AI-Fabricated Citations Infiltrating Medical Literature May Influence How Doctors Treat Patients

A first-of-its-kind audit of 2.5 million academic papers, published on May 7, 2026 in The Lancet, has identified more than 4,000 papers containing fake references that could not be traced to real publications. The audit analyzed over 97 million references across 2.5 million academic papers, marking the first academic study to examine the scale of fake citations in biomedical literature. One of the culprits, according to researchers, is the use of Artificial Intelligence (AI), which the researchers say poses a “rapidly-growing problem” in biomedicine.^{1 2}

Reporting on the findings, Nature.com noted that “the rate of fabrication in 2025 was more than 12 times greater than that in 2023.” A separate analysis by Nature, published a month prior to the Lancet correspondence, estimated that around 1.6 percent of publications from 2025 contained at least one reference corresponding to a nonexistent publication. Corresponding author and AI researcher Maxim Topaz, who is based at Columbia University’s School of Nursing and Data Science Institute in New York City, says the Lancet findings are “conservative estimates.” He adds, “What we identified is the lower bound of true prevalence. We’re scratching the tip of the iceberg.”^{1 2}

Topaz and colleagues, writing in a correspondence piece published in The Lancet, say fabricated references can arise from “paper mill activity, intentional misconduct, or uncritical use of artificial intelligence (AI) writing tools.” The researchers note:

Large language models (LLMs) generate plausible sounding but fictitious references, a well-documented failure mode; previous studies estimate that 30–69 percent of LLM-generated references in biomedical contexts are fabricated. These references are often correctly formatted, attributed to real researchers, and bear plausible publication dates, making them difficult to detect by conventional peer review. To our knowledge, no systematic audit of reference integrity across the biomedical literature has been conducted until now.^{1 2}

Researchers Used AI to Detect AI-Generated Fake Citations Across Biomedical Literature

To conduct the audit, researchers developed a system to inspect the 125.6 million references cited by 2.5 million papers using large language models (LLM), the same framework that powers popular AI chatbots like ChatGPT and Claude. The team focused on the 97 million references that had valid digital object identifiers (DOIs), a unique string of alphanumeric characters used to permanently identify academic papers, research reports, datasets, or digital documents. They then used LLMs to flag mismatches between the referenced article title and the title of the paper linked to its DOI identifier, also searching for the references in four prominent scholarly databases, including PubMed, Crossref, OpenAlex and Google Scholar.^{1 2}

Commenting on the proliferation of fake citations in medical literature articles over the past few years,  Kathryn Weber-Boer, director of scientometrics at Digital Science Ltd. of London, England said:

Whether they’re fabricated by a computer or fabricated by a human, that’s a question that remains open. The growth in the problem suggests that there is a generative AI component.¹

AI-Fueled Fabrications Add to Growing Institutional Mistrust

Experts continue to ring the alarm about the risks of AI hallucinations, the term for when AI generates false, incorrect,  fabricated or biased information based on popular narratives that sounds confident, logical, believable and unbiased but may be misleading. Most frequently used LLMs feature a small disclaimer on their platforms warning the user to fact-check any outputs. ChatGPT’s disclaimer reads, “ChatGPT can make mistakes. Always check important information.”³

A peer-reviewed study published in Computers in Human Behavior: Artificial Humans examined the role of disclaimers in the perception of AI-generated content and found that “across three studies, disclaimer type did not impact message credibility of an AI-generated text.” The study notes:

Generative AI, and large language models (LLMs) in particular, have become a prevalent source of digital content. Despite their widespread availability, these models come with critical weaknesses, such as a lack of factual accuracy. Being informed about the advantages and disadvantages of these tools is essential for using AI safely and adequately, yet not everyone is aware of them.³

The study’s authors strongly suggest the responsibility should be on the developers and applicators to ensure factual information outputs from LLMs rather than the users…

As our studies suggest, even detailed preliminary information that respondents were forced to read missed to be sufficiently recognized. Therefore, we argue that a footnote cannot fulfill this responsibility adequately, and more intensive efforts should be made to ensure safe use of GenAI.³

Doctors Could Be Making Treatment Decisions Based on Studies That Never Existed

Topaz also warns of the risks of AI hallucinations in fabricated citations and how they affect the practice of medicine specifically, saying, “Fabricated citations are dangerous because they influence clinical guidelines, which are based on public research that health care professionals follow in providing care.” He adds:

When those fake references are making it into the literature, they will end up in those guidelines, and that’s how doctors decide how to provide care for you. Your doctor could be making decisions around treatment based on studies that never existed.⁴

Two-Thirds of Americans Have Little Confidence in Health Care Leaders

The audit’s findings come as distrust in the U.S. medical industry has peaked, with rising health care costs, growing patient-doctor tension due to medical gaslighting, regulatory capture, and pandemic-era polarization. A 2026 study published in npj Digital Medicine found that “two-thirds of Americans now report having only ‘some’ or ‘hardly any’ confidence in healthcare leadership.”

At the same time, despite growing concerns about AI hallucinations and privacy, a growing number of Americans are turning to large language models (LLMs). Many are beginning to prefer consulting LLMs, which are based on AI neural networks designed to generate human-like text, instead of interacting with physicians for the diagnosis and treatment of health conditions and for getting answers to questions about complex and chronic health problems.^{5 6}

Despite Hallucination Risks, Americans Increasingly Turn to AI Generated LLMs Over Physicians

Users cite that AI generated large language models (LLMs) are more collaborative, less judgmental of personal medical choices, and significantly cheaper than seeking a qualified doctor or other health care professional for advice. Users also report that LLMs are more helpful in identifying root causes of a health problem, a practice not standard in allopathic medicine, which focuses on symptom treatment and disease management. A Gallup poll conducted in late 2025 found that roughly one-quarter of U.S. adults had used an AI tool to obtain health information or advice in the past 30 days.⁵

The npj Digital Medicine study suggests that AI generated LLMs will revolutionize the medical industry. “Like the internet before it, LLMs will reshape the landscape of medicine. The question is no longer whether to engage, but how. If clinicians and health systems do not take an active role in guiding their integration, they risk surrendering influence to less qualified voices – undermining both trust and the quality of care. The time has come for medicine not just to adopt LLMs, but to lead its responsible, patient-centered implementation.”⁶

Experts Struggle to Keep Pace With AI

“If you just flood the world with information, truth is bound to lose,” said Yuval Noah Harari, historian, Oxford-trained PhD, and professor at the Hebrew University of Jerusalem in a 2024 interview with Noema magazine. “If you want truth to win and to acquire knowledge and wisdom, you must tilt the playing field. Only if we invest in institutions that preserve the hope of reaching the truth and acquiring knowledge and developing wisdom, can we tilt the balance.”⁷

Dr. Harari is among more than 40 voices featured in The A.I. Doc: Or How I Became an Apocaloptimist, a new documentary film that puts the people most responsible for building, funding, and deploying AI, including OpenAI CEO Sam Altman, on the record with a question that should have a simple answer: is AI good or bad, who is making sure the tech advances safely, and what does the public need to know? Producer Diane Becker described it as “the most challenging movie she’s ever made,” in that “literally the minute we started making it, it was out of date.”⁸

If you would like to receive an e-mail notice of the most recent articles published in The Vaccine Reaction each week, click here.