Fake Citations Are Spreading Through Biomedical Literature at Accelerating Rates
Fabricated and AI-generated citations are entering scientific papers faster than peer review can catch them. Two large-scale studies published this year document a sharp rise in false references that threatens the integrity of the evidence base underlying clinical guidelines and policy decisions.
A comprehensive audit of 2.5 million biomedical papers found more than 4,000 fabricated references across 2,810 publications. The problem is accelerating: in 2023, roughly one in 2,800 papers contained fake citations. By 2025, that ratio had jumped to one in 458 papers. Early 2026 data suggests the trend is continuing upward.
Most papers with fabricated citations contained one or two false references, but some included multiple entries that made up a significant portion of their bibliography. Review articles showed a 57% higher rate of fabricated references than other study types.
AI Is Generating Hallucinated Citations at Scale
Large language models are producing non-existent citations that look credible. Researchers examining over 111 million references across academic preprint servers and databases estimated that nearly 147,000 hallucinated citations were generated in 2025 alone.
These fake references are difficult to spot. They feature realistic titles, real author names, and proper formatting. Early-career researchers and smaller teams are disproportionately affected, suggesting less access to verification tools or institutional oversight.
The hallucinated citations frequently credit well-established male researchers, reinforcing existing biases in academic attribution.
Current Systems Aren't Catching the Problem
Most papers containing fabricated or hallucinated citations remain uncorrected and unretracted. Peer reviewers lack automated tools to verify references against databases like PubMed, Crossref, and OpenAlex.
The stakes are high. False citations can distort the evidence base that informs clinical practice, treatment guidelines, and research funding decisions. A single fabricated reference in a review article can propagate through subsequent literature, compounding the damage.
What Researchers Should Know
- Over 4,000 fabricated references identified across 2,810 papers in a single audit
- Citations grew 12-fold from 2023 to 2025
- By early 2026, nearly one in 277 papers contained at least one fabricated reference
- AI-intensive fields and AI-assisted manuscripts show higher error rates
- Most affected papers have not been corrected or retracted
What Comes Next
Experts are calling for automated reference verification integrated into journal submission and review workflows. Improved transparency in indexing systems and stronger research integrity frameworks are also needed.
As AI tools become standard in academic writing, journals face pressure to implement detection systems before false citations become too numerous to manage. The alternative is a gradual erosion of trust in published research.
Researchers working with AI writing tools should manually verify citations against primary sources before submission. Journals should require verification checks as a condition of publication.
For those managing research workflows or policy decisions based on published evidence, understanding this problem is essential. The citation you're reading may not exist.
Your membership also unlocks:
