German AI-skill ads were not linked to employment growth

What changed in the study

Peede and Stops link AI-skill requirements in German job ads to administrative establishment records. The 2024 working-paper abstract reported slightly higher growth in highly skilled jobs; the 2026 journal abstract no longer reports a skill-group employment association. These are versions of one study, not two independent findings. The latest full article was not accessible in this check, so we cannot tell which sample or model changes explain the difference.

Why this does not settle AI job loss

An ad asking for AI skills can signal development, implementation or use. It does not prove that the establishment deployed a particular tool, or identify what would have happened without AI. The long 2015–2023 window mixes earlier AI technologies with only the start of the generative-AI period.

A separate U.S. BEA working paper links reported AI use in 2025–2026 to state-industry employment paths ending in 2024. Employment differences were positive but imprecise. Because use was measured after most outcomes, those results cannot establish that AI produced job growth. They do not contradict the German result: place, unit, period and AI measure differ.

Current answer

This check does not change our conclusion. Observed AI-related hiring evidence remains mixed and narrow; the Copilot studies still count different outcomes. The 6.1 million U.S. vulnerability estimate is conditional exposure and recovery capacity, not observed layoffs. A stronger test would follow the same employers before and after verified AI deployment, with hires, separations, pay and skill mix measured together.