When we talk about research transparency and credibility, the first thing that comes to mind is whether results are reproducible. We spend a great deal of time checking whether code runs correctly, carefully combing through it to catch even the smallest errors. When everything checks out, there is a sense of relief, as if the entire project hinges on perfectly matching results. Often, we go even further, testing how robust those results are to alternative modeling assumptions and data cleaning choices.
But here is a question we rarely ask: have you ever wondered whether the references check out?
Transparent and credible research extends beyond results and methods; it also depends on citations. Yet references often receive far less attention, with minimal time dedicated to ensuring they are complete, accurate, or even properly formatted. It is not uncommon to encounter papers with inconsistent or poorly formatted references (Onwuegbuzie et al., 2010; Rivkin, 2020). While this may seem like a minor issue, it signals something deeper: problems of credibility and transparency can begin at the very foundation of a paper: the literature it builds upon.
And formatting is not even the main concern. The more serious issue is the presence of fake citations. While this may not have been a widespread concern in the past, it is becoming an increasingly important threat in the age of artificial intelligence (AI).
Since the rise of AI, the problem of fake citations has become more pronounced (Naddaf and Quill, 2026). Large language models can generate text that appears highly credible, complete with citations that look convincing at first glance. However, these references are often fabricated, incomplete, or mismatched, what is commonly referred to as “hallucinated” citations. They can easily go unnoticed, especially when reviewers and readers focus primarily on the results rather than the sources.
Recent cases illustrate how serious this issue can be. In one instance (see Retraction Watch, 2026a), a paper published in Digestive Diseases and Sciences, a Springer Nature journal, was found to contain a “preposterous number” of nonexistent references, with 12 out of 14 citations failing to correspond to real articles. The problem was only uncovered when a librarian attempted to locate the sources and discovered that they simply did not exist. What is particularly concerning is not just the presence of these fake references, but the fact that they passed through peer review and editorial checks without being flagged.
In another example, the issue of fake citations was uncovered almost by accident. A researcher checking a routine citation alert on Google Scholar noticed that a newly published paper citing her work seemed suspicious. Upon closer inspection, the paper turned out to be a rewritten version of an existing preprint. What initially appeared to be a case of plagiarism quickly revealed something more troubling. A deeper investigation uncovered a network of fake articles across multiple preprint servers. These papers were generated by rephrasing legitimate research and populated with references that were either irrelevant or strategically inserted. The goal was not to contribute to knowledge, but to artificially inflate citation counts of certain researchers. In some cases, the cited individuals themselves were unaware of, and even harmed by, this activity, as it distorted their citation profiles and raised questions about their academic integrity.
This example highlights an important shift in the nature of fake citations. It is no longer just a matter of fabricated references appearing in otherwise legitimate papers. Instead, we are beginning to see coordinated efforts, sometimes described as “citation mills”, where entire networks of fake or manipulated papers are created to game academic metrics. In such an environment, citations are no longer reliable signals of influence or credibility but can instead become tools for manipulation.
The problem is more serious than it may initially appear. The consequences of fake citations are no longer confined to academic publishing; they are beginning to surface in high-stakes environments such as the legal system.
In a recent case before the Connecticut Supreme Court (see Leavenworth, 2026), lawyers submitted a legal brief containing citations that were entirely fabricated by generative AI. Some of the quoted phrases did not exist in any real case law, effectively inventing legal precedent to support their argument. The issue was not only the use of AI, but the failure to verify its output. The attorneys later acknowledged that they had relied on AI-generated content without properly checking the accuracy of the citations. In another case (see Davis, 2026), two attorneys in New Orleans were sanctioned after submitting a court filing that included multiple fake case citations generated by AI. One of the attorneys admitted to using AI to speed up legal research and failing to verify whether the cited cases were real. Both attorneys faced financial penalties and ultimately resigned following the incident.
What these examples make clear is that fake citations are not a niche academic problem. They undermine systems that rely fundamentally on the integrity of references, whether that is scientific knowledge or legal precedent. When citations can no longer be taken at face value, the entire chain of trust begins to erode. This calls for more serious action, particularly within academia, to ensure that the scholarly record is not contaminated by fabricated references. In an era of citation mills, where generating convincing but false information is easier than ever, reproducibility should extend beyond results to include verifying references.
References
Davis, K. (2026). 2 government attorneys resign over use of fake AI citations. Retrieved from https://www.abajournal.com/news/article/2-new-orleans-government-attorneys-resign-over-use-of-fake-ai-citations#google_vignette
Leavenworth, J. (2026). CT Supreme Court asked to dismiss case after AI made up fake citations in legal brief. Retrieved from https://www.yahoo.com/news/articles/ct-supreme-court-asked-dismiss-110000009.html
Onwuegbuzie, A. J., Frels, R. K., & Slate, J. R. (2010). Evidence-Based Guidelines for Avoiding the Most Prevalent and Serious APA Error in Journal Article Submissions-The Citation Error. Research in the Schools, 17(2), i–xxiv.
Retraction Watch (2026a). Librarian finds ‘preposterous number’ of fake references in paper from Springer Nature journal. Retrieved from https://retractionwatch.com/2026/03/06/librarian-finds-preposterous-number-of-fake-references-in-paper-from-springer-nature-journal/
Retraction Watch (2026b). A citation alert led researchers to a network of fake articles. But who is benefiting? Retrieved from https://retractionwatch.com/2026/03/30/fake-articles-plagiarism-preprints-arxiv-ssrn-citation-network/
Rivkin, A. (2020). Manuscript referencing errors and their impact on shaping current evidence. American Journal of Pharmaceutical Education, 84(7), ajpe7846.