HT Nik John for sending me this fascinating paper by the editors of Organisational Science. If you’re interested in AI and scholarly publishing, read this paper: https://pubsonline.informs.org/doi/10.1287/orsc.2026.ed.v37.n3
The highlight for me was the finding about editorial outcomes:
A clear pattern emerges above 30% AI use. Manuscripts crossing this threshold have desk rejection rates nearly 30 percentage points higher than those below it, a 60%–80% relative increase. By the end of the editorial funnel, only 3.2% of high-AI papers (70%+ AI) receive Revise & Resubmit decisions, compared with over three times that rate among low-AI submissions. In other words, editors, likely without recognizing the writing as AI-generated, consistently judged these manuscripts as lower quality and unworthy of reviewers’ time. Interestingly, submissions in the 15%–30% range show a slight increase in success rates, possibly reflecting a more human-first collaboration mode or differences in authors, topics, and other factors that we cannot disentangle.
This to me feels absolutely key. If this is established across multiple domains it becomes the foundation for a renewed normativity about LLM use. The incentives for individual acceleration are real, the rewards for it are not. Going faster with LLMs doesn’t produce the outcomes you’re actually looking for. It also generates a whole range of externalities that make your working life less pleasant, as well as everyone else’s.
Mark Carrigan 