The rise of generative artificial intelligence has led to a significant increase in the volume of scientific papers submitted for publication, raising alarms about the quality of research materials. A recent analysis by the editorial team at Organization Science examined 6,957 submissions and 10,389 reviews from January 2021 to February 2026. The introduction of ChatGPT in late 2022 marked a turning point, coinciding with a 42% increase in submissions and a notable decline in the quality of the articles.
By February 2026, it was evident that artificial intelligence had been utilized to some extent in the majority of the articles submitted. Researchers employed the Pangram 3.1 algorithm to gauge the probability of machine-generated text in these submissions. The average readability score, according to the Flesch index, dropped significantly, reflecting a decline of 1.28 standard deviations from January 2021 levels.
Notably, submissions where AI contributed more than 70% were rarely accepted, with a rejection rate of 70% prior to reaching peer reviewers. In contrast, articles showing minimal AI involvement faced rejection only 43.7% of the time. The likelihood of receiving a request for revisions also decreased dramatically from 11.9% to 3.2% when significant AI usage was detected.
The findings indicate that even well-structured and stylistically polished works faced higher rejection rates if AI was heavily involved, raising concerns about the scientific rigor of these submissions. AI-generated reviews were found to lack diversity, often focusing more on theoretical frameworks while neglecting data, methodology, and experimental results. The researchers caution that while current algorithms can't reliably identify the influence of AI, they can highlight prevailing trends.
Despite these challenges, some scientists argue that AI does not inherently harm scientific progress. A large-scale experiment led by James Evans encouraged researchers to evaluate ideas generated by language models, resulting in over 25,000 assessments by 6,749 scientists. The study revealed that while popular models often produced similar concepts, more advanced models could indeed surprise users with innovative ideas.
AI reviewers frequently disagreed with human experts, with a specialized model, Qwen3-14B, outperforming general models by up to 27%. The researchers concluded that while AI can expedite data retrieval, programming, and writing, reliance on it without human oversight and a revised evaluation system could lead to an unregulated increase in publications rather than a genuine expansion of knowledge.
This surge in AI-generated content poses significant implications for the research community and competitors, emphasizing the need for stringent quality control measures to ensure the integrity of scientific literature.
Informational material. 18+.
