“This excess word analysis suggests that at least 13.5% of 2024 abstracts were processed with LLMs.”
The team counted style words across more than 15 million PubMed abstracts from 2010 to 2024, and the jump after chatbots arrived beat the vocabulary shift from Covid. At least 13.5% of 2024 abstracts went through an LLM, and some subcorpora hit 40%. That is a floor, because it only counts the authors who left the tell-tale words in. Peer review is now reading machine prose about human experiments.