Year: 2026
Type: article
Abstract: Large language models (LLMs) have reshaped scientific writing, yet how this transformation distributes across publishers remains largely unexamined. We analyze over one million academic documents from 49 international publishers using lexical signals. Our results identify an abrupt rise, with a slope break of +10.2% per year after the ChatGPT launch, with a simultaneous rise across all 12 examined fields within one-quarter. We also note that publisher leniency is large and independent of field, open access status, author country, and team size, with odds ratios ranging from 0.28 (American Physical Society) to 4.25 (Emerald), and this variation is widening. This effect is concentrated in mega-journals, correlates with lower author English proficiency, and confers a persistent citation advantage. In addition, we conduct a within-author fixed-effects design (e.g., the same researcher publishing at different outlets) and report that publishers seem to mediate the textual footprint of LLM-assisted writing (not merely select authors who use it).
Author: M.Z. Naser
Institution: Clemson University
Language: English
Subfield: Statistics, Probability and Uncertainty
Field: Decision Sciences
Domain: Social Sciences
Open Access status: hybrid
Estimated APC paid: $2,000
Funder: National Science Foundation
OpenAlex ID: W7204457875