Scientists are publishing more research papers than ever, helped by artificial intelligence tools like ChatGPT. But many of these papers fail to meet journal standards, a new study warns.
Researchers at Cornell University analysed how large language models are reshaping academic publishing. Their findings show a sharp rise in paper production, especially among non-native English speakers.
A shift in scientific writing
What once seemed like academic gossip is now a measurable trend across many scientific fields. The study found AI use growing in physical, computer, biological and social sciences.
Yian Yin, the study’s lead author, said the change affects the entire research ecosystem. He warned funders and policymakers must look closely at how science is evaluated and supported.
How the study worked
The research examined more than two million papers posted between 2018 and 2024. These papers appeared on three major preprint servers before formal peer review.
Researchers trained an AI system to detect text likely written by language models. They compared papers from before 2023 with those showing clear AI assistance.
They then tracked productivity changes and later journal acceptance rates.
Productivity jumps sharply
Scientists using AI tools posted far more papers than their peers. In physics and computer science, output rose by about one third.
In biology and social sciences, production increased by more than 50 percent. The biggest gains appeared among researchers whose first language is not English.
Some Asian institutions saw publication growth reaching nearly 90 percent.
Quality raises concerns
Despite higher output, AI-assisted papers were less likely to pass peer review. Human-written papers with complex writing were accepted more often.
Reviewers appeared unconvinced by polished language without strong scientific value. Researchers say new rules may be needed to guide responsible AI use.
The key question, Yin said, is how AI is used, not whether it is used.
