Click any bubble on the chart to see the full profile for that occupation — where it sits on the two AI axes, employment and wages, and the model’s predicted employment change.
—
SOC —
Zone
Estimated relative annual employment change · post-ChatGPT (2021–2025)
—
90% CI: —
Substitution effect —
Complement effect —
* Complement effect capped to 0 for this occupation — the fitted parabola turns sharply negative at high C, which is not economically meaningful for hands-on human-services roles like counseling/therapy.
S + C + I = 1 (task composition). The headline is a difference-in-differences estimate: the model’s fitted annual employment growth attributable to the occupation’s S and C, relative to a purely physical baseline (I = 1), net of pre-2021 trend. Formula: Δg = exp(βS·S + βC·C + βC²·C²) − 1 with βS = −0.0563, βC = +0.3002, βC² = −0.4909 (dose-response spec, reference note Figure 4). Not the observed employment change for the occupation. Bands are typically wide; the zone is a more robust signal than the point estimate.
AI Job Impact MapUS Occupations Broken Down by AI Substitution and Complement