---
title: "The AI Jobs Apocalypse for New Graduates Is, So Far, Not in the Data"
description: "Summer 2026 unemployment for recent college graduates was 7.3%, well within the range of recent years, new research finds"
author: "Nate Ledger"
published: 2026-09-25T19:11:05Z
modified: 2026-09-26T17:04:18Z
url: https://rews.cc/a/the-ai-jobs-apocalypse-for-new-graduates-is-so-far-not-in-th-044d99
language: en
tags: ["ai", "jobs", "economy", "unemployment", "workforce", "us"]
publisher: "Rews (https://rews.cc)"
---

# The AI Jobs Apocalypse for New Graduates Is, So Far, Not in the Data

*Summer 2026 unemployment for recent college graduates was 7.3%, well within the range of recent years, new research finds*

By Nate Ledger · September 25, 2026 · https://rews.cc/a/the-ai-jobs-apocalypse-for-new-graduates-is-so-far-not-in-th-044d99

## In brief

- A CESifo working paper finds summer 2026 unemployment of 7.3% for recent college graduates, within recent years’ range
- The 2022-2024 range ran from 6.3% to 7.8%; 2022 was the year ChatGPT launched
- Comparisons with non-graduates, older graduates and AI-exposed roles showed no statistically significant divergence
- The result contrasts with a Stanford study using ADP payroll data, which counts jobs rather than job-seekers
- Researchers caution the class of 2027 and later could feel more impact as workplace AI use deepens

By now there is a settled story about artificial intelligence and the labor market, and it goes like this: AI is coming for entry-level white-collar work first, so the class of recent college graduates is the canary in the coal mine. The concern that companies are using generative AI to cut back on entry-level hiring is widespread enough that, as [TechSpot put it](https://www.techspot.com/news/113997-ai-job-apocalypse-hasnt-hit-recent-college-grads.html), the question is mostly how bad the damage is. A much-discussed Stanford study, built on payroll data from the HR firm ADP, seemed to find exactly the expected damage. But a new working paper from researchers at CESifo goes and looks at a different dataset — the Census Bureau’s Current Population Survey — and finds, well, nothing.

The headline numbers are almost boring. The summer 2026 unemployment rate for young college graduates was 7.3%. In previous years the figure ranged from 6.3% in 2022 — the year ChatGPT was released — to 7.8% in 2024. So 2026 sits comfortably inside the recent band. The same held when the researchers widened the lens to include graduates who told the CPS they “want a job” but weren’t actively looking, and so don’t count as officially unemployed: unremarkable again.

Maybe the averages hide the pain, so the researchers poked at the data. They compared recent graduates with non-college workers of the same age, and with older college graduates aged 30 to 49. They also split jobs by their potential “AI exposure,” based on a 2023 study of which roles AI systems are best equipped to do — the idea being that if AI is eating entry-level work, the eating should show up first in the most exposed occupations. In almost all of these comparisons, the differences in trends between 2022 and 2026 were not statistically significant. The data, the researchers wrote, “tell a consistent story in which unemployment among recent college graduates in summer 2026 was not unusually high relative to earlier summers.”

How can that square with the Stanford study finding nearly the opposite? Here’s the mechanism, and it’s genuinely interesting. The ADP payroll data counts jobs — the total supply of positions in various fields, as seen by a big HR processor that covers a decent but incomplete slice of the economy. The CPS unemployment rate counts people: graduates who want work and can’t find it, which reflects not just the supply of jobs but the aggregate demand for labor. Those two things can move independently. If AI contracts the supply of entry-level jobs in, say, software, but the economy’s demand for young graduates holds up — they get hired into other roles — then a payroll dataset shows a decline in coding jobs while an unemployment dataset shows nothing at all. Both studies can be right, and the economy simply reshuffles.

The CESifo authors are careful not to overclaim. They call the summer 2026 data a “useful first test” of how accelerating AI use is — or isn’t — hitting the job market, and they warn that “if the intensity of AI use in the workplace continues to increase, the graduating classes of 2027 and later might be more affected than the class of 2026,” with more years of data needed to know whether effects emerge as workplace AI deepens. Which is fair. The claim was never really “AI has already wiped out the class of 2026” so much as “AI will get the entry level eventually.” Eventually has not arrived. Check back next summer.