A research team analyzed 1,194,287 biomedical articles from PubMed Central published between 2017 and 2025. By the end of 2025, 89% of open-access biomedical papers showed signs of LLM-assisted writing or editing. This figure significantly exceeds previous estimates of 12–16% for 2024.
The new statistical method tracks the frequency of 379 specific marker words that surged after late 2022. Researchers compared actual usage against linear regression trends from 2018–2022. The validation process, using 100,000 simulated documents annually, confirmed an error margin of no more than 2 percentage points.
The study highlights significant regional variations in LLM adoption. South Korea leads with 85%, followed by China at 82% and Taiwan at 80%. In contrast, the UK shows the lowest usage at 28%. Aggregated data reveals a 37% adoption rate for English-speaking countries versus 72% for the rest of the world.
Section-specific analysis shows that discussion sections and abstracts are most frequently affected by AI, reaching 68% and 67% respectively by December 2025. Methods sections are less impacted at 32%, though full-text analysis of these sections shows a 54% usage rate, suggesting authors prefer AI for interpretation over procedural descriptions.
