Fake entries have also surged in references cited in medical papers. It is evidence that "paper mills," which produce fake papers for money, are thriving. While cases of fake papers with manipulated research data have been caught from time to time, this is the first attempt to check how many references cited in papers are fake.
A team led by Professor Maxim Topaz of the Columbia University School of Nursing said it "identified 4,046 references that do not actually exist in peer-reviewed medical papers," in a study published in The Lancet on May 9 local time.
◇ 246 papers had three or more fake references
The Columbia team developed an artificial intelligence (AI) verification system that examined 125.6 million references cited in 2.5 million papers posted on PubMed Central from Jan. 1, 2023, to Feb. 18 this year. PubMed Central is a free databases of life science papers.
The AI verification focused on 97 million references with a digital object identifier (DOI), a unique identifier assigned to digital content, or an ID assigned in the PubMed databases. The review found 4,046 false citations in 2,810 papers. Analysis showed 2,564 papers had one or two fake references, and as many as 246 papers had three or more fake references.
Cases of fake references in papers have surged recently. In 2025 they were 12 times higher than in 2023. The researchers said, "Starting in mid-2024, coinciding with the spread of AI writing tools, we saw the sharpest increase." Kathryn Weber-Boer, a researcher at Digital Science, a U.K. research information management firm, also suspected AI.
She told Nature, "We still do not know whether fake references are fabricated by computers or by humans," but added, "the recent upward trend in these problems suggests the influence of Generative AI."
◇ Could adversely affect patient care
An increase in fake references poses a serious problem. If the sources cited to support a paper's findings are fake, the paper's scientific basis is inevitably weak. That can trigger a vicious cycle in which fake papers spawn more fakes. In particular, because medical papers are directly tied to patient care, the consequences could be fatal.
Topaz said, "Because clinicians make treatment decisions based on clinical guidelines presented in papers, this directly affects patients." In other words, papers that cite fake studies can steer patients toward the wrong treatments. Topaz said, "In one paper we reviewed, 18 of 30 references were fake," adding, "some of them have already been cited elsewhere and even appeared in review papers used to inform clinical practice."
Even so, journals reportedly took no action. The researchers said that at the time of the investigation, 98.4% of problematic papers had received no action from publishers. In a commentary published alongside the study, Howard Bauchner of the Boston University School of Medicine and Frederick Rivara of the University of Washington School of Medicine said, "This study shows the need to improve research ethics."
Based on the findings, Topaz's team recommended that publishers verify references before accepting papers for publication. They also suggested that providers of indexing services for papers supply additional information so users can assess the accuracy of references.
References
Lancet (2026), DOI: https://doi.org/10.1016/S0140-6736(26)00603-3