Study Finds Over 70% of English Biomedical Papers Use AI-Assisted Writing
A research team led by Lena Holzwarth analyzed over 1.19 million papers on PubMed Central and found that the proportion of English biomedical papers showing signs of AI-assisted writing rose to approximately 77% in 2025, up from 19% in 2023. The study detected AI usage by tracking characteristic high-frequency vocabulary like "delves" and "invaluable." This rapid adoption highlights how generative AI has become deeply integrated into scientific publishing, particularly helping non-native English speakers bridge language barriers. However, it also raises concerns about AI-generated hallucinations, fake references, and the potential erosion of public trust in scientific research. The study revealed that AI usage was highest in the "Discussion" section of papers (78%) and exceeded 80% in papers from non-native English-speaking regions like China and South Korea. Rather than banning AI, the researchers advocate for establishing clear regulatory frameworks for disclosure and fact-checking.
## BACKGROUND
PubMed Central (PMC) is a free digital repository archiving open-access full-text scholarly articles in biomedical and life sciences, managed by the U.S. National Center for Biotechnology Information (NCBI). As large language models (LLMs) like ChatGPT became widespread, researchers noticed a surge in specific transition words and adjectives (like "delves" or "testament") that are statistically overrepresented in AI-generated text compared to human writing.