Abstract:

Aims: This study examines whether U.S. occupations with greater exposure to large language models showed weaker wage and employment performance during the first measurable years after the public release of ChatGPT, and whether the observed pattern has implications for workforce education and social inequality.
Study design: Longitudinal occupation-level empirical analysis using national U.S. occupational wage and employment data.
Place and Duration of Study: United States national labor market; Bureau of Labor Statistics Occupational Employment and Wage Statistics data for 2018-2025, analyzed in 2026.
Methodology: Annual occupation-level median wages and employment estimates were merged with published large-language-model exposure scores by six-digit Standard Occupational Classification code. Occupations were grouped into four exposure bins, and post-2022 changes in real median wages and employment were compared. A descriptive ordinary least squares trend was estimated between continuous AI exposure and real wage growth.
Results: Employment growth from 2022 to 2025 declined as exposure increased: low-exposure occupations grew by about 5.0%, medium-low exposure occupations by 6.1%, medium-high exposure occupations by 2.6%, and high-exposure occupations by only 0.5%. High-exposure occupations did not receive a broad real wage premium. The occupation-level wage trend was negative, with an estimated slope of approximately -2.9 percentage points from the least-exposed to the most-exposed end of the distribution, although dispersion across occupations remained substantial.
Conclusion: The evidence is best interpreted as an early descriptive signal, not a causal estimate. The results are more consistent with early AI-related wage stagnation and slower job-count growth in highly exposed occupations than with an immediate high-skill wage boom. Workforce education should therefore emphasize AI literacy, domain expertise, statistical reasoning, communication, ethics, and human judgment rather than narrow tool use alone.