What it measures
- Occupation exposure (6-digit SOC), industry exposure (4-digit NAICS) and geographic exposure (state and county)
- Language-modeling and image-generation variants, each kept as its own measure
What it does not measure
Most misreadings of labor data come from asking a dataset a question it was not built to answer. These are the limits worth holding onto.
- Observed use of AI
- Employment outcomes
Update cadence
Static research release; checked annually.
Occupation pages name the specific release behind each figure and when it was imported, so you can see how current a number is rather than assuming it is new. More on that on about the data.
How JobMarketHealth uses it
- Exposure context on the AI & Jobs page and occupation AI panels.
- The authors make the original datasets available for scholars, policymakers and practitioners and request citation when using AIOE data. JobMarketHealth cites Felten, Raj & Seamans (2021) and identifies the 2023 language-modeling and image-generation variants separately. The repository does not state a standalone workbook license, so unrestricted commercial redistribution of those workbooks has not been confirmed. Never an input to the Job Market Health score.
Where it sits
This dataset is AI & Jobs research context. It is never an input to the Job Market Health score. Read the methodology.
Official source
Felten, Raj & Seamans (2021) publishes AI Occupational Exposure (AIOE). JobMarketHealth organizes and derives from that public data and is not its publisher.
Ready to look at an occupation? Search occupations to see the score, wage context and outlook for a specific job.