AI ‘Hallucinated’ Citations Are Flooding Medical Research: What It Means for Your Health

The next time your doctor recommends a treatment based on the latest medical study, you might assume that study is built on solid, verifiable science. But a troubling new investigation reveals that a growing number of research papers contain fake references—citations that look real but lead nowhere. Experts call this problem the “tip of the iceberg,” and it has serious implications for patient safety, medical guidelines, and the trust we place in scientific literature.

A team of researchers led by Dr. Maxim Topaz from Columbia University analyzed nearly 2.5 million biomedical papers published between January 2023 and February 2026. They discovered that fabricated citations—references that appear to be real but do not correspond to any actual publication—are skyrocketing. The findings, published as a correspondence in The Lancet, show that the rate of these fake references has jumped more than 12-fold in just three years. In 2023, only about 4 out of every 10,000 papers contained a fabricated citation. By early 2026, that number had climbed to nearly 57 out of every 10,000 papers.

The culprit? Artificial intelligence. Large language models (LLMs)—the same technology behind popular AI writing assistants—are known to “hallucinate.” They generate information that sounds convincing but is completely made up. In the context of research papers, this means AI tools can create references that are correctly formatted, attributed to real researchers, and given plausible publication dates. These fake citations are extremely difficult for human reviewers to catch during the standard peer-review process.

Dr. Topaz told MedPage Today that this research was inspired by a personal experience. He regularly uses AI to polish his own writing. After submitting a paper to a journal, the editorial staff notified him that one of his citations did not lead to an actual source. He described the experience as “deeply embarrassing” and decided to investigate how widespread the problem really is.

How the Study Was Conducted

To understand the scope of the issue, Topaz and his team built an automated reference verification system. They scanned the PubMed Central Open Access database, which contains a large collection of biomedical research papers. Out of 125.6 million references across nearly 2.5 million papers, they were able to verify 97.1 million references that had a PubMed identifier (PMID). The remaining references were excluded from the analysis.

For each verified reference, the team used text-similarity scoring to compare the citing paper’s claimed metadata with the actual publication record. References that appeared mismatched went through additional filters to rule out simple formatting errors. Any reference that passed all filters was then checked against four major databases: PubMed, Crossref, OpenAlex, and Google Scholar. If a reference could not be found in any of these databases, it was classified as fabricated. References that were found but linked to the wrong identifier were considered reference errors.

In total, the team identified 4,046 fabricated references across 2,810 papers. The numbers tell a stark story. In 2023, about 1 in every 2,828 papers contained at least one fake reference. By 2025, that ratio had dropped to 1 in 458 papers. In the first seven weeks of 2026 alone, it was 1 in 277 papers. The fabrication rate increased from 4.0 per 10,000 papers in 2023 to 56.9 per 10,000 papers in early 2026. The researchers noted a sharp spike in fabricated references in mid-2024, which coincides with the lag time between the growing adoption of LLMs and the eventual publication of papers that used them.

Alarming Examples Discovered

The study uncovered some particularly disturbing cases. In one 2025 paper published in an open-access oncology journal, a full 60% of the references were fabricated. The topic was ureteroileal anastomotic techniques—a highly specific surgical procedure. Each fake reference was tailored to that narrow topic and attributed to real urologists. This suggests that the AI was not just generating random citations but was creating references that seemed perfectly relevant to the paper’s subject matter.

The researchers also found patterns consistent with “paper mill” activity—organizations that produce fake or low-quality research for profit. For example, an author duo appeared across 11 papers in a single surgical journal in 2025, and those papers contained 15 fabricated references in total.

Dr. Topaz emphasized that these findings highlight a critical need for human oversight. “AI can hallucinate at any point, and we need to really be careful about that,” he said. He stressed that both authors and journals share the responsibility of double-checking every part of the research process.

What Experts Say About the Growing Crisis

In an accompanying editorial in The Lancet, Dr. Howard Bauchner of Boston University and Dr. Frederick Rivara of the University of Washington—both former editors of the prestigious JAMA—argued that the primary responsibility for catching fabricated citations falls on the authors themselves. They warned that the problem is only going to get worse as AI and LLMs become more common in writing and editing.

Bauchner and Rivara took a firm stance: any paper in which a fabricated reference is detected should be retracted. “A fabricated reference in a published manuscript is consistent with research misconduct and should be treated as such,” they wrote. This is a strong statement, as retraction is usually reserved for cases of fraud, data fabrication, or serious ethical violations.

Dr. Arjun K. Manrai of Harvard Medical School, who was not involved in the study, told MedPage Today that many AI researchers have known about the problem of fabricated references for some time, but it has barely been studied until now. “Fabricated references should concern us all,” Manrai said. “If our scientific foundation erodes, we are in deep trouble.”

How This Affects You as a Reader and Patient

You might wonder: Why should I care about fake citations in obscure academic journals? The answer is that these citations can have a direct impact on your health. Medical research is the foundation of clinical practice. Doctors rely on studies, systematic reviews, and clinical guidelines to make treatment decisions. If those guidelines are built on papers with fabricated references, the recommendations could be flawed or even dangerous.

Dr. Topaz noted that his team is continuing their analysis and has found that fabricated citations are spreading deeper into the literature—into systematic reviews and clinical guidelines. “This is the most alarming piece,” he said. “You would find yourself providing clinical care as a clinician based on the guideline, and then you look at the articles that are cited in this guideline and you see that those citations lead to nowhere. That’s very concerning.”

For the average person, this means it is more important than ever to be an informed patient. If your doctor recommends a treatment based on a recent study, you can ask questions. Where was the study published? Has it been verified by other researchers? Is the journal reputable? While you cannot check every citation yourself, being aware that the problem exists is the first step.

Four Strategies to Fix the Problem

Dr. Topaz and his co-authors proposed four concrete strategies to address the growing issue of fabricated citations:

– Automated reference verification before peer review: Publishers should implement software that automatically checks every reference against existing databases before a paper is even sent out for review. This would catch fake citations early in the process.

• Integrity metadata in indexing services: Databases like PubMed should add integrity flags to article records. If a paper is found to contain fabricated references, that information should be permanently attached to the record so future researchers know to be cautious.
• Retroactive screening of published research: Publishers should go back and check already-published papers for fabricated references. This is a massive undertaking, but it is necessary to clean up the scientific record.
• Establish fabricated references as a category in research integrity databases: Currently, there is no standard way to report or track fabricated citations. Creating a dedicated category would help researchers and publishers monitor the problem over time.

Limitations of the Study

No study is perfect, and this one has important limitations. The researchers excluded nearly a quarter of all references because they did not have a PubMed identifier. This could bias the estimates, as papers without PMIDs may come from less rigorous journals. Additionally, the team only looked at papers in PubMed Central, which is just one subset of all biomedical research. The true scale of the problem could be much larger.

Dr. Topaz acknowledged that his team is continuing to analyze the data and expects to find even more fabricated citations as they expand their search. The “tip of the iceberg” comment suggests that what they have found so far may only be a small fraction of the total.

Practical Takeaways for Readers

– Be skeptical of new studies: Not all published research is trustworthy. Look for papers in well-known, peer-reviewed journals.

• Ask your doctor questions: If a treatment recommendation is based on a recent study, ask where it was published and whether the findings have been confirmed by other research.
• Support transparency: Encourage medical organizations and publishers to adopt automated verification tools. The more we demand accountability, the safer the scientific literature becomes.
• Understand the limits of AI: AI writing tools are convenient, but they can make mistakes. Never assume that a citation generated by an AI is real. Always verify.

The Bottom Line

The rise of AI-generated fabricated citations is a serious threat to the integrity of medical research. With rates increasing more than 12-fold in just three years, experts warn that the problem is only going to get worse. The responsibility falls on authors, publishers, and indexing services to catch these fake references before they contaminate clinical guidelines and patient care. For now, the best defense is awareness. If the scientific foundation erodes, as Dr. Manrai warns, we are all in deep trouble. Staying informed and asking questions is the most powerful tool you have.

Source: MedPage Today