A major new audit published in The Lancet has exposed the growing infiltration of fabricated citations into the peer-reviewed biomedical literature. Researchers at Columbia University analyzed approximately 2.5 million papers published between January 2023 and early 2026, identifying 4,046 fabricated references across 2,810 papers. These non-existent citations (references whose claimed titles do not match any real publication in major databases) rose dramatically: from roughly 1 in 2,828 papers in 2023 to 1 in 458 in 2025, and further to 1 in 277 in the first weeks of 2026. The surge aligns with the widespread adoption of generative AI tools and the activities of paper mills.
Fabricated citations threaten the reliability of scientific knowledge. They appear in systematic reviews, clinical guidelines, and meta-analyses, potentially skewing evidence-based decisions. Contributors include uncritical use of large language models (which can hallucinate plausible but false references at rates of 30–69% in biomedical contexts), organized paper mills producing low-quality or fraudulent manuscripts, and occasional intentional misconduct. Publishers are responding with enhanced integrity tools: AI-driven reference validation, image screening, and manuscript triage systems designed to flag suspicious patterns before peer review.
For authors, rigorous verification is now essential. Best practices include manually cross-checking every reference against trusted databases such as PubMed, Crossref, or Google Scholar at the drafting stage, using reference management software with built-in validation features, and documenting methodological details with full transparency—including exact protocols, data sources, and any AI assistance. Journals increasingly expect clear disclosures of AI use and may reject submissions with unverifiable citations or inadequate methods sections. Proactive steps, such as depositing raw data and detailed protocols in public repositories, further strengthen credibility.
While these challenges test the foundations of scholarly publishing, the scientific community’s rapid development of detection tools and updated policies signals a strong commitment to safeguarding integrity. By prioritizing verification and transparency in every submission, researchers can help maintain trust in the literature and ensure that genuine advances continue to drive discovery.
Click here for the Japanese version.