Key Takeaways
A Lancet audit of 2.47 million open-access biomedical papers found 4,046 fabricated citations across 2,810 papers.
Fabricated references rose from about 4 to 57 per 10,000 papers after mid-2024, a more than twelvefold jump that tracks AI writing-tool adoption.
The authors stop short of blaming AI alone, also flagging paper mills and misconduct, and found 98.4 percent of affected papers saw no publisher action.
An automated audit of nearly 2.5 million open-access biomedical papers found fabricated citations climbing sharply after mid-2024, according to a study published in The Lancet on May 7.
Researchers led by Maxim Topaz at Columbia University screened 2,471,758 papers and more than 125 million references in PubMed Central from January 2023 to February 2026, and flagged 4,046 fabricated citations across 2,810 papers.
The rate held near 4 fabricated references per 10,000 papers through 2023, then jumped more than twelvefold, reaching about 57 per 10,000 in early 2026. By the first weeks of this year, roughly 1 in 277 papers cited a source that did not exist, up from 1 in 2,828 in 2023, according to Retraction Watch.
The sharp climb began in mid-2024, when papers written with help from large language models started appearing in the index. The timing lines up, but the authors do not pin it on AI alone. Fabricated references can come from paper mills, deliberate misconduct, or uncritical use of AI tools, and indexing changes could play a part. Review articles were hit hardest, with a fabrication rate 57 percent higher than other papers.
The audit also found little cleanup. At the time of analysis, 98.4 percent of affected papers had received no publisher action. The findings drew fresh attention this week as researchers and technologists debated how exposed the permanent record now is to AI-generated work.
The pattern here is hard to ignore. Citations are the trust layer of science, the assumption that a reference points to something real, and that layer is cracking just as the tools to generate convincing fakes go mainstream. It echoes the push to put cryptographic receipts on the data that trains AI models, and it lands on the same problem Yale named when it audited its own collapse in public trust. The fix is detection and disclosure, not faith. Worth watching whether publishers move.
People Also Ask
What did the Lancet citation audit find?
It screened 2.47 million open-access biomedical papers and found 4,046 fabricated citations across 2,810 papers, with the rate rising sharply after mid-2024.
Are AI writing tools causing fabricated citations?
The timing tracks the spread of large language models, but the authors also cite paper mills and misconduct and stop short of blaming AI alone.
How common are fake references in research papers now?
About 1 in 277 papers in early 2026 cited a source that did not exist, up from 1 in 2,828 in 2023, roughly 57 per 10,000.
What happens to papers with fabricated citations?
The audit found 98.4 percent of affected papers had received no publisher action at the time of analysis.
Sources
The Lancet, Retraction Watch, STAT News