ChatGPT’s public debut widened pay gaps inside Chinese listed firms: generative AI raised managerial productivity and pay more than ordinary employees, shifting returns toward management—especially in large firms and where worker bargaining power is weak.

Main Finding

The release of ChatGPT‑3.5 produced an exogenous shock that widened the internal pay gap between management and ordinary employees at Chinese A‑share listed firms (2018–2023). Structural decomposition shows the widening is driven by an increase in efficiency‑based pay (higher marginal productivity for managers), not by increased managerial rent‑seeking. Generative AI acts as a cognitive complement to management—boosting technological innovation and operational performance—and the distributional gains accrue disproportionately to managers, especially in large firms and where employees have limited bargaining power.

Key Points

• Identification: The paper treats the public release of ChatGPT‑3.5 as an exogenous technology shock and implements a difference‑in‑differences design to estimate effects on intra‑firm pay structure.
• Exposure measure: Firm technological exposure to generative AI is measured through textual analysis of firms’ business‑scope descriptions (matching AI‑relevant terms).
• Main outcome: The internal pay gap (management vs. ordinary employees) widens after the shock for AI‑exposed firms.
• Mechanism: Structural decomposition attributes the widening to efficiency‑based pay premia (higher manager marginal productivity), not to increases in managerial rent extraction.
• Channels: Evidence points to two complementary channels—enhanced technological innovation and improved operational performance tied to managerial use of generative AI.
• Heterogeneity: Effects are strongest in large firms and in firms where worker bargaining power is weaker.
• Theory tested: Provides micro‑level empirical support for task‑based displacement/complementarity frameworks—AI complements cognitive managerial tasks while substituting or compressing routine employee tasks.
• Institutional role: Institutional frictions (e.g., bargaining power, firm size) condition how AI gains are distributed within firms.

Data & Methods

• Sample: Chinese A‑share listed firms, 2018–2023.
• Empirical strategy: Difference‑in‑differences exploiting timing of ChatGPT‑3.5 release as an exogenous shock; likely includes firm and year fixed effects and controls for firm characteristics (as described in the abstract).
• Exposure measurement: Textual analysis of corporate business‑scope statements to construct a firm‑level technological exposure index to generative AI.
• Outcome measurement: Firm‑level internal pay gap between management and ordinary employees (derived from firm disclosures).
• Decomposition: Structural decomposition separates the change in the pay gap into an efficiency‑based component (linked to productivity/performance) and a rent‑seeking component (managerial extraction). The widening is driven by the former.
• Robustness & heterogeneity: Results hold more strongly in larger firms and in contexts with limited employee bargaining power, consistent with institutional amplification of distributional effects.

Implications for AI Economics

• Distributional effects within firms: Generative AI can raise overall firm productivity while reallocating returns toward managerial cohorts—heightening within‑firm wage inequality even absent outright rent‑seeking.
• Task‑based theory confirmation: The paper provides micro‑level empirical support for models where AI complements higher‑order cognitive tasks and may substitute routine employee tasks, reshaping wage structures.
• Role of institutions: Worker bargaining power, firm size, and other institutional frictions importantly mediate how AI gains are shared—policy interventions (collective bargaining, wage negotiation frameworks, redistribution mechanisms) will affect the incidence of AI‑driven inequality.
• Policy and corporate governance: Regulators and firms should consider measures to (a) support upskilling and task reallocation for nonmanagerial workers, (b) strengthen employee bargaining channels or compensation frameworks, and (c) monitor compensation governance to ensure productivity gains translate into broader welfare gains.
• Research directions: Extend analysis to non‑listed firms and other countries, examine long‑run labor reallocation and career dynamics, quantify external labor‑market spillovers, and test specific policies (bargaining, taxation, training) that alter intra‑firm distributional outcomes.