89% of Biomedical Papers Show LLM-Assisted Writing by End of 2025

A recent study available on arXiv (2608.10715) indicates that by the conclusion of 2025, 89% of open-access biomedical articles in PubMed Central will exhibit characteristics of writing assisted by large language models (LLMs). The researchers introduced an innovative technique to assess LLM utilization through variations in word frequencies, which they confirmed to be impartial. Their findings reveal that LLMs are utilized more frequently in Discussion sections (68%) than in Methods sections (32%). This research underscores the extensive integration of LLM tools in academic writing, prompting concerns regarding potential misconduct and fraud, and seeks to guide policy-making. The method provides a dependable approach to track the occurrence of LLM-modified texts in academic literature.