Occupations most exposed to AI pay a premium rather than a penalty, with the wage advantage concentrated among cognitively demanding roles; the pattern holds across the wage distribution and in instrumental-variable tests.

Artificial intelligence exposure and occupational wages: Evidence from the United States

Ozan Atalay · June 24, 2026 · Journal of Economic Studies

openalex quasi_experimental medium evidence 7/10 relevance Summary only summary available; pdf_status=not_found DOI Source

Using occupation-level data for 671 U.S. occupations, the paper finds that higher measured potential exposure to AI is associated with higher wages — a relationship that is stronger in cognitively intensive occupations and persists across wage quantiles and under IV checks.

Purpose This paper investigates the relationship between occupational exposure to artificial intelligence (AI) and wage structures in the United States. While much of the literature focuses on the displacement effects of AI, less attention has been given to its implications for wage outcomes across occupations. Design/methodology/approach The study uses occupation-level data for 671 occupations, combining wage information with an AI exposure index that captures the extent to which occupations are affected by AI technologies. Cross-sectional regression models with robust standard errors are employed, controlling for employment size and occupational characteristics. Quantile regression is also used to examine variation across the wage distribution. Findings The results indicate a positive and statistically significant association between AI exposure and wages. Occupations with higher exposure tend to have higher wage levels. This pattern is consistent across model specifications and across the wage distribution. The findings are broadly consistent when using an instrumental variable approach. The association is stronger in occupations with higher cognitive skill intensity. Research limitations/implications This study is based on occupation-level data rather than individual-level observations, which limits the ability to capture within-occupation wage heterogeneity. In addition, the AI exposure index reflects potential exposure rather than actual adoption at the firm level. Future research could extend this analysis using firm-level or longitudinal data. Practical implications The findings suggest that occupations with higher exposure to artificial intelligence tend to exhibit higher wages, highlighting the importance of skill upgrading and targeted workforce policies. Policymakers and organizations should focus on enhancing digital skills and supporting workforce transition to maximize the benefits of AI. Social implications The results indicate that artificial intelligence may contribute to wage differences across occupations by enhancing productivity in certain roles. This highlights the need to ensure equal access to skills and training opportunities so that the benefits of AI are distributed more evenly across the labor market. Originality/value This study contributes by providing occupation-level evidence on the relationship between AI exposure and wages, shifting attention from employment effects to wage structures. It also highlights that this relationship is partly explained by occupational skill composition, while a residual association remains.

Summary

Main Finding

Occupations with higher measured exposure to AI have higher wage levels in the U.S. labor market. This positive association is statistically robust across cross-sectional specifications, quantile regressions (across the wage distribution), and an instrumental-variable (IV) robustness check. The link is particularly strong for occupations with higher cognitive-skill intensity; skill composition explains part, but not all, of the relationship.

Key Points

Data: occupation-level analysis covering 671 occupations.
Core result: AI exposure index is positively and significantly associated with occupation-level wages.
Robustness: result holds across model specifications, across the wage distribution (quantile regressions), and when using an IV approach.
Heterogeneity: the wage premium associated with AI exposure is larger in occupations with higher cognitive skill intensity.
Partial mediation: occupational skill composition explains some of the association; a residual positive relationship remains.
Limitations noted by authors: use of occupation-level (not individual- or firm-level) data and an AI exposure index that captures potential exposure rather than observed firm-level AI adoption.

Data & Methods

Sample: 671 occupations (occupation-level dataset).
Key variables:
- Dependent variable: occupational wage levels.
- Main independent variable: AI exposure index (measures extent an occupation is affected by AI technologies; reflects potential exposure).
- Controls: employment size and other occupational characteristics (e.g., skill composition).
Estimation:
- Cross-sectional OLS regressions with robust standard errors.
- Quantile regressions to assess effects across the wage distribution.
- Instrumental-variable (IV) approach used as a robustness check (details not specified in the abstract).
Identification & limitations:
- Cross-sectional design limits causal inference and cannot observe within-occupation wage heterogeneity or actual firm-level AI adoption.
- AI exposure index measures potential rather than realized AI use.

Implications for AI Economics

Complementarity and skill premiums: Results are consistent with AI acting as a complement to cognitive skills, increasing returns to occupations with higher cognitive intensity and contributing to higher wages in those roles.
Distributional impact: AI exposure is linked to wage differences across occupations, implying AI adoption may widen or reshape wage structures depending on which occupations are exposed and complemented.
Policy relevance: Findings support policies that emphasize digital and cognitive-skill upgrading, targeted retraining, and broader access to AI-related education to distribute benefits more evenly.
Research agenda: To strengthen causal claims and inform policy, future work should use individual- or firm-level and longitudinal data to measure actual AI adoption, trace within-occupation heterogeneity, and identify mechanisms (e.g., productivity gains, task reallocation, selection into occupations).
Caution for interpretation: Because the index captures potential exposure and the analysis is cross-sectional, observed associations may reflect selection (e.g., higher-wage occupations attracting AI applications) as well as complementarity effects; deeper causal analysis is required.

Assessment

Paper Typequasi_experimental Evidence Strengthmedium — The paper presents consistent positive associations across OLS, quantile regressions, and an IV robustness check, which strengthens the claim; however the analysis is cross-sectional at the occupation level (ecological aggregation), the AI measure reflects potential exposure rather than observed adoption, and details on the instrument and its validity are not provided, leaving concerns about omitted variable bias, reverse causality, and measurement error. Methods Rigormedium — Analytical methods are appropriate and varied (robust SEs, quantile regression, IV robustness), and models control for basic occupational covariates; nevertheless the reliance on cross-sectional aggregated data, limited control for firm- or individual-level confounders, and lack of transparency about the IV weaken causal interpretation and internal validity. SampleOccupation-level dataset covering 671 U.S. occupations, combining occupation mean wages, employment size, occupational characteristics (e.g., skill composition), and an AI exposure index that measures potential exposure of occupations to AI technologies; cross-sectional (single-period) analysis. Themeslabor_markets skills_training IdentificationCross-sectional occupation-level OLS regressions relating an occupation AI-exposure index to mean wages, controlling for employment size and occupational characteristics; quantile regressions to examine the wage distribution; results checked with an instrumental-variable (IV) estimator (instrument not specified in the summary). GeneralizabilityOccupation-level (aggregated) data — cannot capture within-occupation heterogeneity or individual-level effects (ecological inference problem), Measures potential exposure to AI, not actual firm- or worker-level AI adoption/use, U.S.-only occupations — results may not generalize to other countries or institutional contexts, Cross-sectional snapshot — limits inference about dynamics, causal timing, or long-term impacts, Instrument validity and strength not fully described — causal claims may not generalize if IV is weak or context-specific

Claims (11)

Claim	Direction	Outcome	Confidence & Evidence	Details
There is a positive and statistically significant association between AI exposure and wages. Wages	positive	wage levels	Reading fidelity high Study strength medium	n=671 0.48
The positive association between AI exposure and wages is robust across different model specifications. Wages	positive	wage levels	Reading fidelity high Study strength medium	n=671 0.48
The positive association between AI exposure and wages holds across the wage distribution. Wages	positive	wage distribution (quantiles)	Reading fidelity high Study strength medium	n=671 0.48
Findings are broadly consistent when using an instrumental variable (IV) approach. Wages	positive	wage levels	Reading fidelity high Study strength medium	n=671 0.48
The positive association between AI exposure and wages is stronger in occupations with higher cognitive skill intensity. Wages	positive	wage levels	Reading fidelity medium Study strength medium	n=671 0.29
Occupational skill composition partly explains the relationship between AI exposure and wages, but a residual association remains after accounting for skills. Wages	positive	wage levels (conditional on occupational skill composition)	Reading fidelity medium Study strength medium	n=671 0.29
The study uses occupation-level data for 671 occupations combining wage information with an AI exposure index. Other	null_result	data coverage (occupational wages and AI exposure index)	Reading fidelity high Study strength high	n=671 0.8
The AI exposure index reflects potential exposure to AI technologies rather than actual firm-level AI adoption. Other	null_result	measurement scope of the AI exposure index	Reading fidelity high Study strength high	n=671 0.8
Policy implication: occupations with higher exposure to AI tend to exhibit higher wages, suggesting the importance of skill upgrading and targeted workforce policies. Wages	positive	wage levels and policy-relevant recommendations	Reading fidelity high Study strength speculative	n=671 0.08
Social implication: AI may contribute to wage differences across occupations by enhancing productivity in certain roles, so equitable access to skills and training is important to distribute benefits. Wages	positive	wage differences across occupations	Reading fidelity high Study strength speculative	n=671 0.08
A limitation of the study is that using occupation-level data prevents capturing within-occupation wage heterogeneity. Other	null_result	ability to capture within-occupation wage heterogeneity	Reading fidelity high Study strength high	n=671 0.8