Global retractions due to randomly generated content: Characterization and trends

The aim of the study was to characterize retractions due to randomly generated content. A cross-sectional study was performed, using Retraction Watch database, Journal Citation Reports, Scopus, and journal official websites as data sources. Papers retracted up to 28 May 2024 with reasons related to randomly generated content were included. A total of 3540 retractions were identified. The first retraction was conducted in 2010, and the number of retractions per year escalated from 3 in 2010 to 2302 in 2023. The delay in retraction for papers published between 2020 and 2023 was shorter than that for those published prior to 2020 [2248 (1293, 2687) days vs. 387 (335, 516) days, P < 0.001]. The papers were distributed across the seven primary subject categories classified by the Retraction Watch database. Most retractions fell under the primary category of “business and technology” (2445/69.07%), with technology, computer science, and data science being the most susceptible specific fields. Hindawi published the majority of the retracted papers (68.98%), significantly more than other publishers combined. Of the top 20 journals with the most retractions, 19 were open access journals with a large volume of publications. Six of these 20 journals have ceased submissions and will stop publishing soon, while three have already ceased publication and no longer accept submissions. These retractions spanned 73 countries, with Asian countries making up a significant portion, particularly China (2827 papers, 79.86%) and India (556 papers, 15.71%). These results suggest that the number of retractions due to randomly generated content is on the rise, highlighting an urgent need to establish guidelines for the responsible and transparent use of AI tools and implement disciplinary measures.