AI has flooded scientific journals with articles — and so far there is little good about it

The editorial team of the journal Organization Science analyzed 6,957 submissions and 10,389 reviews received from January 2021 to February 2026. The appearance of ChatGPT in late 2022 coincided with a sharp increase in submissions and a decline in quality. After ChatGPT's spread, submissions rose by 42%, and by February 2026 AI was likely used in some form in most submitted articles. Using the Pangram 3.1 algorithm to detect machine text, researchers found that the average readability (Flesch index) dropped by 1.28 standard deviations by January 2026 compared to January 2021. Papers with an AI participation index above 70% were rejected in 70% of cases before peer review, while those with minimal AI usage were rejected only 43.7% of the time. The chance of receiving a revision request dropped from 11.9% to 3.2% for papers with heavy AI use. Statistical models showed that the higher rejection rate persisted even for stylistically polished texts, indicating poor scientific content. AI-generated reviews were less diverse, focusing more on theoretical frameworks and less on data, methods, and results. The authors note that algorithms cannot reliably identify AI influence, so only general trends are discussed. However, a large experiment by James Evans had researchers evaluate ideas generated by language models: 6,749 scientists gave over 25,000 ratings on novelty, practical value, and plausibility. Popular models often suggested similar ideas, while more advanced models could surprise with novelty. AI reviewers often disagreed with human experts. The Qwen3-14B model, specifically trained on scientist-written reviews, outperformed general-purpose models by up to 27%. The researchers conclude that AI already helps with data search, programming, and text preparation, but without human verification and changes in the evaluation system, it may lead to poorly controlled growth in publications rather than an increase in new knowledge.