Generative AI systems can create references to papers that do not exist and make them look real. In medicine, a fabricated or poorly verified reference is not simply a bibliographic error. It weakens the chain of evidence on which reviews, guidelines, teaching, and clinical judgment may depend.
A recent article by Topaz and colleagues in The Lancet highlights an important and growing threat to scientific integrity: the increasing appearance of fabricated references within the biomedical literature, with implications for scholarly publishing more broadly.
Studies by Ibrahim and colleagues on citation manipulation through citation mills and preprint servers, and by Smarsly on the vulnerability of citation metrics in the era of generative AI, have shown that AI-generated or AI-assisted documents can contribute to citation-metric manipulation, especially at the level of authors, profiles, and individual articles. Whether such practices have already contributed to inflated official journal impact factors is less certain, although the possibility cannot be dismissed.
When we encounter a reference clearly out of place in a student assignment, many of us may initially dismiss it as AI-generated, assuming that the original source was never carefully verified. Walters and Wilder, in their study of fabricated and erroneous ChatGPT-generated citations, showed how AI systems can produce references that appear authentic and convincing but do not correspond to real publications. Verification therefore remains fundamental.
Real physician voices, twice a week
Free, and one click to unsubscribe.
Part of the solution may lie in how AI systems handle references. Wherever possible, AI-assisted writing tools should anchor citations to verified records, including DOIs, PubMed entries, publisher pages, or other authoritative metadata. The citation should begin with a verifiable source, not end with a plausible reconstruction. At the same time, improved algorithms cannot replace author responsibility. A reference is valid only when it can be traced to a real source that the author has examined.
Apart from AI-generated inaccuracies, there are several other ways in which inaccurate references can enter the scientific literature. One familiar practice from earlier years was indirect or secondary citation, where authors cited references they had not personally examined in their original form. Earlier studies on reference accuracy in medical journals and citation-copying behavior had already shown that such problems existed well before AI. Practical limitations often contributed to this problem. Journals were expensive, library access depended heavily on institutional resources, and obtaining full articles was often difficult and time-consuming.
The emergence of MEDLINE and resources such as Current Contents and the Science Citation Index greatly improved discoverability by making biomedical and scientific literature easier to identify and track. Later, online indexing and search systems expanded access further, although many investigators still relied heavily on abstracts, citation listings, or secondary summaries rather than complete original articles. Earlier studies of reference accuracy in medical journals and quotation and reference accuracy in surgical journals showed that concerns about citation verification existed well before the emergence of artificial intelligence, although AI now has the ability to generate such inaccuracies at a much larger scale and with far greater plausibility.
Another related issue is the growing reliance on review articles as primary citation sources. Citation patterns are increasingly influenced by journal visibility, search-engine ranking, and article accessibility, leading highly visible secondary sources to be cited preferentially over original foundational studies. As scientific literature becomes increasingly searchable and algorithm-driven, what is most visible is not always identical to what is most original or carefully verified. Although distinct from fabricated references, such patterns can still distort how scientific contributions are recognized and remembered over time.
This essay is cited in the KevinMD record on artificial intelligence.
Rao M. Uppu is professor of environmental toxicology and chemistry at Southern University and A&M College in Baton Rouge, Louisiana, where he has served since 2002. He is also an adjunct professor of chemistry and pathobiological sciences at Louisiana State University.
Trained in biochemistry, physical organic chemistry, and free radical chemistry, his research spans bioanalytical methods, biomarker discovery and validation, chemical toxicology, environmental chemistry, computational genomics, reactive intermediates, and the molecular mechanisms of disease. His scholarship is indexed on ORCID, ResearchGate, and Google Scholar.
Dr. Uppu is a Fellow of the American Association for the Advancement of Science (AAAS), the Royal Society of Chemistry (RSC), the Royal Society of Biology (RSB), the Royal Society of Medicine (RSM), the Royal Society for Public Health (RSPH), and the Academy of Toxicological Sciences (ATS).
In addition to his scientific research, he writes reflective essays on scientific culture, mentorship, peer review, ethics, and the human dimensions of academic life. He shares updates on LinkedIn.
