This site is a work in progress and has not been widely shared. Content may contain errors. Feedback is welcome.
This site is undergoing review. Some annotations were human-generated, some AI-generated — all are being verified.
Back to papers

Generative AI as Seniority-Biased Technological Change: Evidence from U.S. Resume and Job Posting Data

Lichtinger, Maasoum

2025Working paper7 citations
Observational labor marketCausal
LLM / Generative AIAI AdoptionJunior / entry-levelHuman-AI collaborationAugmentation vs. substitution
DOI: 10.2139/ssrn.5425555
Summary

Lichtinger and Hosseini use LinkedIn resume and job posting data (2015-2025) covering 62 million workers in 285,000 U.S. firms to study whether generative AI adoption constitutes seniority-biased technological change, identifying adopters via LLM-classified postings for AI integrator roles and employing difference-in-differences and triple-difference designs to compare within-firm employment dynamics by seniority level

Main Finding

Junior employment in AI-adopting firms declined 7.7% (DiD) to 12% (triple-diff) relative to non-adopters beginning in 2023Q1, driven primarily by a 22% reduction in junior hiring rather than increased separations, while senior employment continued rising; wholesale/retail trade saw the largest effects (~40% decline in junior hiring)

Primary Datasets

Revelio Labs

Labor outcomes

Lightcast (formerly Burning Glass)

Labor outcomes

U.S. resume data (~62M workers, 285K firms); GenAI job postings

Secondary Datasets

O*NET

Other

Eloundou et al. (2024) task-level LLM exposure scores (O*NET tasks)

Key Methods
Difference-in-differences and triple-difference design comparing employment dynamics by seniority (junior vs senior) in AI-adopting versus non-adopting firms; AI adoption identified via LLM-classified job postings for dedicated AI integrator roles
Sample Period
2015-2025
Geographic Coverage
US
Sample Size
284,974 firms; 156,765,776 employment positions from ~62 million unique workers; 245,838,118 job postings (198,773,384 with text); 10,599 AI adopters (3.7%)
Level of Analysis
Firm, Individual
Occupation Classification
SOC
Industry Classification
NAICS
Notes
SSRN; Stanford Digital Economy Lab; junior employment declines 15% in adopting firms; senior unchanged; coins seniority-biased TC [Claude classification]: Novel AI adoption identification using LLM-classified job postings for AI integrator roles; adopters represent only 3.7% of sample (10,599 firms); acknowledges selection concerns despite triple-diff design; U-shaped education pattern with mid-tier graduates most affected; increased promotions for remaining juniors; separation effects small relative to hiring effects [Claude classification]: Novel AI adoption identification using LLM-classified job postings for AI integrator roles; adopters represent only 3.7% of sample (10,599 firms); acknowledges selection concerns despite triple-diff design; U-shaped education pattern with mid-tier graduates most affected; increased promotions for remaining juniors; separation effects small relative to hiring effects [Claude classification]: Novel AI adoption identification using LLM-classified job postings for AI integrator roles; adopters represent only 3.7% of sample (10,599 firms); acknowledges selection concerns despite triple-diff design; U-shaped education pattern with mid-tier graduates most affected; increased promotions for remaining juniors; separation effects small relative to hiring effects [Claude classification]: Novel AI adoption identification using LLM-classified job postings for AI integrator roles; adopters represent only 3.7% of sample (10,599 firms); acknowledges selection concerns despite triple-diff design; U-shaped education pattern with mid-tier graduates most affected; increased promotions for remaining juniors; separation effects small relative to hiring effects [Claude classification]: Novel AI adoption identification using LLM-classified job postings for AI integrator roles; adopters represent only 3.7% of sample (10,599 firms); acknowledges selection concerns despite triple-diff design; U-shaped education pattern with mid-tier graduates most affected; increased promotions for remaining juniors; separation effects small relative to hiring effects [Claude classification]: Novel AI adoption identification using LLM-classified job postings for AI integrator roles; adopters represent only 3.7% of sample (10,599 firms); acknowledges selection concerns despite triple-diff design; U-shaped education pattern with mid-tier graduates most affected; increased promotions for remaining juniors; separation effects small relative to hiring effects [Claude classification]: Novel AI adoption identification using LLM-classified job postings for AI integrator roles; adopters represent only 3.7% of sample (10,599 firms); acknowledges selection concerns despite triple-diff design; U-shaped education pattern with mid-tier graduates most affected; increased promotions for remaining juniors; separation effects small relative to hiring effects [Claude classification]: Novel AI adoption identification using LLM-classified job postings for AI integrator roles; adopters represent only 3.7% of sample (10,599 firms); acknowledges selection concerns despite triple-diff design; U-shaped education pattern with mid-tier graduates most affected; increased promotions for remaining juniors; separation effects small relative to hiring effects [Claude classification]: Novel AI adoption identification using LLM-classified job postings for AI integrator roles; adopters represent only 3.7% of sample (10,599 firms); acknowledges selection concerns despite triple-diff design; U-shaped education pattern with mid-tier graduates most affected; increased promotions for remaining juniors; separation effects small relative to hiring effects [Claude classification]: Novel AI adoption identification using LLM-classified job postings for AI integrator roles; adopters represent only 3.7% of sample (10,599 firms); acknowledges selection concerns despite triple-diff design; U-shaped education pattern with mid-tier graduates most affected; increased promotions for remaining juniors; separation effects small relative to hiring effects [Claude classification]: Novel AI adoption identification using LLM-classified job postings for AI integrator roles; adopters represent only 3.7% of sample (10,599 firms); acknowledges selection concerns despite triple-diff design; U-shaped education pattern with mid-tier graduates most affected; increased promotions for remaining juniors; separation effects small relative to hiring effects [Claude classification]: Novel AI adoption identification using LLM-classified job postings for AI integrator roles; adopters represent only 3.7% of sample (10,599 firms); acknowledges selection concerns despite triple-diff design; U-shaped education pattern with mid-tier graduates most affected; increased promotions for remaining juniors; separation effects small relative to hiring effects [Claude classification]: Novel AI adoption identification using LLM-classified job postings for AI integrator roles; adopters represent only 3.7% of sample (10,599 firms); acknowledges selection concerns despite triple-diff design; U-shaped education pattern with mid-tier graduates most affected; increased promotions for remaining juniors; separation effects small relative to hiring effects [Claude classification]: Novel AI adoption identification using LLM-classified job postings for AI integrator roles; adopters represent only 3.7% of sample (10,599 firms); acknowledges selection concerns despite triple-diff design; U-shaped education pattern with mid-tier graduates most affected; increased promotions for remaining juniors; separation effects small relative to hiring effects [Claude classification]: Novel AI adoption identification using LLM-classified job postings for AI integrator roles; adopters represent only 3.7% of sample (10,599 firms); acknowledges selection concerns despite triple-diff design; U-shaped education pattern with mid-tier graduates most affected; increased promotions for remaining juniors; separation effects small relative to hiring effects [Claude classification]: Novel AI adoption identification using LLM-classified job postings for AI integrator roles; adopters represent only 3.7% of sample (10,599 firms); acknowledges selection concerns despite triple-diff design; U-shaped education pattern with mid-tier graduates most affected; increased promotions for remaining juniors; separation effects small relative to hiring effects [Claude classification]: Novel AI adoption identification using LLM-classified job postings for AI integrator roles; adopters represent only 3.7% of sample (10,599 firms); acknowledges selection concerns despite triple-diff design; U-shaped education pattern with mid-tier graduates most affected; increased promotions for remaining juniors; separation effects small relative to hiring effects [Claude classification]: Novel AI adoption identification using LLM-classified job postings for AI integrator roles; adopters represent only 3.7% of sample (10,599 firms); acknowledges selection concerns despite triple-diff design; U-shaped education pattern with mid-tier graduates most affected; increased promotions for remaining juniors; separation effects small relative to hiring effects [Claude classification]: Novel AI adoption identification using LLM-classified job postings for AI integrator roles; adopters represent only 3.7% of sample (10,599 firms); acknowledges selection concerns despite triple-diff design; U-shaped education pattern with mid-tier graduates most affected; increased promotions for remaining juniors; separation effects small relative to hiring effects [Claude classification]: Novel AI adoption identification using LLM-classified job postings for AI integrator roles; adopters represent only 3.7% of sample (10,599 firms); acknowledges selection concerns despite triple-diff design; U-shaped education pattern with mid-tier graduates most affected; increased promotions for remaining juniors; separation effects small relative to hiring effects [Claude classification]: Novel AI adoption identification using LLM-classified job postings for AI integrator roles; adopters represent only 3.7% of sample (10,599 firms); acknowledges selection concerns despite triple-diff design; U-shaped education pattern with mid-tier graduates most affected; increased promotions for remaining juniors; separation effects small relative to hiring effects [Claude classification]: Novel AI adoption identification using LLM-classified job postings for AI integrator roles; adopters represent only 3.7% of sample (10,599 firms); acknowledges selection concerns despite triple-diff design; U-shaped education pattern with mid-tier graduates most affected; increased promotions for remaining juniors; separation effects small relative to hiring effects [Claude classification]: Novel AI adoption identification using LLM-classified job postings for AI integrator roles; adopters represent only 3.7% of sample (10,599 firms); acknowledges selection concerns despite triple-diff design; U-shaped education pattern with mid-tier graduates most affected; increased promotions for remaining juniors; separation effects small relative to hiring effects [Claude classification]: Novel AI adoption identification using LLM-classified job postings for AI integrator roles; adopters represent only 3.7% of sample (10,599 firms); acknowledges selection concerns despite triple-diff design; U-shaped education pattern with mid-tier graduates most affected; increased promotions for remaining juniors; separation effects small relative to hiring effects [Claude classification]: Novel AI adoption identification using LLM-classified job postings for AI integrator roles; adopters represent only 3.7% of sample (10,599 firms); acknowledges selection concerns despite triple-diff design; U-shaped education pattern with mid-tier graduates most affected; increased promotions for remaining juniors; separation effects small relative to hiring effects [Claude classification]: Novel AI adoption identification using LLM-classified job postings for AI integrator roles; adopters represent only 3.7% of sample (10,599 firms); acknowledges selection concerns despite triple-diff design; U-shaped education pattern with mid-tier graduates most affected; increased promotions for remaining juniors; separation effects small relative to hiring effects