An extensive audit of 2.5 million biomedical science papers has uncovered nearly 3,000 studies containing fake citations, exposing a growing challenge in research integrity.

The investigation, described in a paper published in The Lancet on May 7, marks the first attempt to quantify the scale of fabricated references within the biomedical literature.

Researchers developed an automated pipeline that scanned papers indexed in PubMed Central, focusing on publications issued between January 2023 and February 2026.

Their analysis revealed that the presence of fake citations has surged over time, with 12 times more papers containing fabricated references in 2025 compared with 2023.

Here's What They're Not Telling You About Your Retirement

Maxim Topaz, an AI researcher at Columbia University and a co-author of the study, said the true scope of the problem is likely far larger than what was detected.

“What we identified is the lower bound of true prevalence. We’re scratching the tip of the iceberg,” he said.

Kathryn Weber-Boer, director of scientometrics at Digital Science in London, also considered the research a strong opening step toward addressing the issue. She described the study as “a solid first initial contribution to the problem.”

This Could Be the Most Important Video Gun Owners Watch All Year

A separate Nature analysis released in April estimated that about 1.6 percent of 2025 publications included at least one citation linked to a publication that appeared not to exist.

Topaz and his team examined a total of 125.6 million references cited by the 2.5 million papers. They concentrated on 97 million references that contained valid Digital Object Identifiers (DOIs) or PubMed IDs.

Using large language models, the researchers compared each cited title with the article title associated with its DOI or PubMed ID.

They also searched four major databases — PubMed, Crossref, OpenAlex, and Google Scholar — to determine whether each reference truly existed.

If a citation did not appear in any of these databases, it was flagged as fabricated.

The audit eventually identified 2,564 papers with one or two fake citations and another 246 papers with three or more.

Weber-Boer said it is still unclear whether the fabricated references were created by humans or by artificial intelligence tools.

However, she noted that “the growth in the problem suggests that there is a generative AI component.”

A manual review of 500 flagged references by three independent experts confirmed that the majority were indeed fabricated, validating seven out of ten suspected cases.

Even with these findings, the authors say the problem is likely underestimated.

According to Weber-Boer, “Google Scholar is not a reliable source” for verifying citations because some fabricated references can appear on the platform without linking to genuine publications.

The study underscores how easily false information can infiltrate the scientific record and highlights the urgent need for tools and standards to maintain trust in biomedical research.