AI, Wages and Firm Structure in Knowledge Intensive Industries

Abstract

We study how generative AI affects wages and firm structure in knowledge-intensive industries. We exploit the public release of ChatGPT on November 30, 2022 as a sharp shock and combine U.S. Census OEWS data with Revelio workforce data. Our empirical strategy compares wage and employment dynamics across occupations with different pre-shock exposure to generative AI, using a continuous-treatment event-study difference-in-differences design. AI exposure is measured at the occupation level using Eloundou et al.’s task-based exposure scores. To address the concern that AI-exposed occupations are also more suitable for remote work, we construct nonparametric generalized-propensity-score weights that control for alternative work-from-home measures. We find that generative AI exposure is associated with a sizable decline in wages, both in nationally representative OEWS occupation-by-industry cells and within firm-by-occupation cells in Revelio. The effect appears as wage repricing rather than immediate employment displacement: overall employment responses are small and statistically unstable. Within firms, wage declines are concentrated among low- and high-seniority workers. Firms also reorganize job ladders, reducing the share of low- and high-seniority positions while increasing the share of mid-seniority positions. The evidence suggests that generative AI is already reshaping the price and structure of exposed work inside firms.