Analysis code, results and derived data for: How much of a generative-AI prevalence estimate is the word list? Seven word lists compared on Japanese society journals and on PubMed

Abstract

Code, result files, derived integer-count tables and figures for a comparison of seven word lists used to estimate generative-AI-assisted writing, applied under identical rules to the English abstracts of 25 Japanese society journals and to PubMed (a random sample of 20,000 abstracts per year, 2010-2025, and the PubMed word counts released by Kobak et al., 2025). Includes the harvest scripts, the comparison word groups of both corpora, the PubMed comparison protocol as fixed before PubMed data from 2022 onward were analyzed, and a clean-room script (code/reproduce.py) that recomputes the reported quantities from the derived tables. No abstract text is redistributed. Version 1.8.0 accompanies the revision: two corrections to the construction of the submitted analysis, the PubMed comparison, and the revision's table and figure generators. Code under the MIT License; derived tables under CC0 (derived/DATA_LICENSE.txt). The deposit is uploaded as four zip files; extracting all four into the same folder gives the complete jstage_wordlists_v1.8.0 folder. Funded by Waseda University Grant for Special Research Projects (Tokutei Kadai), an internal scheme that assigns no external grant number, and MEXT Supporting Pioneering Research through AI for 1,000 Discovery challenges Program (SPReAD) Japan Grant Number JPMXP1726306059.