. If review mills shift from recycling human-written boilerplate to generating AI-produced variations that convey the same hollow praise in different words, the detection challenge changes fundamentally. No published study has confirmed this shift yet, but the incentive structure makes it a logical next move for operators whose business model depends on staying invisible.
How review mills work and who they serve
Published research describes review mills as commercial services that sell fabricated or manipulated peer reviews to authors seeking to get papers accepted in journals. The typical operation works by exploiting the reviewer-suggestion feature many journals offer during submission: authors (or agents acting on their behalf) suggest reviewers whose email addresses are actually controlled by the mill. When the journal sends a review invitation to that address, the mill responds with a pre-written, generic report recommending acceptance. Oviedo-Garcia’s Scientometrics study found that these reports often include coercive citation demands, meaning the mill also profits by inflating the citation counts of specific papers or journals. The customers are typically researchers under intense pressure to publish, whether for career advancement, institutional quotas, or degree requirements. No reliable public data exists on what mills charge per review, and the operators remain largely anonymous, making enforcement difficult.
Why this matters beyond academia
Peer review is not just an academic ritual. It is the mechanism that governments, hospitals, regulatory agencies, and journalists use to distinguish credible research from unreliable claims. Drug approvals, environmental regulations, and public health guidance all rest, at some point, on the assumption that published studies have been vetted by independent experts acting in good faith.
The existence of review mills means that assumption has been compromised in ways that were invisible until very recently. The DRC and the research behind it represent the first systematic attempt to detect the problem. But detection alone does not fix it. Journals that keep reviewer reports confidential make it nearly impossible for outside observers to spot manipulation. And even when suspicious patterns surface, publishers face difficult decisions about whether to retract, correct, or simply flag affected articles, with little transparency about which path they choose.
What publishers and researchers must do next
Until more publishers adopt detection tools and move toward open peer review, the scientific community faces an uncomfortable reality: peer review now requires its own layer of independent scrutiny. The process that was supposed to guarantee quality has itself become a target, and the tools to defend it are only just arriving. None of the sources reviewed for this article include direct statements from researchers, editors, or publisher spokespeople beyond what Nature’s reporting conveys. That absence itself underscores how little transparency currently exists around a problem that touches every field of science.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
