Data sources & attribution

The index uses documented public data, coverage-dependent empirical research, and clearly labeled curated role models. Here is what each source contributes.

O*NET 29.1 database

CC BY 4.0
U.S. Department of Labor / O*NET Resource Center · source

The backbone for O*NET-linked roles: occupations, tasks, skills, work activities, work context, job zones, and related occupations. Curated roles without a matched SOC are labeled separately.

Anthropic Economic Index

Open dataset
Anthropic · source

Observed Claude usage by occupation. For roles with a matched signal, release 2026-06-26 nudges whether observed use leans toward automation or augmentation. Usage volume is a reliability weight, never exposure, and Claude.ai usage is a proxy, not a census.

“GPTs are GPTs”

Published research
OpenAI / University of Pennsylvania (Eloundou et al.) · source

Task-level exposure modeling. Where a role matches, Core tasks receive greater weight and the occupation aggregate helps set the automation/exposure magnitude.

AI Occupational Exposure (AIOE) - Language Modeling

Published research (cited, not redistributed)
Felten, Raj & Seamans (2023) · source

A second, independent exposure measure built from O*NET abilities mapped to AI capabilities. The Language-Modeling variant is blended for matched 6-digit SOC roles, so an ability-based method can corroborate the task-based Penn measure. It is standardized exposure, not a prediction of job loss.

BLS Employment Projections

Public domain (U.S. government)
U.S. Bureau of Labor Statistics (via ProjectionsCentral) · source

Separate U.S. labor-market context: wages, employment, projected growth or decline, and average annual openings (2024–34, including growth and replacement needs) where published. These figures are not components of the AI-exposure score.

Corroborating research - context, not score inputs

These shape how we frame AI exposure, but are not blended into your score. Exposure measures disagree in magnitude, and none of them equals observed job loss.

AI is showing up in hiring
Share of US job postings that mention AI: 6.3% (2026-07), up from 1.7% in 2019.
Source: Indeed Hiring Lab AI Tracker (CC BY 4.0). A macro hiring trend - not a per-occupation signal and not part of your score.
  • Stanford "Canaries in the Coal Mine" (Brynjolfsson, Chandar & Chen, 2025) - observed early-career strain: a 16% relative employment decline for workers ages 22-25 in the most AI-exposed U.S. occupations, after firm-level controls, relative to workers in less-exposed fields and more experienced workers in the same occupations. Context only, not a score input or a personal forecast.
  • ILO Working Paper 140 - a global generative-AI occupational-exposure index (ISCO-08; CC BY 4.0).
  • OECD AI Exposure Measure (2026) - a forward-looking, occupation-level AI capability-gap measure.
  • Yale Budget Lab - AI & the labor market - finds the published exposure metrics disagree in magnitude and do not equal job loss.
  • Stanford HAI AI Index - the broad annual benchmark for AI's economic context.

O*NET data is used under the Creative Commons Attribution 4.0 license. Government datasets (BLS, O*NET) are used in accordance with their terms; research is cited, not redistributed. AI-Safe Careers is not affiliated with or endorsed by these organizations.

Important: This is an estimate of AI exposure, not a prediction that your job will disappear. It is designed to help you understand how your role may change and improve your career resilience.

← How the score is calculated