← News

Models & toolsSources: arXiv · 18 Nov 2026

Study of 2.5 million papers finds little sign LLMs have changed science yet

Researchers tracking a decade of scientific writing report that most trends predate late 2022 and continued without major breaks after LLMs arrived.

In plain words

Large language models are AI systems that generate text. A research team examined 2.5 million full papers from 2015 to 2025 to see whether such models changed how science is written and organized. They found that, apart from shifts in some word frequencies, earlier trends mostly continued with no clear break.

What was announced

The authors analyzed full texts across five disciplines, tracing changes in lexical, structural, epistemological, and social dimensions of scientific practice. Their conclusion is that most pre-2023 trends continued with limited detectable breaks after LLMs emerged in late 2022, though they note the analysis is meant as a baseline for monitoring science as adoption deepens rather than a final verdict.

Some exceptions stand out. Growth in computer science coauthor teams accelerated significantly in Asia, and apparent declines in biomedical topic diversity largely reflected standardization in article formats rather than a real narrowing of research questions. The team presents the work as an empirical foundation for tracking the health of global science over time.

What it does not show

  • The evidence is an abstract, so methods, discipline list, and full findings are not available for verification.
  • The study covers 2015 to 2025 and five disciplines, so it may not capture effects outside that window or those fields.
  • The authors report limited detectable breaks, which is not the same as proof that LLMs have had no influence on science.

Sources

Source-support review completed 19 Nov 2026. This story was drafted with AI assistance and checked against retained source passages. Automated checking is not independent verification.

Coverage: arXiv: abstract.

Updates and corrections

No corrections so far.

First published 19 Nov 2026

New to the subject? AI from the Ground Up starts from the beginning.