Preprint
Delving into LLM-assisted writing in biomedical publications through excess vocabulary
Dmitry Kobak et al., arXiv, 2024
Read the originalRead on October 17, 2026
- What it is
- An analysis of more than 15 million PubMed abstracts from 2010 to 2024. The authors look for an abrupt rise in certain style words after large language models appeared, and use the excess to estimate how many abstracts were processed with LLMs.
- How much weight it carries
- A preprint, revised through version 5 (July 2025). The later journal version was not checked. The figure is a lower bound from word counts, not a test of individual texts, and it describes scientific abstracts rather than dictation. A search summary reported lower earlier figures, which I did not check.
Facts taken from this source
- An excess vocabulary analysis of PubMed abstracts suggests that at least 13.5 percent of 2024 abstracts were processed with LLMs.
“This excess word analysis suggests that at least 13.5% of 2024 abstracts were processed with LLMs.” (Abstract)