17% of biology preprints now say they used an AI chatbot, nearly as many as use tools like AlphaFold
Maths papers show a bigger jump, from 4% to 48% in six months. Both count only what authors choose to disclose.
Epoch AI counted how often research preprints acknowledge using AI. A preprint is a paper posted publicly before peer review. In biology (the bioRxiv archive), 17% of September 2026 preprints acknowledged using a large language model (LLM, the kind of AI behind chatbots). That is close to the 19% that acknowledged specialised AI tools such as AlphaFold, which predicts protein structures and appears in about half of those. Specialised tools have climbed steadily since 2021, while LLM acknowledgments grew from about 0.3% in January 2023.
Most acknowledged LLM use is not research itself. In Epoch’s downloadable September data for bioRxiv, by our count, writing appears in about three quarters of the acknowledgments and code in about half. About 4% of all biology preprints acknowledged LLM use for research or analysis, up from about 1% at the start of 2026. One acknowledgment can name several uses.
In mathematics (the arXiv archive), Epoch’s newsletter of 11 November reports acknowledgments of AI use growing from about 4% in April 2026 to 48% by October. Epoch’s downloadable data gives 160 of 4,089 preprints in April and 4,065 of 8,515 in October.
What it means
Authors in two fields now say, in print, that they use AI, and in maths about half of new preprints do. Biology shows how that use compares with purpose-built science tools, though most of it is help with writing and code. The numbers do not say how much AI contributed to the science.
Caveats
- An acknowledgment is a disclosure authors chose to make. Epoch says its figures may understate real use, and norms for disclosing it differ by field and change over time, so a rise can partly be more people disclosing.
- Biology and maths come from different archives with their own validation steps, so compare each with its own past, not with the other.
- An AI model (GPT-5.6 Sol) did the reading. A human checked a sample of 238 decisions and agreed with 226. Epoch’s list of 150 specialised tools may be incomplete.
- Preprints are not a typical sample of papers. Epoch estimates about 30% of biology papers in top journals first appear on bioRxiv, against over 95% in maths on arXiv.