RESEARCH · GLOBAL
Occupational AI-exposure scores partly reflect platform user base, not workforce distribution
arXiv cross-disciplinary study: AI-exposure metrics derived from ChatGPT logs overstate true occupational AI risk because they measure platform adoption patterns, not workforce prevalence. Reweighting to BLS data attenuates estimates by 42–93%.
WHY IT MATTERS
Popular AI-exposure studies used by policymakers and researchers overstate risk. BFSI staffing analyses claiming 30%+ job displacement may be inflating the number; true exposure is likely 15–20% lower.
Source: arXiv · 2026-05-22