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

AI-Exposed Jobs Deteriorated Before ChatGPT

Frank, Sabet, Simon, Bana, Yu

2026arXiv

Observational labor marketInterdisciplinary

LLM / Generative AISoftware / codingJunior / entry-levelEducation

View Repository DOI: 10.48550/arXiv.2601.02554

Abstract

Public debate links worsening job prospects for AI-exposed occupations to the release of ChatGPT in late 2022. Using monthly U.S. unemployment insurance records, we measure occupation- and location-specific unemployment risk and find that risk rose in AI-exposed occupations beginning in early 2022, months before ChatGPT. Analyzing millions of LinkedIn profiles, we show that graduate cohorts from 2021 onward entered AI-exposed jobs at lower rates than earlier cohorts, with gaps opening before late 2022. Finally, from millions of university syllabi, we find that graduates taking more AI-exposed curricula had higher first-job pay and shorter job searches after ChatGPT. Together, these results point to forces pre-dating generative AI and to the ongoing value of LLM-relevant education.

Summary

Frank et al. combine monthly U.S. unemployment insurance records, millions of LinkedIn profiles, and university syllabi to show that unemployment risk and job-entry rates for LLM-exposed occupations began deteriorating in early 2022, months before ChatGPT's November 2022 release, while educational exposure to LLM-relevant skills predicted better post-ChatGPT outcomes.

Main Finding

Unemployment risk in highly LLM-exposed occupations (especially computer/math) rose beginning in early 2022—well before ChatGPT's November 2022 launch—and graduates from 2021-2023 cohorts entered highly exposed jobs at lower rates with gaps opening before late 2022, yet higher curricular exposure to LLM-related tasks predicted higher first-job salaries and shorter job searches after ChatGPT.

Primary Datasets

UI Wage Records

Labor outcomes

LinkedIn Economic Graph

Labor outcomes

U.S. unemployment insurance records; LinkedIn profiles; University syllabi

Secondary Datasets

O*NET

Other

NSCG

Labor outcomes

BLS Occupational Employment and Wage Statistics (OEWS); O*NET task data; National Survey of College Graduates (NSCG); BLS JOLTS

Key Methods: Occupation- and location-specific unemployment risk measurement using administrative UI claims; event-study regression analysis of LinkedIn graduate cohorts' job delays and entry rates by LLM exposure; regression analysis linking syllabus-derived educational exposure to post-graduation labor market outcomes with detailed fixed effects.
Sample Period: 2018-2024
Geographic Coverage: US
Sample Size: 10,584,980 LinkedIn profiles; 3 million syllabi; monthly UI claims for 22 major occupation groups across 50 states
Level of Analysis: Individual, Occupation, Region
Occupation Classification: Standard Occupation Classification (SOC) codes (2-digit major groups and 6-digit detailed occupations)
Industry Classification: 2-digit NAICS codes
Replication Package: Partial

Notes

arXiv preprint arXiv:2601.02554. Finds AI-exposed occupation deterioration began early 2022, months before ChatGPT release. [Claude classification]: Paper explicitly cautions against treating ChatGPT's November 2022 launch as a clean natural experiment. Uses proprietary Revelio Labs salary estimates (ML-based, trained on visa data). Educational exposure measure derived from pre-2020 syllabi, representing baseline curricular differences rather than post-ChatGPT responses. Excludes graduates whose first job begins >3 years after degree completion. Some state-occupation cells have missing UI data (see Table S1). Connecticut data shows anomalies for certain occupations. [Claude classification]: Paper explicitly cautions against treating ChatGPT's November 2022 launch as a clean natural experiment. Uses proprietary Revelio Labs salary estimates (ML-based, trained on visa data). Educational exposure measure derived from pre-2020 syllabi, representing baseline curricular differences rather than post-ChatGPT responses. Excludes graduates whose first job begins >3 years after degree completion. Some state-occupation cells have missing UI data (see Table S1). Connecticut data shows anomalies for certain occupations. [Claude classification]: Paper explicitly cautions against treating ChatGPT's November 2022 launch as a clean natural experiment. Uses proprietary Revelio Labs salary estimates (ML-based, trained on visa data). Educational exposure measure derived from pre-2020 syllabi, representing baseline curricular differences rather than post-ChatGPT responses. Excludes graduates whose first job begins >3 years after degree completion. Some state-occupation cells have missing UI data (see Table S1). Connecticut data shows anomalies for certain occupations. [Claude classification]: Paper explicitly cautions against treating ChatGPT's November 2022 launch as a clean natural experiment. Uses proprietary Revelio Labs salary estimates (ML-based, trained on visa data). Educational exposure measure derived from pre-2020 syllabi, representing baseline curricular differences rather than post-ChatGPT responses. Excludes graduates whose first job begins >3 years after degree completion. Some state-occupation cells have missing UI data (see Table S1). Connecticut data shows anomalies for certain occupations. [Claude classification]: Paper explicitly cautions against treating ChatGPT's November 2022 launch as a clean natural experiment. Uses proprietary Revelio Labs salary estimates (ML-based, trained on visa data). Educational exposure measure derived from pre-2020 syllabi, representing baseline curricular differences rather than post-ChatGPT responses. Excludes graduates whose first job begins >3 years after degree completion. Some state-occupation cells have missing UI data (see Table S1). Connecticut data shows anomalies for certain occupations. [Claude classification]: Paper explicitly cautions against treating ChatGPT's November 2022 launch as a clean natural experiment. Uses proprietary Revelio Labs salary estimates (ML-based, trained on visa data). Educational exposure measure derived from pre-2020 syllabi, representing baseline curricular differences rather than post-ChatGPT responses. Excludes graduates whose first job begins >3 years after degree completion. Some state-occupation cells have missing UI data (see Table S1). Connecticut data shows anomalies for certain occupations. [Claude classification]: Paper explicitly cautions against treating ChatGPT's November 2022 launch as a clean natural experiment. Uses proprietary Revelio Labs salary estimates (ML-based, trained on visa data). Educational exposure measure derived from pre-2020 syllabi, representing baseline curricular differences rather than post-ChatGPT responses. Excludes graduates whose first job begins >3 years after degree completion. Some state-occupation cells have missing UI data (see Table S1). Connecticut data shows anomalies for certain occupations. [Claude classification]: Paper explicitly cautions against treating ChatGPT's November 2022 launch as a clean natural experiment. Uses proprietary Revelio Labs salary estimates (ML-based, trained on visa data). Educational exposure measure derived from pre-2020 syllabi, representing baseline curricular differences rather than post-ChatGPT responses. Excludes graduates whose first job begins >3 years after degree completion. Some state-occupation cells have missing UI data (see Table S1). Connecticut data shows anomalies for certain occupations. [Claude classification]: Paper explicitly cautions against treating ChatGPT's November 2022 launch as a clean natural experiment. Uses proprietary Revelio Labs salary estimates (ML-based, trained on visa data). Educational exposure measure derived from pre-2020 syllabi, representing baseline curricular differences rather than post-ChatGPT responses. Excludes graduates whose first job begins >3 years after degree completion. Some state-occupation cells have missing UI data (see Table S1). Connecticut data shows anomalies for certain occupations. [Claude classification]: Paper explicitly cautions against treating ChatGPT's November 2022 launch as a clean natural experiment. Uses proprietary Revelio Labs salary estimates (ML-based, trained on visa data). Educational exposure measure derived from pre-2020 syllabi, representing baseline curricular differences rather than post-ChatGPT responses. Excludes graduates whose first job begins >3 years after degree completion. Some state-occupation cells have missing UI data (see Table S1). Connecticut data shows anomalies for certain occupations. [Claude classification]: Paper explicitly cautions against treating ChatGPT's November 2022 launch as a clean natural experiment. Uses proprietary Revelio Labs salary estimates (ML-based, trained on visa data). Educational exposure measure derived from pre-2020 syllabi, representing baseline curricular differences rather than post-ChatGPT responses. Excludes graduates whose first job begins >3 years after degree completion. Some state-occupation cells have missing UI data (see Table S1). Connecticut data shows anomalies for certain occupations. [Claude classification]: Paper explicitly cautions against treating ChatGPT's November 2022 launch as a clean natural experiment. Uses proprietary Revelio Labs salary estimates (ML-based, trained on visa data). Educational exposure measure derived from pre-2020 syllabi, representing baseline curricular differences rather than post-ChatGPT responses. Excludes graduates whose first job begins >3 years after degree completion. Some state-occupation cells have missing UI data (see Table S1). Connecticut data shows anomalies for certain occupations. [Claude classification]: Paper explicitly cautions against treating ChatGPT's November 2022 launch as a clean natural experiment. Uses proprietary Revelio Labs salary estimates (ML-based, trained on visa data). Educational exposure measure derived from pre-2020 syllabi, representing baseline curricular differences rather than post-ChatGPT responses. Excludes graduates whose first job begins >3 years after degree completion. Some state-occupation cells have missing UI data (see Table S1). Connecticut data shows anomalies for certain occupations. [Claude classification]: Paper explicitly cautions against treating ChatGPT's November 2022 launch as a clean natural experiment. Uses proprietary Revelio Labs salary estimates (ML-based, trained on visa data). Educational exposure measure derived from pre-2020 syllabi, representing baseline curricular differences rather than post-ChatGPT responses. Excludes graduates whose first job begins >3 years after degree completion. Some state-occupation cells have missing UI data (see Table S1). Connecticut data shows anomalies for certain occupations. [Claude classification]: Paper explicitly cautions against treating ChatGPT's November 2022 launch as a clean natural experiment. Uses proprietary Revelio Labs salary estimates (ML-based, trained on visa data). Educational exposure measure derived from pre-2020 syllabi, representing baseline curricular differences rather than post-ChatGPT responses. Excludes graduates whose first job begins >3 years after degree completion. Some state-occupation cells have missing UI data (see Table S1). Connecticut data shows anomalies for certain occupations. [Claude classification]: Paper explicitly cautions against treating ChatGPT's November 2022 launch as a clean natural experiment. Uses proprietary Revelio Labs salary estimates (ML-based, trained on visa data). Educational exposure measure derived from pre-2020 syllabi, representing baseline curricular differences rather than post-ChatGPT responses. Excludes graduates whose first job begins >3 years after degree completion. Some state-occupation cells have missing UI data (see Table S1). Connecticut data shows anomalies for certain occupations. [Claude classification]: Paper explicitly cautions against treating ChatGPT's November 2022 launch as a clean natural experiment. Uses proprietary Revelio Labs salary estimates (ML-based, trained on visa data). Educational exposure measure derived from pre-2020 syllabi, representing baseline curricular differences rather than post-ChatGPT responses. Excludes graduates whose first job begins >3 years after degree completion. Some state-occupation cells have missing UI data (see Table S1). Connecticut data shows anomalies for certain occupations. [Claude classification]: Paper explicitly cautions against treating ChatGPT's November 2022 launch as a clean natural experiment. Uses proprietary Revelio Labs salary estimates (ML-based, trained on visa data). Educational exposure measure derived from pre-2020 syllabi, representing baseline curricular differences rather than post-ChatGPT responses. Excludes graduates whose first job begins >3 years after degree completion. Some state-occupation cells have missing UI data (see Table S1). Connecticut data shows anomalies for certain occupations. [Claude classification]: Paper explicitly cautions against treating ChatGPT's November 2022 launch as a clean natural experiment. Uses proprietary Revelio Labs salary estimates (ML-based, trained on visa data). Educational exposure measure derived from pre-2020 syllabi, representing baseline curricular differences rather than post-ChatGPT responses. Excludes graduates whose first job begins >3 years after degree completion. Some state-occupation cells have missing UI data (see Table S1). Connecticut data shows anomalies for certain occupations. [Claude classification]: Paper explicitly cautions against treating ChatGPT's November 2022 launch as a clean natural experiment. Uses proprietary Revelio Labs salary estimates (ML-based, trained on visa data). Educational exposure measure derived from pre-2020 syllabi, representing baseline curricular differences rather than post-ChatGPT responses. Excludes graduates whose first job begins >3 years after degree completion. Some state-occupation cells have missing UI data (see Table S1). Connecticut data shows anomalies for certain occupations. [Claude classification]: Paper explicitly cautions against treating ChatGPT's November 2022 launch as a clean natural experiment. Uses proprietary Revelio Labs salary estimates (ML-based, trained on visa data). Educational exposure measure derived from pre-2020 syllabi, representing baseline curricular differences rather than post-ChatGPT responses. Excludes graduates whose first job begins >3 years after degree completion. Some state-occupation cells have missing UI data (see Table S1). Connecticut data shows anomalies for certain occupations. [Claude classification]: Paper explicitly cautions against treating ChatGPT's November 2022 launch as a clean natural experiment. Uses proprietary Revelio Labs salary estimates (ML-based, trained on visa data). Educational exposure measure derived from pre-2020 syllabi, representing baseline curricular differences rather than post-ChatGPT responses. Excludes graduates whose first job begins >3 years after degree completion. Some state-occupation cells have missing UI data (see Table S1). Connecticut data shows anomalies for certain occupations. [Claude classification]: Paper explicitly cautions against treating ChatGPT's November 2022 launch as a clean natural experiment. Uses proprietary Revelio Labs salary estimates (ML-based, trained on visa data). Educational exposure measure derived from pre-2020 syllabi, representing baseline curricular differences rather than post-ChatGPT responses. Excludes graduates whose first job begins >3 years after degree completion. Some state-occupation cells have missing UI data (see Table S1). Connecticut data shows anomalies for certain occupations. [Claude classification]: Paper explicitly cautions against treating ChatGPT's November 2022 launch as a clean natural experiment. Uses proprietary Revelio Labs salary estimates (ML-based, trained on visa data). Educational exposure measure derived from pre-2020 syllabi, representing baseline curricular differences rather than post-ChatGPT responses. Excludes graduates whose first job begins >3 years after degree completion. Some state-occupation cells have missing UI data (see Table S1). Connecticut data shows anomalies for certain occupations. [Claude classification]: Paper explicitly cautions against treating ChatGPT's November 2022 launch as a clean natural experiment. Uses proprietary Revelio Labs salary estimates (ML-based, trained on visa data). Educational exposure measure derived from pre-2020 syllabi, representing baseline curricular differences rather than post-ChatGPT responses. Excludes graduates whose first job begins >3 years after degree completion. Some state-occupation cells have missing UI data (see Table S1). Connecticut data shows anomalies for certain occupations.