AI-Generated Fake Citations Surge in Scientific Papers, Bypassing Peer Review
At least 146,932 fabricated references created by artificial intelligence entered the scientific record in 2025 alone, with most surviving peer review and appearing in published journal articles. Researchers from Cornell University, UCLA, and UC Berkeley analyzed 111 million citations across 2.5 million papers published between 2020 and 2025 to reach this finding.
The study tracked citations whose titles could not be verified against major academic databases including Semantic Scholar, OpenAlex, and Google Scholar. By comparing post-2022 trends with pre-ChatGPT baselines, researchers isolated the role of AI-generated hallucinations in the sharp increase.
Hallucination Rates Climb Across Platforms
By August 2025, hallucinated citation rates had reached nearly 2% in SSRN papers, 0.4% in arXiv, 0.3% in PubMed Central, and 0.2% in bioRxiv. Monthly estimates of fake citations in PubMed Central alone touched 8,140.
The steepest increase began around mid-2024, roughly 18 months after ChatGPT's public release, as AI tools evolved from writing assistants into citation-generation engines.
Fake references were often sparsely distributed across otherwise legitimate manuscripts, suggesting many researchers copied AI-generated citations without verification. The contamination was not limited to obviously fraudulent papers.
Less Experienced Authors More Affected
Authors linked to hallucinated citations tended to be less experienced, but their publication output grew rapidly-increasing 3.13 times faster on SSRN and more than doubling on bioRxiv compared with matched peers by 2025.
Solo researchers and smaller teams were overrepresented. When hallucinated references pointed to real scientists, they favored prominent scholars-those cited had 68.8% more prior publications and 58.3% more citations than average.
Peer Review Offers Little Protection
Existing safeguards proved inadequate. Nearly 78.8% of fake citations passed arXiv moderation. Among bioRxiv preprints later published in PubMed Central-indexed journals, 85.3% of hallucinated references remained in final published versions.
The problem could become self-reinforcing. As fabricated references become embedded in open-access repositories and citation databases, future AI models trained on those datasets may absorb and reproduce the same hallucinations.
Separate Audit Confirms Escalating Problem
A second study published in The Lancet analyzed biomedical papers from 2023 through early 2026 and identified more than 4,000 fabricated references across 2,810 peer-reviewed papers. The rate of fabricated references rose sharply over the period.
In 2023, roughly one in 2,828 papers contained at least one fabricated citation. By 2025, the figure had worsened to one in 458 papers, and by early 2026, it had climbed to one in 277 papers.
One 2025 paper on ureteroileal surgical techniques in an open-access oncology journal contained 18 fabricated references out of 30 verified sources-60% of the paper's citations.
Nearly 98% of affected papers had faced no publisher action at the time of the audit.
Publishers Need Automated Verification
Researchers warned that fabricated citations could compromise clinical guidelines and systematic reviews. They urged publishers to introduce automated reference verification systems before papers are accepted for publication.
For researchers working in fields where literature review and citation accuracy matter-clinical medicine, systematic reviews, meta-analyses-understanding how large language models generate hallucinations has become essential. The same applies to those using AI tools for research automation.
Your membership also unlocks:
