Current results
AI screening and REFORMS item-level assessment are under way. Human validation will be used to measure the AI reviewer's accuracy and place uncertainty bounds around later results.
Read the preliminary results, including what counted as ML-based science, what reporting was usually missing, and the first pre/post comparison.
The Crossref density scan is complete. Under the conservative title-or-abstract query, explicit ML language rose from 1.568% of journal papers in 2020 to 4.852% in 2025.
Explore the ML density atlas for journal distributions, growth rankings, high-density venues, high-volume contributors, abstract coverage, and the exact scan method.
Open the manual full-text queue for papers that still need a verified PDF.