Answers are grounded in this paper and cite their sources.
The rapid adoption of AI in academic writing, particularly with tools like ChatGPT, has raised significant concerns regarding the accuracy of generated content. This study explores the phenomenon of “hallucitation” in scientific writing, where AI models fabricate citations, analyzing responses from ChatGPT versions 3.5 and 4o in the context of librarianship. Through an experimental design, scientific content with citations was generated and systematically verified using Google Scholar and the publisher’s website. The findings reveal a disturbingly high frequency of false or non-existent citations—42.9% in ChatGPT-3.5 and 51.8% in ChatGPT-4o. Despite slight improvements in citation accuracy from version 3.5 to 4o, with accuracy rates of 3.92% and 6.35%, respectively, both versions exhibit significant limitations. Notably, ChatGPT 3.5 frequently generated completely fabricated sources, while ChatGPT-4o introduced subtle errors, such as mismatched journals. The study indicates no significant difference in accuracy between the two versions, underscoring the persistent risks associated with AI-generated citations. These findings highlight the urgent need for rigorous verification of AI-generated content to safeguard the integrity of scholarly work.
No takes yet. Share an insight, caveat, or question.
Oladokun et al. (2025) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: