Summary: Researchers at the University of Colorado Boulder have developed an artificial intelligence system that automatically scans open-access journals to identify potentially predatory publications. These outlets often accept papers for a fee without rigorous peer review, threatening the credibility and reliability of scientific literature.
The AI examined more than 15,000 open-access journals and initially flagged over 1,400 as suspicious, with a final set of more than 1,000 journals identified as potentially problematic after human review. The platform is intended as a scalable prescreening tool to help researchers, institutions, and publishers spot risky outlets quickly; human experts remain essential for final determinations.
Key Facts
- Predatory publishing: Some journals charge publication fees but do not provide genuine peer review or editorial oversight.
- AI screening: The system analyzed about 15,200 journals and initially flagged more than 1,400, with roughly 1,000 remaining as questionable after expert checks.
- Protecting research integrity: The tool functions as an automated first filter to reduce the burden on human reviewers and to limit the spread of unreliable scientific findings.
Source: University of Colorado
A team of computer scientists led by the University of Colorado Boulder has created a new AI platform designed to detect “questionable” scientific journals.
The study, published Aug. 27 in Science Advances, addresses a growing problem in scholarly publishing: journals that prioritize revenue over rigorous peer review. Daniel Acuña, the study’s lead author and an associate professor in the Department of Computer Science, describes frequent unsolicited emails from unfamiliar journal editors offering rapid publication for a fee—an increasingly common solicitation aimed at researchers.
Predatory journals frequently target researchers in regions where institutional infrastructure is still developing and the pressure to publish is high. By promising rapid publication in exchange for fees—sometimes hundreds or thousands of dollars—these outlets exploit authors who expect legitimate peer review and editorial standards. In many cases, the papers are posted online with minimal or no review.
Acuña notes that efforts to identify and remove predatory outlets can feel like “whack-a-mole”: when one journal is exposed, a new title often appears under a different name or website. To meet the scale of the problem, his team turned to automated methods.
The AI platform inspects journal websites and metadata for a variety of signals: the presence and credibility of editorial boards, the clarity and accuracy of peer review policies, language quality on the site, publication volume, authorship patterns, and citation behaviors. These features help the system estimate the likelihood that a journal is operating without adequate editorial standards.
The researchers trained their model using data from the Directory of Open Access Journals (DOAJ), a volunteer-run registry that has long worked to catalog and vet open-access publications based on established criteria. After training, the AI processed nearly 15,200 open-access titles and initially flagged more than 1,400 as suspicious.
Human experts then reviewed a sample of those flagged journals. According to the expert assessments, the system produced roughly 350 false positives—journals flagged by the AI that appear to be legitimate. Even accounting for those errors, the team identified more than 1,000 journals warranting closer scrutiny. Acuña emphasizes that the tool is meant to prescreen large collections of journals, speeding up triage while leaving final judgments to trained professionals.
The shake down
Peer review remains the cornerstone of scholarly publishing: independent experts evaluate manuscripts to assess methods, results, and interpretations. Predatory publishers undermine that process by charging authors without delivering genuine review or editorial oversight. The term “predatory journals” was popularized in 2009 by librarian Jeffrey Beall to describe this exploitative model.
Such publishers often focus on authors in countries where academic evaluation systems are evolving, offering the promise of quick publication in exchange for payment. In many instances, the advertised review process is superficial or nonexistent—the article is simply posted online.
Organizations like the Directory of Open Access Journals have used volunteer-based screening to identify suspect titles, but manual vetting struggles to keep up with the volume of new and rebranded outlets. Automated systems can help by narrowing the field for human reviewers and highlighting journals that most urgently need attention.
A firewall for science
The team intentionally designed their AI to be interpretable rather than a “black box.” Unlike some large language models, their system provides insight into the features that drive its decisions. For example, the researchers found that many questionable journals publish unusually large numbers of articles, list authors with multiple affiliations more frequently, and show higher rates of self-citation compared with established journals.
The AI platform is not yet publicly available, but the researchers aim to share it with universities and publishing organizations. Acuña describes the tool as a kind of “firewall for science”—a way to protect fields from the contamination of unreliable studies by intercepting suspect outlets early in the process.
“In science, you don’t start from scratch; you build on previous work,” Acuña said. “If that foundation is weak, the entire structure can crumble. Automated screening can help identify and remove weak links before they mislead other researchers.”
About this AI and science research news
Author: Daniel Strain
Source: University of Colorado
Contact: Daniel Strain – University of Colorado
Image: The image is credited to Neuroscience News
Original Research: Open access. “Estimating the predictability of questionable open-access journals” by Daniel Acuña et al., Science Advances
Abstract
Estimating the predictability of questionable open-access journals
Questionable journals threaten global research integrity, yet manual vetting is slow and often cannot keep pace. This study evaluates AI as a scalable approach to identifying suspect venues by analyzing website design, published content, and metadata. Compared against human-annotated datasets, the method achieves useful accuracy and reveals previously overlooked markers of journal legitimacy. By tuning decision thresholds, the approach can emphasize broad screening or low-noise identification. At a balanced threshold, the system flags over 1,000 suspect journals that together publish large numbers of articles, receive millions of citations, report funding from major agencies, and attract authors from developing countries. Error analysis highlights challenges such as discontinued titles, misclassified book series, and small society outlets with limited online footprints—issues that can be mitigated with improved data. The findings demonstrate AI’s potential for scalable integrity checks while underscoring the need to pair automated triage with expert review.