AI-Assisted Writing Is Growing Fastest Among Non-English-Speaking and Less Established Scientists

The dominance of English in global science has long created significant barriers for non-native speakers, but the rise of generative AI is reshaping this landscape in profound ways. A groundbreaking study analyzing over two million biomedical publications from 2021 to 2024 reveals that AI-assisted writing is surging, with adoption growing fastest among researchers from non-English-speaking countries and those with less established careers. This research, which employs a distribution-based framework to estimate AI-generated content, highlights how tools like ChatGPT are being adopted unevenly, potentially mitigating long-standing linguistic inequalities while reflecting existing structural disparities in the scientific community. offer a critical look at how technology is transforming academic communication, with for equity, productivity, and the future of global research.

To conduct this analysis, the researchers utilized full-text biomedical papers from PubMed Central, linked to OpenAlex for detailed author metadata, and focused on the Introduction and Discussion sections where narrative writing is most prevalent. They estimated the fraction of AI-generated sentences using a maximum likelihood estimation that models documents as mixtures of human-written and AI-generated content, providing a continuous measure rather than a binary classification. This approach, validated in prior studies with a prediction error below 3.5%, allowed for robust comparisons across linguistic and professional contexts. The study employed difference-in-differences and difference-in-difference-in-differences models to analyze changes post-ChatGPT, controlling for factors like journal, subfield, and time to isolate the effects of language background, author productivity, citation impact, career stage, and institutional prestige on AI adoption.

Show a dramatic post-ChatGPT surge in AI-assisted writing, with publications from non-English-speaking countries experiencing a 400% increase in AI-generated content compared to 183% in English-speaking countries. This disparity is strongly correlated with national English proficiency, as countries with lower English Proficiency Index scores saw the highest adoption rates. For instance, China, with a low EPI rank, saw a 250% growth, while the Netherlands, with high proficiency, had only 60% growth. At the author level, less-established scientists—those with fewer publications, citations, or in early career stages—adopted AI tools more aggressively, as did those from lower-ranked institutions. Corresponding authors, who oversee manuscripts, showed more conservative adoption, whereas first authors, who handle initial drafting, exhibited broader use, especially in non-English-speaking contexts where linguistic barriers are most acute.

These have significant for global scientific equity, suggesting that generative AI could help level the playing field by reducing language-related publication barriers. The study notes a modest positive correlation between increased AI use and productivity, narrowing the gap between English-speaking and non-English-speaking researchers at higher adoption levels. However, this potential benefit is tempered by risks, such as overreliance on AI threatening originality and accountability, and the reinforcement of existing hierarchies where early-career and less-resourced scientists may use AI to meet 'publish or perish' pressures. The research underscores the need for ethical guidelines, transparent disclosure standards, and a cultural shift toward valuing quality over quantity in scientific evaluation to ensure AI integration enhances rather than undermines research integrity.

Despite its large scale, the study has limitations, including its correlational nature, which prevents causal inferences, and its focus on biomedical publications, which may not generalize to other fields. The estimation captures lexical patterns but may miss nuanced AI uses, such as heavily revised drafts or conceptual contributions. Additionally, the analysis relies on author affiliations and may not fully account for multilingual or international collaborations. Future research should incorporate surveys and longitudinal data across disciplines to track evolving adoption patterns and their long-term impacts on research quality and equity. As AI tools become embedded in scientific workflows, understanding these dynamics will be crucial for fostering an inclusive and rigorous global research ecosystem. (Liu et al., 2025, Nature Human Behaviour)