Academic publishing faces a significant challenge to its integrity as a new study reveals a dramatic increase in "hallucinated" or fabricated citations within scholarly papers. While a recent social media post by Crémieux projected that almost 2% of citations on papers uploaded to the Social Science Research Network (SSRN) could be hallucinated by mid-2025, a more comprehensive analysis published in The Lancet provides concrete figures on this growing issue. The study, conducted by Columbia University researchers, found a sixfold increase in papers containing fabricated references between 2023 and 2025, reaching a rate of 1 in 458 papers in 2025 and escalating further to 1 in 277 papers in the first seven weeks of 2026.
Hallucinated citations are non-existent or misattributed references, often inadvertently generated by artificial intelligence (AI) language models, which can undermine research credibility. Maxim Topaz, a Columbia nurse and health AI researcher who led the Lancet study, noted that this phenomenon reflects potentially deeper issues within the academic publishing ecosystem. Mohammad Hosseini, a professor at Northwestern University, highlighted that this trend signals a shift in citation practices, moving from genuine engagement with literature to a mere "box-checking exercise" facilitated by AI tools.
The rapid proliferation of AI-generated content in academic writing has prompted urgent discussions among researchers and publishers about maintaining scholarly standards. Experts are increasingly concerned that the ease with which AI can produce plausible-sounding but false citations could lead to a propagation of misinformation and waste valuable research time. Some scholars argue that, in certain contexts, including hallucinated citations could even constitute research misconduct, particularly when these citations function as data in review articles or bibliometric studies.
Publishers are actively grappling with strategies to combat this issue. While some, like the Science family of journals, report no issues with fabricated citations in published papers due to automated checks, others like the Public Library of Science (PLOS) have observed "numerous" unverifiable references in submissions. The academic community is calling for enhanced verification processes, greater AI literacy among authors, and the development of more robust tools to detect and prevent the inclusion of these erroneous references, underscoring the critical need to preserve the trustworthiness of academic databases.
