Ai-assisted Writing Surges 400% Among Non-English and Less Established Scientists, Reducing Drafting Costs

The increasing dominance of English in scientific communication presents a persistent challenge for researchers worldwide, and recent advances in generative artificial intelligence offer a potential, yet complex, solution. Jialin Liu from the University of Wisconsin-Madison, Yongyuan He, and Zhihan Zheng from Peking University, along with Yi Bu and Chaoqun Ni, investigate how this technology is reshaping scientific writing practices. Their study presents the first large-scale analysis of over two million biomedical publications, revealing a substantial surge in AI-assisted writing following the emergence of ChatGPT. Importantly, the team demonstrates that adoption of these tools is growing at a significantly faster rate in non-English-speaking countries and among scientists with fewer established publications, suggesting that AI may be helping to level the playing field and address long-standing linguistic inequalities in global science.

LLMs Transform Scientific Writing and Output

This study investigates the impact of Large Language Models (LLMs), such as ChatGPT, on scientific publishing, employing quantitative analysis of extensive datasets of scientific papers to understand evolving publication patterns. Researchers conducted a comprehensive analysis, examining authorship trends, publication rates, and potential shifts in research output, focusing on how and to what extent LLMs are being integrated into scientific writing. Investigations also explore whether LLMs are increasing or decreasing overall publication numbers, and if these effects differ based on author type. The study considers how LLM usage varies based on author characteristics, including experience, country of origin, and English proficiency, addressing concerns about the quality and originality of research when LLMs are used.

The analysis utilizes large datasets from sources including PubMed Central, Web of Science, Scopus, and OpenAlex, spanning from 2021 to 2024, allowing for comparisons before and after the release of ChatGPT. Results confirm widespread and increasing LLM usage in scientific writing, with some indications of changes in publication rates that vary across author groups. Corresponding authors appear more likely to utilize LLMs, and authors with lower English proficiency may rely on these tools more heavily. Researchers emphasize that correlation does not equal causation, noting the difficulty of proving that LLM usage directly causes changes in publication patterns, and acknowledge the limitations of current AI detection methods. In summary, this rigorous and comprehensive study provides valuable insights into how LLMs are being used in scientific publishing and the potential consequences of this integration, representing a timely and important contribution to the ongoing discussion about the future of scientific research.

AI Involvement in Biomedical Publications Measured

This approach avoids binary classifications and instead provides a nuanced assessment of the degree to which AI tools contributed to a given publication. Researchers employed a difference-in-differences analysis to compare changes in AI-generated content between publications originating from English-speaking and non-English-speaking countries, controlling for other relevant variables. Further analysis involved a finer-grained, country-level examination, calculating the change in the average fraction of AI-generated content between 2021-2022 and 2023-2024 cohorts for each country. This revealed a strong negative correlation between a country’s increase in AI content and its English Proficiency Index. Stratifying publications according to productivity, citation impact, career stage, and institutional prestige, researchers used advanced statistical models to identify specific author profiles most likely to adopt AI-assisted writing tools, revealing that less-established scientists at lower-ranked institutions demonstrated the highest rates of adoption. Results demonstrate a significant increase in AI-generated content in papers from non-English-speaking countries, approximately four times the increase observed in English-speaking countries.

AI Writing Surges in Global Science Publications

This research demonstrates a significant surge in the adoption of AI-assisted writing tools within the scientific community, particularly following the release of ChatGPT. Analyzing over two million biomedical publications from PubMed Central between 2021 and 2024, scientists measured a substantial increase in the fraction of AI-generated sentences, revealing a clear post-ChatGPT trend. Prior to 2022, AI-writing assistance remained relatively stable across both English- and non-English-speaking countries. However, after ChatGPT’s release, publications from non-English-speaking countries experienced a dramatic escalation in AI-generated content, increasing from 0.

04 to 0. 20 of total content, while English-speaking countries saw an increase from 0. 06 to 0. 17. Statistical analysis, employing a difference-in-differences approach, confirmed this observation, revealing that the fraction of AI-generated sentences in publications from non-English-speaking countries increased by an average of 0.

02 units (p For example, publications from China exhibited approximately 250% growth in AI-generated content, while those from the Netherlands showed approximately 60% growth. Further investigation into author profiles revealed that papers by non-top scientists and junior researchers, serving as corresponding authors, exhibited significantly greater increases in AI-assisted writing compared to their top or senior counterparts, suggesting that established scientists adopt these tools more conservatively in supervisory roles. Importantly, author profile can partially mitigate disparities associated with language background. These findings demonstrate that generative AI is being adopted unevenly, reflecting existing structural disparities while offering a potential opportunity to mitigate long-standing linguistic inequalities.

AI Writing Surges Beyond English-Speaking Science

This research demonstrates a significant surge in the adoption of generative artificial intelligence tools within the biomedical research community, particularly following the release of ChatGPT. Analysis of over two million publications reveals that AI-assisted writing is growing at a substantially faster rate in non-English-speaking countries, approximately four times the rate observed in English-speaking nations. This increased uptake is especially prominent among.