A striking new study has revealed that the integration of artificial intelligence into academic writing is far more widespread than previously anticipated. According to research posted on the arXiv preprint server on 12 August, an astonishing 90% of biomedical papers published in December 2025 within a major database displayed clear linguistic markers of AI assistance.
The Exponential Rise of LLMs in Scholarly Research
Authored by researchers including Lena Holzwarth, Rita González-Márquez, and Dmitry Kobak, the study evaluated open-access papers archived in the PubMed Central repository. The findings indicate that the overall usage rate of large language models for papers published throughout 2025 stood at 77%, a dramatic leap from 52% recorded in 2024. These figures dwarf earlier estimates, such as a 2025 analysis that placed the figure at a mere 13.5% by focusing solely on abstracts.
Where Scientists Rely Most on Artificial Intelligence
The research team utilized a novel, unbiased methodology based on shifting word frequencies to detect alterations across full texts. The analysis demonstrated that AI tools are heavily favored during specific sections of a manuscript. Specifically, scientists are twice as likely to employ LLMs when drafting a paragraph in the Discussion section, hitting 68%, compared to 32% in the Methods section. Nevertheless, the authors noted that even within the Methods section, overall AI prevalence surpassed 50%.
Balancing Academic Integrity with Language Accessibility
Since the launch of OpenAI’s ChatGPT in November 2022, chatbots have transformed modern workplaces, including academic institutions. While proponents argue that AI-assisted writing bridges language gaps for non-native English speakers and democratizes publishing, critics warn of rising risks involving academic misconduct, hallucinations, and scientific fraud. ‘To inform policy decisions, it is necessary to monitor the prevalence of LLM-altered texts in scholarly publications,’ the researchers emphasized, urging academic bodies to establish robust new guidelines.
