For most of the last 35 years a new college graduate was less likely to be unemployed than the average worker. That edge is gone. Since recent graduates have been out of work more often than workers overall, and in the gap reached its widest since the New York Fed's series begins in 1990. The question is how much of that is AI. One payroll study finds young workers in AI-exposed jobs falling well behind; several survey-based studies find little sign of AI in the numbers yet. Below are the data and both sides, with what each can and cannot show.
Unemployment rate, recent college graduates and all workers
Percent, seasonally adjusted, three-month moving average. Below: recent graduates minus all workers, in percentage points.
Two more ways to look at it
The New York Fed series are built from the Current Population Survey. Here is the raw CPS for everyone aged 20 to 24, degree or not, against 25 to 34 year-olds and all workers; and the share of recent graduates working in jobs that do not need a degree.
Unemployment by age, all education levels
Percent, seasonally adjusted, monthly. BLS did not publish October 2025.
Underemployment of college graduates
Share working in jobs that typically do not require a degree, percent.
By major
Recent graduates (22 to 27) by field, from the 2024 American Community Survey. Computer science and computer engineering, the fields most often named in the AI argument, sit near the top for unemployment and near the bottom for underemployment: when their graduates find work, it is usually a degree job. Click a column to sort.
Major
Unemployment
Underemployment
Median wage, early
Median wage, mid
Grad degree
The New York Fed replaces this table each year and does not keep earlier editions on its site, so I can only show the current one. A year-over-year change by major would need the previous release, which I do not have from the publisher.
What the studies say
Four groups looked at whether AI is behind this, with different data and different questions. They are less opposed than the headlines made them sound: the payroll study finds no economy-wide effect either, and its signal is about hiring of 22 to 25 year-olds in particular occupations, which unemployment surveys are too small to see clearly.
Canaries in the Coal Mine?
Brynjolfsson, Chandar and Chen, Stanford Digital Economy Lab. August 2025, revised August 2026.
Finds a young-worker gap
Data
ADP payroll records covering millions of U.S. workers through June 2026; occupations sorted into five AI-exposure groups (Eloundou et al. 2024).
Findings
Employment of 22 to 25 year-olds in the two most exposed groups fell about 11% from November 2022 to June 2026, while the same age group in the three least exposed grew about 10%. Their employment now stands 19% below where it would be had it kept pace (pp. 1, 3, 10–11; Fig. 3).
Earlier versions reported a 13% relative decline (July 2025 data) and 16% (September 2025 data) from regressions with firm controls; by the simpler measure the shortfall was 15% in July 2025 (p. 3).
No economy-wide displacement: all ADP employment rose about 6% over the period, the most exposed group about 4% (p. 8, Fig. 1). Workers 22 to 25 overall were about 2% below November 2022; workers 35 to 40 were up 11% (p. 8, Fig. 2).
The decline runs through less hiring rather than more separations, and shows up where AI use substitutes for tasks rather than complements them (p. 3).
Limits
Descriptive, not causal, by the authors' own account (p. 4). The gap shrinks when controlling for occupations' education levels, some divergence predates ChatGPT, and it is weaker in the Census's American Community Survey: a 2022 to 2024 gap of −0.022 (interval −0.055 to +0.011) against −0.132 in ADP, though the two agree closely in professional, information and financial services (p. 28). ADP data are not public.
AI and Jobs: The Final Word (Until the Next One)
Sarah Eckhardt and Nathan Goldschlag, Economic Innovation Group. August 10, 2025.
Little sign yet
Data
Current Population Survey; five AI-exposure measures, Felten et al. (2021) as the main one.
Findings
From 2022 to early 2025 unemployment for the most exposed fifth of workers rose 0.30 points; for the least exposed it rose 0.94 points (Fig. 1).
Unemployment crept up for young workers and recent graduates whether or not their jobs were exposed (Fig. 6).
Two of five measures, both built with language models, show a small relative rise for exposed workers, about 0.2 and 0.3 points (Fig. 8).
Limits
Unemployment does not capture someone who was never hired into an exposed job. The main exposure measure does not separate automation from augmentation, and exposed workers differ from others in education and pay, as the authors note.
Evaluating the Impact of AI on the Labor Market
Martha Gimbel, Molly Kinder, Joshua Kendall and Maddie Lee, Yale Budget Lab. October 1, 2025. Follow-up by Ryan Nunn, May 7, 2026.
Little sign yet
Data
CPS microdata: how fast the mix of occupations changes, compared with the arrival of computers and the internet.
Findings
The occupational mix is changing a little faster than in past technology waves, about 1 point above the internet era's pace, and the speed-up began before generative AI.
The difference between the jobs held by recent graduates (20 to 24) and by 25 to 34 year-olds has mostly stayed within 30 to 33% since 2021, with some recent upward drift the authors call consistent with Brynjolfsson et al. or with a slowing market.
The 2026 follow-up uses synthetic difference-in-differences and finds AI's estimated effect on employment and wages in exposed occupations close to zero and not statistically distinguishable from it.
Limits
Small samples for young graduates (about 1,100 observations a year, by the authors' count). A measure of occupational churn is not built to catch slower hiring into the same occupations.
No Country for Young Grads
Gad Levanon, Matt Sigelman and others, Burning Glass Institute. July 2025.
Finds fewer junior postings
Data
Job postings plus CPS.
Findings
Unemployment for 20 to 24 year-olds with at least a bachelor's rose from 5.2% in 2018–19 to 6.2% in the two years through June 2025 (p. 5).
In computer, business and legal occupations, relative demand in postings for people with three years' experience or less fell sharply and stayed down, while demand for six-plus years held or grew (Chart 7, p. 10).
AI is one of four drivers they name, with lean post-pandemic staffing and a growing supply of graduates.
Limits
Postings are not hires. The report argues a structural story but does not separate AI from the other three drivers.
The New York Fed's own series do not test for AI. They show who is unemployed, not why.
Sources
Federal Reserve Bank of New York. The Labor Market for Recent College Graduates. Unemployment and underemployment data, ; outcomes by major, . Built from Census/BLS Current Population Survey and American Community Survey microdata (IPUMS) and O*NET. newyorkfed.org/research/college-labor-market
U.S. Bureau of Labor Statistics. Current Population Survey, unemployment rate, seasonally adjusted: ages 20–24 (LNS14000036), 25–34 (LNS14000089), 16 and over (LNS14000000). Public Data API v2, retrieved . bls.gov/cps
Erik Brynjolfsson, Bharat Chandar and Ruyu Chen. Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence. Stanford Digital Economy Lab, August 2025, revised August 12, 2026. Page numbers above refer to the August 2026 PDF. digitaleconomy.stanford.edu/publications/canaries-in-the-coal-mine
Sarah Eckhardt and Nathan Goldschlag. AI and Jobs: The Final Word (Until the Next One). Economic Innovation Group, August 10, 2025. eig.org/ai-and-jobs-the-final-word
Martha Gimbel, Molly Kinder, Joshua Kendall and Maddie Lee. Evaluating the Impact of AI on the Labor Market: Current State of Affairs. The Budget Lab at Yale, October 1, 2025. budgetlab.yale.edu
Ryan Nunn. AI Is Probably Not (Yet) the Reason for Labor Market Weakening. The Budget Lab at Yale, May 7, 2026. budgetlab.yale.edu
Gad Levanon, Matt Sigelman, Mariano Mamertino, Mels de Zeeuw and Gwynn Guilford. No Country for Young Grads: The Structural Forces That Are Reshaping Entry-Level Employment. Burning Glass Institute, July 2025. burningglassinstitute.org/research/no-country-for-young-grads