AI Is Fabricating Citations in Biomedical Studies, Researchers Find

Why It Matters

Fake citations erode the evidentiary foundation of medical guidelines, potentially compromising patient care and undermining trust in scientific publishing.

Key Takeaways

• Over 4,000 fake citations identified in biomedical literature.
• Fabricated references increased twelvefold in the past three years.
• No retractions; bogus citations remain in about 3,000 papers.
• AI tools can invent studies and attribute them to real authors.
• False citations risk shaping clinical guidelines and patient treatment decisions.

Pulse Analysis

Artificial intelligence’s ability to generate plausible‑looking text has outpaced its capacity for factual verification, leading to a new kind of scholarly error: fabricated citations. Researchers at Columbia’s School of Nursing scanned a massive corpus of biomedical literature and flagged more than 4,000 references that point to studies that never existed. The phenomenon, often called "AI hallucination," has accelerated dramatically, with the volume of bogus citations growing twelvefold in just three years. This surge reflects both the rapid adoption of AI writing assistants and the limited safeguards built into many of these tools.

The ripple effects extend far beyond academic embarrassment. Clinical guidelines, which synthesize peer‑reviewed evidence to inform treatment protocols, rely on the integrity of cited research. When phantom studies infiltrate the citation network, they can subtly shift recommendations, leading physicians to base decisions on nonexistent data. Such contamination threatens patient safety, erodes confidence in the medical literature, and places additional burdens on editors and reviewers who must now verify not only the content but also the existence of every referenced work.

Addressing the crisis will require a multi‑pronged approach. Publishers should implement automated citation‑validation pipelines that cross‑check references against reputable databases before acceptance. AI developers must embed stricter source‑verification mechanisms and clearly disclose when a citation is generated. Researchers, too, bear responsibility for meticulous fact‑checking, especially when using generative tools. As AI becomes entrenched across scientific domains, the community’s vigilance will determine whether these fabricated references remain an isolated glitch or become a systemic threat to the credibility of scholarly communication.