AI is straining peer review

Organization Science's AI Task Force analyzed 6,957 initial submissions and 10,389 text-entry reviews between January 2021 and February 2026.

• Submissions increased 42% after ChatGPT, and by February 2026 a majority were estimated to contain some degree of AI writing.
• Schools more sensitive to UT-Dallas ranking increased their share of high-AI submissions more after ChatGPT.
• The higher the AI share, the harder the writing became to read. Flesch Reading Ease in January 2026 was 1.28 standard deviations lower than in 2021, and the correlation between AI score and readability was rho=-0.4 (p<=0.001).
• Jargon and nominalization increased, and manuscripts with higher AI share had more initial rejections and more rejections after full review.

The review stage showed similar patterns. AI use was detected in 30%+ of text-entry reviews, and reviews with a higher AI share were more abstract, less data-driven, and less associated with editorial decisions.

To respond, the editorial office increased deputy editors from 6 to 11 and active senior editors from about 30 to about 60. Published papers still remained mostly human-written, but authors held that while AI can be used for drafting, coding, and organizing, the process of writing is itself the process of thinking, so large-scale ghostwriting worsens both research quality and review burden at the same time. The core cause was seen as less AI itself and more the publish-or-perish incentive.