AI-Fabricated Citations In Over 2,800 Biomedical Journal Articles

More than 2,800 published biomedical journal articles contain citations fabricated by artificial intelligence, a new correspondence in The Lancet reveals. Over a three-year period, researchers identified 4,046 references that were entirely made up, likely generated by large language models and other AI tools.

The Lancet correspondence, authored by a team of scientists and journal editors, details how a systematic review of articles from 2023 to 2025 uncovered these phantom references. The fabricated citations appeared in papers across high-impact biomedical journals, including those from major publishers like Elsevier and Springer Nature. The discovery came after routine checks flagged citations that did not exist in any database.

This crisis stems from the rapid adoption of AI writing assistants—especially ChatGPT and similar models—by researchers under pressure to publish. While some journals have updated policies to require disclosure of AI use, enforcement remains inconsistent. The problem is compounded by peer reviewers who rarely verify every reference.

Key details from the Lancet analysis: 2,810 articles contained at least one fabricated reference, with some papers featuring multiple fake citations. The highest concentration was in articles published in 2024 and early 2025, coinciding with the peak of generative AI adoption in academia. Affected fields include oncology, cardiology, and neurology. The authors manually sampled 500 of the suspect references and confirmed 100% were nonexistent.

"This is a systemic failure of the peer review and editorial process," said Dr. Amelia Zhou, a co-author of the correspondence and editor-in-chief of a leading internal medicine journal. "AI is not inherently bad, but its misuse is undermining the integrity of scientific literature." The study did not name specific authors or institutions, but noted the problem is global.

The broader implications are severe: biomedical research is the foundation of clinical decision-making. Fabricated citations can lead researchers down dead ends, waste funding, and potentially harm patients if erroneous data is cited in future studies. Medical librarians and fact-checkers now face a monumental cleanup task.

Looking ahead, publishers are expected to deploy automated citation verification tools at scale. Crossref, the DOI registration agency, has announced a pilot program to validate references in real time. Some journals are retroactively auditing papers published since 2023. The Lancet correspondence recommends that all journals require authors to submit verification of every reference as a condition of acceptance. Without drastic action, the contamination of the scientific record could worsen as AI generation tools become more sophisticated.

The takeaway: the biomedical literature is suffering from an invisible AI-generated citation infection that threatens decades of research integrity. The next battle will be detection at scale.