AI is writing more in science, is this a good thing?

The number of scientific publications written by artificial intelligence (AI) is increasing – and not necessarily leading to better quality, according to researchers.  

ChatGPT was released for public use in November 2022, and the growing prevalence of AI has transformed scientific research. As a part of the AI Task Force for Organisation Science, a research team led by Claudine Gartenberg at the University of Pennsylvania aims to discover how AI is changing scientific publication. 

Their recent paper analysed trends in the submission and review of scientific papers at Organisation Science since the release of ChatGPT. Close to 7000 submissions and more than 10,000 reviews were considered for this research.  

The trend is striking. Since the release of ChatGPT, submission volume has increased by more than 40%, and papers with higher AI usage are on an upward trend.  

Number of monthly submissions of scientific papers and their AI use [Gartengery et al. (2026); DOI https://doi.org/10.1287/orsc.2026.ed.v37.n3.]  

A greater number of scientific papers is not inherently problematic. More publications mean more knowledge, and AI is certainly useful in accelerating research processes, which is the likely reason for this overall trend. Even before the release of ChatGPT, overall publication numbers were increasing, mostly because that would provide greater financial bonuses. 

What underlies this trend, however, is worrying. Alongside overall quantity, the team measured the writing quality of submissions, using a range of calculation measures such as the Flesch Reading Ease formula. The calculations were made based on readability and style. The team found the score for submissions between 2013 to 2022 did not show meaningful changes, whereas the post-ChatGPT years showed significant reduction. The drop in writing quality scores and AI usage also correlated, suggesting AI is contributing to poorer quality papers.  

A similar result was found in the reviews. In addition to having poorer writing quality, AI-generated reviews also had narrower analysis and emphasis on theory, rather than data. 

These changes are likely occurring because AI is used to ramp up the number of publications, rather than to improve rigorous research methods. As the authors point out in the paper – “When institutional incentives push “pubs” over rigorous research, AI will be used to drive volume rather than to fulfill its potential as critical research technology.”.  

With only 4 years passing since the introduction of ChatGPT, it is difficult to make a definite statement on whether AI is good for scientific research. However, this study makes a few things very clear. The current way we use AI is not improving the quality of scientific publications. To ensure that its effect does not remain detrimental, change in practice and AI use must be made at an institutional level. 

Written by Kazuma Oura, 4 July, 2026.

Edited by Eloise Trawick.