---
id: "lib-2026-09-00-graduating-into-disruption-labor-market-outcomes"
type: "library"
title: "Graduating Into Disruption: Labor Market Outcomes for AI-Exposed College Majors (CES WP 26-56)"
lang: "EN"
org: "US Census Bureau (CES)"
source_type: "Paper"
date: "2026-09-01"
week: "2026-W44"
url: "https://nosov.ai/library/lib-2026-09-00-graduating-into-disruption-labor-market-outcomes/"
original_url: "https://www2.census.gov/library/working-papers/2026/adrm/ces/CES-WP-26-56.pdf"
tier: "A"
tier_name: "Strong evidence"
tier_reason: "US Census Bureau economists using linked administrative records on 6.7 million graduate records, with an event-time design, alternative exposure measures and robustness checks against remote work and graduate supply."
summary_basis: "full-text"
original_minutes: "75"
tags: ["future of work", "graduate labour market", "ai exposure", "administrative data"]
---

# Graduating Into Disruption: Labor Market Outcomes for AI-Exposed College Majors (CES WP 26-56)

Tier A (Strong evidence): US Census Bureau economists using linked administrative records on 6.7 million graduate records, with an event-time design, alternative exposure measures and robustness checks against remote work and graduate supply.

Summary based on the full text.

## In short

A Census Bureau Center for Economic Studies working paper (CES WP 26-56, September 2026) by Cody Orr, Lee C. Tucker and Lawrence Warren. It assigns AI exposure to college majors from the jobs their graduates held before ChatGPT, then follows new graduates through linked employer records. From late 2022, graduates in the most exposed decile of majors (mostly computer science, information systems and other software-related fields) became less likely to find work right after graduation and earned less when they did. The authors put the earnings loss in the same range as graduating into a large recession.

## Key numbers

- 5 percentage points: drop in initial employment for the most AI-exposed decile of majors after ChatGPT, against the bottom six deciles
- 13%: decline in full-quarter initial earnings for the most exposed decile, attenuating to about 5% after two years
- about half: share of the earnings decline explained by a shift into lower-paying sectors such as retail and food service
- nearly six percentage points: contraction in the share of top-decile graduate jobs in Professional, Scientific and Technical Services
- 29%: share of US bachelor's degrees conferred 2016-2024 covered by the sample

## What it reports

- Top-decile majors saw a five percentage point fall in the probability of employment in the quarter after graduation, relative to the bottom six deciles.
- Initial quarterly earnings fell about 13%, attenuating to about 7% after one year and about 5% after two years.
- About half of the earnings loss came from sector mix. Information lost over three percentage points of exposed graduates' employment and Professional, Scientific and Technical Services nearly six; Accommodation and Food Services and Retail Trade each gained more than two.
- Deciles seven and eight also show smaller, statistically significant employment declines, about two percentage points.
- Graduate supply in the top decile rose sharply before 2022, yet employment kept pace until the post-ChatGPT period, so supply growth alone does not explain the drop.
- Self-employment (1.01% to 1.49%) and graduate school enrollment (16.20% to 17.86%) rose for top-decile graduates from 2022 to 2024, which may account for up to half of the employment decline; the authors read this as a response to weak demand.

## How they got it

Graduate rosters from the Post-Secondary Employment Outcomes (PSEO) data, 27 partners in 24 states covering more than 350 institutions, linked to LEHD employer-employee records. Major-level exposure combines the Eloundou et al. (2024) task scores with each major's pre-period occupation mix from the American Community Survey. An event-time regression compares graduation cohorts across exposure deciles around late 2022. Results hold under alternative exposure measures (including the usage-based measure of Handa et al.) and controls for work-from-home intensity.

## Limits

- The institution sample skews to large, public, urban research universities; effects may differ elsewhere.
- Bachelor's degrees only.
- The post-ChatGPT window is short, so persistence and recovery cannot yet be assessed.
- The design identifies differential outcomes by exposure, not a single mechanism: lower demand for entry tasks, changed hiring and weaker degree signals are all possible.
- A working paper that has not had the Census Bureau's publication review.

## Why it matters for companies

This is the clearest administrative evidence so far that the entry-level pipeline in software-adjacent fields has narrowed since 2022, and that it shows up as fewer first jobs and lower-quality first jobs rather than unemployment. Firms that cut graduate hiring because tools cover junior tasks should expect a thinner pool of experienced staff in a few years. Studies that condition on people already employed in exposed occupations will miss this effect, which is worth remembering when reading reassuring headline unemployment numbers.

Original: https://www2.census.gov/library/working-papers/2026/adrm/ces/CES-WP-26-56.pdf