17% of biology preprints acknowledge chatbot-style AI use, Epoch AI says

Epoch AI used AI models to read every biology preprint on bioRxiv from January 2020 to September 2026 for statements about AI use. It says voluntary acknowledgements mean its figures may underestimate real use.

Share
17% of biology preprints acknowledge chatbot-style AI use, Epoch AI says
Illustration: AI-generated

Biology researchers acknowledged using a large language model (LLM), a chatbot-style AI, in 17% of the preprints Epoch AI examined for September 2026, in its analysis. A preprint is a research paper shared publicly before it has been through peer review.

Epoch used an AI model, GPT-5.6 Sol, to check every preprint on the preprint site bioRxiv from January 2020 to September 2026 for statements that the authors used AI. It counted LLM use and use of specialised AI tools separately.

Specialised tools, in Epoch's count, include models for protein structure and design, image analysis, genomics (the study of genes) and single-cell sequencing (reading the genes of individual cells). The most commonly acknowledged is AlphaFold, which predicts protein structure. It became publicly available in July 2021, and Epoch says it appears in about half of the preprints that acknowledge specialised tools.

In September 2026, 846 of the 4,922 preprints Epoch examined acknowledged LLM use for any purpose. In January 2023, the figure was 9 of 2,987 (0.30%). The September share is close to the 19% that acknowledged a specialised tool.

Epoch measures voluntary acknowledgements, which reflect both real AI use and how willing researchers are to disclose it. For that reason, it says, its figures may underestimate true AI use.

LLM use is more varied than tool use, Epoch says. It includes literature review, writing and coding help. A second AI system, the OpenAI Decisions API based on GPT-6 Luna, sorted each LLM acknowledgement into one of four uses: research, code and computation, writing, or literature review.

Epoch defines research use as ideas such as experimental design, methods work such as labelling data, and analysis such as statistics. About 4% of biology preprints in September 2026 acknowledged LLM use for research or analysis, up from about 1% at the start of that year.

Epoch names three more limits. Preprints may not be representative of peer-reviewed papers. AI models did both the detecting and the sorting, so both can contain errors. Its list of 150 specialised tools came from sampling recent bioRxiv preprints, and it may not include every tool used in biology research.

Epoch publishes the monthly data as a downloadable spreadsheet file (CSV), updated on 21 October 2026. It is free to use and reproduce under a Creative Commons BY licence, which lets anyone reuse it if Epoch and its authors are credited.

Sources