The evolution of ChatGPT-generated text in scientific medical publications

Background: ChatGPT is an Artificial intelligence (AI) language model that enhances readability in medical writing. However, the prevalence of ChatGPT-generated articles in English medical literature has not been systematically studied. This research aims to evaluate the frequency of these manuscripts, their adoption across medical subspecialties, and the countries of origin for the publishing companies. Methods: The term “delve” is primarily linked to ChatGPT-generated medical literature. In this observational study, we identified manuscripts as ChatGPT-generated based on this term in the title or abstract. We examined PubMed ® from December 01, 2022, to May 14, 2024, and applied descriptive statistics as needed. Results: During the study period, 2.193.871 manuscripts were published in Pubmed ® . The term “delve” was found in the title and abstract of 1996 scientific manuscripts, suggesting that at least 0.09% of all publications were ChatGPT-generated. At the same time, PubMed ® hosted 982 journals. We report on those that published at least ten or more CHATGPT-derived manuscripts (top 25). Publishers with the highest number of ChatGPT-generated manuscripts were located in Switzerland (564; 27.6%), followed by the United States (552; 27.0%) and the United Kingdom (430; 21.0%). The Multidisciplinary Digital Publishing Institute (MDPI: 373; 20.3%) was the top publisher of ChatGPT-driven publications, followed by Elsevier (340; 18.5%) and Frontiers (160; 8.7%). Conclusion: Since the introduction of ChatGPT, medical publications are increasingly ChatGPT-supported or generated with an uncertain amount of author auditing and editing. High-frequency publishers were located in first-world countries.