Field diffusion · General science · Social science

OpenAlex study maps AI adoption across 227M scholarly works

A large-scale bibliometric measurement of AI adoption, topic concentration, visibility, geography, and retractions across science.

Summary

Castro Torres, Giner-Miguelez, and Crosas analyze AI adoption across OpenAlex works and report exponential growth after 2015. They also find concentration in limited topics, close ties to computer science and conventional statistical frameworks, citation premiums, higher retraction rates, and uneven global adoption patterns.

AI role

AI methods were classified as research methods in scholarly works, including through an AI-assisted semantic pipeline validated against human coding.

Narrative role

This broadens diffusion evidence beyond one tool or survey by measuring where AI is actually used as a research method and what warning signs accompany rapid uptake.

Caveat

The work is preprint-stage and depends on semantic classification of publications, so measured adoption and retraction associations may reflect classification and field-composition effects.

Event

Date: 2026-05-07
Timeline role: Notable
Progress relevance: Caution signal
Validation: Downstream measured

Evidence

Preprint · Bibliometric analysis

Observed metrics

scholarly works analyzed: 227 million; The paper analyzes OpenAlex works from 1960 through 2024.
fields analyzed: 46; The analysis spans four scientific domains and 46 fields.
human-coded abstracts: 911; The AI-adoption classifier was validated against human coding of 911 abstracts.
full-text robustness sample: 348,000; The authors report a robustness check on PLOS One full-text articles.

Source: arXiv ↗