docs/us_economy/economy_ai_occupation_exposure_county_year

economy_ai_occupation_exposure_county_year

County-year (2024 baseline). Employment-weighted AI occupational exposure. County occupation employment (BLS QCEW county industry mix expanded by OEWS national NAICS×SOC staffing, shift-share) is weighted against two published occupation-level exposure indices: AIOE (Felten-Raj-Seamans; z-scored, >0 = above-average) and Eloundou et al. "GPTs are GPTs" beta (0-1 share of tasks LLM-exposed). Reports employment-weighted exposure, the share of employment above average, the high- vs low-exposure mean-wage split, and an exposure_coverage flag. The d/r-bar displacement shock sweep is parameterized in the analysis layer, not in this table.

tier cTier C: the exposure weights come from academic indices (Felten-Raj-Seamans AIOE; Eloundou et al. GPT exposure) that are contested and not official statistics, even though the employment substrate (BLS QCEW × OEWS) is Tier A. The composite inherits the lower tier of its weakest input.aiautomationoccupation-exposurelaboremploymentshift-share
grain
county-year
years
2024
cadence
Annual
overview

overview

3,170 counties for 2024 (single baseline year). exposure_coverage reports the share of each county's employment carrying an AIOE score; counties with thin scored employment have lower coverage and noisier weighted means.

current vintage — 2024

history — No — single 2024 baseline vintage (the exposure indices are static)

provenance

source & licensing

authority
Point Luna (derived) — BLS QCEW & OEWS employment substrate; Felten, Raj & Seamans AIOE; Eloundou et al. GPT exposure
dataset
AI occupational exposure (employment-weighted, county-year)
license
Public domain (BLS QCEW/OEWS substrate, U.S. Federal Government work); academic exposure indices (AIOE, Eloundou et al.) used under fair use for research.
citation
Felten, E., Raj, M., & Seamans, R. (2021). Occupational, industry, and geographic exposure to artificial intelligence. Strategic Management Journal. Eloundou, T., Manning, S., Mishkin, P., & Rock, D. (2023). GPTs are GPTs. Employment substrate: U.S. Bureau of Labor Statistics, QCEW (2024) and OEWS (May 2024).
schema

fields

nametypedefinition
country_idstringISO alpha-2 country code (always 'US' for domestic tables).
county_idkeystring5-character FIPS code identifying the county.Part of primary key. Joins dim.counties on county_id.
employment_weighted_aioefloat64 · index (z-scored, mean≈0)Employment-weighted mean AIOE (Felten-Raj-Seamans) occupational exposure index, z-scored so >0 means above-average AI exposure; weighted only over employment carrying an AIOE score. Higher values indicate the county's job mix is more exposed to AI.
employment_weighted_eloundou_beta_dvfloat64 · share (0–1)Employment-weighted mean Eloundou et al. ("GPTs are GPTs") GPT exposure beta using the data+vision (dv) rubric, where the beta is the 0–1 share of an occupation's tasks that are LLM-exposed. Higher values indicate a greater share of county work tasks exposed to LLMs.
employment_weighted_eloundou_beta_humanfloat64 · share (0–1)Employment-weighted mean Eloundou et al. GPT exposure beta using the human-rater rubric, where the beta is the 0–1 share of an occupation's tasks that are LLM-exposed. Higher values indicate a greater share of county work tasks exposed to LLMs.
employment_weighted_lm_aioefloat64 · index (z-scored, mean≈0)Employment-weighted mean of the language-modeling AIOE variant (LM AIOE), the LLM-specific exposure cut of the AIOE index. Higher values indicate greater exposure of the county's job mix to language-model capabilities specifically.
exposure_coveragefloat64 · share (0–1)Share of total county employment carrying an AIOE score, serving as a coverage flag. Lower values mean a larger fraction of employment is unscored and excluded from the weighted means, so the exposure metrics are less representative.
mean_wage_high_exposurefloat64 · dollars (nominal)Employment-weighted mean OEWS annual wage of high-exposure (AIOE > 0) county employment. Higher values indicate better-paid work among the more AI-exposed occupations.
mean_wage_low_exposurefloat64 · dollars (nominal)Employment-weighted mean OEWS annual wage of low-exposure (AIOE <= 0) county employment. Comparison against mean_wage_high_exposure shows whether AI exposure concentrates in higher- or lower-paid work.
share_employment_above_avg_aioefloat64 · share (0–1)Share of scored county employment in occupations with AIOE > 0 (above-average AI exposure). Higher values indicate more of the county workforce is in above-average-exposure occupations.
state_idstring2-character FIPS code identifying the state.Joins dim.states on state_id.
total_employmentfloat64 · workersTotal county employment summed across occupations, derived from the QCEW county industry mix expanded by OEWS national NAICS×SOC staffing shares (shift-share estimate). Higher values indicate a larger county workforce.
yearkeyint64Reference year of the observation.Part of primary key.
relationships

joins

primary key
county_id, year
common joins
not provided
usage

how to use this table

method

County NAICS employment (QCEW) is expanded into occupation employment via OEWS national NAICS×SOC staffing shares (shift-share), priced at the OEWS SOC mean annual wage. Each occupation's employment is weighted by its exposure score and aggregated to county-year. AIOE is z-scored (mean≈0); the Eloundou beta is a 0-1 task share, with O*NET-SOC scores collapsed to 6-digit SOC by unweighted mean before weighting. Employment-weighted means are computed only over employment carrying a score; exposure_coverage records that scored share.

do not use for

Treating cross-county exposure differences as precise (the substrate is a national- staffing shift-share estimate, not a county occupation census); single-year change detection (single 2024 vintage); reading AIOE z-scores as probabilities or Eloundou betas as realized displacement. The actual displacement shock (d/r-bar sweep) is not in this table.

known issues

QCEW statewide aggregate-area pseudo-counties (000, 996-999) and CT/territory FIPS without a dim.geographies match are filtered upstream in int_county_occupation_employment, so a small amount of employment is dropped before exposure weighting. AIOE (SOC-2010) and Eloundou (O*NET-SOC 2019) keys are matched to OEWS 6-digit SOC; occupations without a score are excluded from the weighted means and lower exposure_coverage.

last updated · Jul 7, 2026