Research check · 10 November 2026
German entry-level employment fell in exposed work. AI causation is still open.
Full-time employment among German 20–24-year-olds in the most AI-exposed occupations fell 14% from 2022 Q3 to 2026 Q1. Employment across ages 20–64 in those occupations rose 1.8% over the same period. The age split is a warning, not a tally of jobs AI destroyed.
What the new report measured
Falck, Fuest and Guo's November EconPol report follows German social-security records covering roughly 22 million full-time workers in 140 occupations. In the most exposed fifth, 20–24-year-old employment fell from about 279,000 to 239,000. The authors' within-occupation comparison finds that by 2026 Q1 the young-to-older employment ratio was 5.5% lower per standard deviation of task exposure than before ChatGPT.
Across all ages studied, employment in the most exposed fifth rose 1.8%; it fell 6.4% in the least exposed fifth. Much of that broad divergence predates generative AI. The report also uses EU-27 household-survey data through 2024. Its estimate for 20–24-year-old new hires is 2.9% lower relative to older hires per standard deviation of automation-weighted exposure. That result is suggestive: only this highlighted EU hiring estimate is statistically significant, and the EU young-employment result is not.
What it cannot answer
The German decline of about 40,000 young workers in the most exposed fifth is not a count of AI layoffs. The report scores tasks that a language model could perform faster; it does not identify employers that adopted AI. Its Germany series excludes part-time workers and apprentices. Changes in schooling, remote work, demographics and demand could also shift young employment. The EU series is shorter and its new-hire measure counts workers who joined their employer in the prior 12 months.
An October German IAB analysis used ages 20–29 and a different comparison, finding a small adjusted exposure link not statistically distinguishable from zero. The studies are not direct replications; the difference is a reason to test the age band and design before attributing the new gap to AI.
The current answer
German and selected U.S. payroll data now both show a concentrated early-career warning. Australian young-worker outcomes look weaker too, but its job-finding flows have not clearly diverged; broader Norwegian and U.S. tracking remain quieter. These sources measure different places and outcomes. A worker can ask whether an employer is reducing junior entry or supervised practice, but none of these comparisons can give personal job-loss odds or justify paying for a course.