This site is undergoing review. Some annotations were human-generated, some AI-generated — all are being verified.

AI and the Extended Workday: Productivity, Contracting Efficiency, and Distribution of Rents

Jiang, Park, Xiao, Zhang

2025NBER Working Paper4 citations

Observational labor marketCausalTheoretical model

LLM / Generative AIAI (General)Human-AI collaborationAugmentation vs. substitutionGeneral automationPlatforms / gig economy

DOI: 10.3386/w33536

Summary

Jiang et al. use American Time Use Survey data (2004-2023) with occupation-level AI exposure measures constructed from patent-task text similarity to study how AI affects work hours and time allocation, exploiting the ChatGPT launch as a natural experiment

Main Finding

Workers in occupations with higher AI exposure work 2.2 more hours per week (cross-sectional) and experience reduced leisure time, with effects strongest when AI complements labor and in competitive markets where workers have limited bargaining power to capture productivity gains

Primary Datasets

ATUS

Other

American Time Use Survey (ATUS)

Secondary Datasets

LinkedIn Economic Graph

Revelio Labs

Compustat

Google Trends

LinkedIn employment data via Revelio Lab; Glassdoor employee reviews via Revelio Lab; Compustat; CRSP; Google search trends; Hoberg-Phillips product similarity; Oxfam Best States to Work Index

Key Methods: Difference-in-differences around ChatGPT launch; weighted OLS with occupation-level AI exposure measures; extensive fixed effects; natural experiments using COVID remote work shock
Sample Period: 2004-2023
Geographic Coverage: US
Sample Size: 123,603 individuals in ATUS; 905,667 AI patents (2000-2023); 9,270 top-cited AI patents used for exposure construction
Level of Analysis: Individual, Occupation, Firm
Occupation Classification: O*NET SOC codes aggregated to occ1990dd (Dorn 2009)
Industry Classification: Census industry code (trdtind1)

Notes

NBER WP 33536; higher AI exposure associated with longer work hours and less leisure [Claude classification]: Uses ChatGPT launch (November 2022) as natural experiment. Theoretical framework based on Holmstrom-Milgrom principal-agent model. Combines multiple AI exposure measures including general AI, generative AI, AI complementarity, and AI surveillance. Effect heterogeneity examined by labor and product market competition. Employee satisfaction data from Glassdoor shows declining welfare despite wage increases. [Claude classification]: Uses ChatGPT launch (November 2022) as natural experiment. Theoretical framework based on Holmstrom-Milgrom principal-agent model. Combines multiple AI exposure measures including general AI, generative AI, AI complementarity, and AI surveillance. Effect heterogeneity examined by labor and product market competition. Employee satisfaction data from Glassdoor shows declining welfare despite wage increases. [Claude classification]: Uses ChatGPT launch (November 2022) as natural experiment. Theoretical framework based on Holmstrom-Milgrom principal-agent model. Combines multiple AI exposure measures including general AI, generative AI, AI complementarity, and AI surveillance. Effect heterogeneity examined by labor and product market competition. Employee satisfaction data from Glassdoor shows declining welfare despite wage increases. [Claude classification]: Uses ChatGPT launch (November 2022) as natural experiment. Theoretical framework based on Holmstrom-Milgrom principal-agent model. Combines multiple AI exposure measures including general AI, generative AI, AI complementarity, and AI surveillance. Effect heterogeneity examined by labor and product market competition. Employee satisfaction data from Glassdoor shows declining welfare despite wage increases. [Claude classification]: Uses ChatGPT launch (November 2022) as natural experiment. Theoretical framework based on Holmstrom-Milgrom principal-agent model. Combines multiple AI exposure measures including general AI, generative AI, AI complementarity, and AI surveillance. Effect heterogeneity examined by labor and product market competition. Employee satisfaction data from Glassdoor shows declining welfare despite wage increases. [Claude classification]: Uses ChatGPT launch (November 2022) as natural experiment. Theoretical framework based on Holmstrom-Milgrom principal-agent model. Combines multiple AI exposure measures including general AI, generative AI, AI complementarity, and AI surveillance. Effect heterogeneity examined by labor and product market competition. Employee satisfaction data from Glassdoor shows declining welfare despite wage increases. [Claude classification]: Uses ChatGPT launch (November 2022) as natural experiment. Theoretical framework based on Holmstrom-Milgrom principal-agent model. Combines multiple AI exposure measures including general AI, generative AI, AI complementarity, and AI surveillance. Effect heterogeneity examined by labor and product market competition. Employee satisfaction data from Glassdoor shows declining welfare despite wage increases. [Claude classification]: Uses ChatGPT launch (November 2022) as natural experiment. Theoretical framework based on Holmstrom-Milgrom principal-agent model. Combines multiple AI exposure measures including general AI, generative AI, AI complementarity, and AI surveillance. Effect heterogeneity examined by labor and product market competition. Employee satisfaction data from Glassdoor shows declining welfare despite wage increases. [Claude classification]: Uses ChatGPT launch (November 2022) as natural experiment. Theoretical framework based on Holmstrom-Milgrom principal-agent model. Combines multiple AI exposure measures including general AI, generative AI, AI complementarity, and AI surveillance. Effect heterogeneity examined by labor and product market competition. Employee satisfaction data from Glassdoor shows declining welfare despite wage increases. [Claude classification]: Uses ChatGPT launch (November 2022) as natural experiment. Theoretical framework based on Holmstrom-Milgrom principal-agent model. Combines multiple AI exposure measures including general AI, generative AI, AI complementarity, and AI surveillance. Effect heterogeneity examined by labor and product market competition. Employee satisfaction data from Glassdoor shows declining welfare despite wage increases. [Claude classification]: Uses ChatGPT launch (November 2022) as natural experiment. Theoretical framework based on Holmstrom-Milgrom principal-agent model. Combines multiple AI exposure measures including general AI, generative AI, AI complementarity, and AI surveillance. Effect heterogeneity examined by labor and product market competition. Employee satisfaction data from Glassdoor shows declining welfare despite wage increases. [Claude classification]: Uses ChatGPT launch (November 2022) as natural experiment. Theoretical framework based on Holmstrom-Milgrom principal-agent model. Combines multiple AI exposure measures including general AI, generative AI, AI complementarity, and AI surveillance. Effect heterogeneity examined by labor and product market competition. Employee satisfaction data from Glassdoor shows declining welfare despite wage increases. [Claude classification]: Uses ChatGPT launch (November 2022) as natural experiment. Theoretical framework based on Holmstrom-Milgrom principal-agent model. Combines multiple AI exposure measures including general AI, generative AI, AI complementarity, and AI surveillance. Effect heterogeneity examined by labor and product market competition. Employee satisfaction data from Glassdoor shows declining welfare despite wage increases. [Claude classification]: Uses ChatGPT launch (November 2022) as natural experiment. Theoretical framework based on Holmstrom-Milgrom principal-agent model. Combines multiple AI exposure measures including general AI, generative AI, AI complementarity, and AI surveillance. Effect heterogeneity examined by labor and product market competition. Employee satisfaction data from Glassdoor shows declining welfare despite wage increases. [Claude classification]: Uses ChatGPT launch (November 2022) as natural experiment. Theoretical framework based on Holmstrom-Milgrom principal-agent model. Combines multiple AI exposure measures including general AI, generative AI, AI complementarity, and AI surveillance. Effect heterogeneity examined by labor and product market competition. Employee satisfaction data from Glassdoor shows declining welfare despite wage increases. [Claude classification]: Uses ChatGPT launch (November 2022) as natural experiment. Theoretical framework based on Holmstrom-Milgrom principal-agent model. Combines multiple AI exposure measures including general AI, generative AI, AI complementarity, and AI surveillance. Effect heterogeneity examined by labor and product market competition. Employee satisfaction data from Glassdoor shows declining welfare despite wage increases. [Claude classification]: Uses ChatGPT launch (November 2022) as natural experiment. Theoretical framework based on Holmstrom-Milgrom principal-agent model. Combines multiple AI exposure measures including general AI, generative AI, AI complementarity, and AI surveillance. Effect heterogeneity examined by labor and product market competition. Employee satisfaction data from Glassdoor shows declining welfare despite wage increases. [Claude classification]: Uses ChatGPT launch (November 2022) as natural experiment. Theoretical framework based on Holmstrom-Milgrom principal-agent model. Combines multiple AI exposure measures including general AI, generative AI, AI complementarity, and AI surveillance. Effect heterogeneity examined by labor and product market competition. Employee satisfaction data from Glassdoor shows declining welfare despite wage increases. [Claude classification]: Uses ChatGPT launch (November 2022) as natural experiment. Theoretical framework based on Holmstrom-Milgrom principal-agent model. Combines multiple AI exposure measures including general AI, generative AI, AI complementarity, and AI surveillance. Effect heterogeneity examined by labor and product market competition. Employee satisfaction data from Glassdoor shows declining welfare despite wage increases. [Claude classification]: Uses ChatGPT launch (November 2022) as natural experiment. Theoretical framework based on Holmstrom-Milgrom principal-agent model. Combines multiple AI exposure measures including general AI, generative AI, AI complementarity, and AI surveillance. Effect heterogeneity examined by labor and product market competition. Employee satisfaction data from Glassdoor shows declining welfare despite wage increases. [Claude classification]: Uses ChatGPT launch (November 2022) as natural experiment. Theoretical framework based on Holmstrom-Milgrom principal-agent model. Combines multiple AI exposure measures including general AI, generative AI, AI complementarity, and AI surveillance. Effect heterogeneity examined by labor and product market competition. Employee satisfaction data from Glassdoor shows declining welfare despite wage increases. [Claude classification]: Uses ChatGPT launch (November 2022) as natural experiment. Theoretical framework based on Holmstrom-Milgrom principal-agent model. Combines multiple AI exposure measures including general AI, generative AI, AI complementarity, and AI surveillance. Effect heterogeneity examined by labor and product market competition. Employee satisfaction data from Glassdoor shows declining welfare despite wage increases. [Claude classification]: Uses ChatGPT launch (November 2022) as natural experiment. Theoretical framework based on Holmstrom-Milgrom principal-agent model. Combines multiple AI exposure measures including general AI, generative AI, AI complementarity, and AI surveillance. Effect heterogeneity examined by labor and product market competition. Employee satisfaction data from Glassdoor shows declining welfare despite wage increases. [Claude classification]: Uses ChatGPT launch (November 2022) as natural experiment. Theoretical framework based on Holmstrom-Milgrom principal-agent model. Combines multiple AI exposure measures including general AI, generative AI, AI complementarity, and AI surveillance. Effect heterogeneity examined by labor and product market competition. Employee satisfaction data from Glassdoor shows declining welfare despite wage increases. [Claude classification]: Uses ChatGPT launch (November 2022) as natural experiment. Theoretical framework based on Holmstrom-Milgrom principal-agent model. Combines multiple AI exposure measures including general AI, generative AI, AI complementarity, and AI surveillance. Effect heterogeneity examined by labor and product market competition. Employee satisfaction data from Glassdoor shows declining welfare despite wage increases.