A research team analyzed over 1.19 million English biomedical papers on the PubMed Central platform and found that generative AI is rapidly infiltrating academic writing. By 2025, about 77% of papers showed clear signs of AI-assisted writing or editing, a significant increase from 19% in 2023 and 52% in 2024.

Significant Differences by Region and Section

Looking at sections, the discussion section showed the highest signs of AI assistance (about 78%), followed by the abstract, introduction, results, and methods, with 68%, 63%, 58%, and 54% respectively. In terms of regions, non-English speaking researchers from countries such as South Korea and China used AI at a rate exceeding 80%, significantly higher than English-speaking countries like the UK and the US. Researchers believe this is due to AI helping improve grammar and expression.

Risks are also emerging: large models may fabricate false facts and non-existent references, often presenting errors with confidence. Additionally, some authors fail to disclose their use of AI, further eroding scientific credibility. Researchers advocate for establishing guidelines rather than banning AI. They propose clear requirements for fact-checking, reference authenticity, and AI disclosure. When AI assistance has become common, defining reasonable boundaries is more critical than simply banning it.