Researchers at the Institute for Employment Research (IAB) compared full-time employment trends across occupations with different levels of potential exposure to artificial intelligence. While figures differed for the 20–29 age group, the authors emphasize that this descriptive comparison does not prove that AI caused the divergence.
- Potential exposure indicates which tasks in an occupation could be performed faster using large language models and related software; it does not measure actual AI adoption or employee replacement.
- Between 2022 and 2025, full-time employment among 20–29-year-olds fell by 3.8% in high-exposure occupations and by 2.6% in lower-exposure occupations. The 1.2 percentage point gap does not establish causality.
- After controlling for other factors, the negative relationship between AI exposure and employment for this age cohort is statistically insignificant; cyclical economic vulnerability showed a stronger correlation.
What Potential AI Exposure Means
The metric used by the IAB assesses the share of an occupation’s tasks that large language models (LLMs) or tools built on them could perform substantially faster. This reflects technical potential at the occupational level. It does not indicate whether individual employers actually deploy these tools or whether jobs were eliminated as a result.
Comparing Employment Among Young Workers
According to the study, the number of full-time employees aged 20–29 fell by 3.8% in high-exposure occupations between 2022 and 2025, compared to a 2.6% decrease in lower-exposure occupations. The difference stands at 1.2 percentage points. This is a descriptive comparison between two occupational groups: it does not isolate the impact of AI from all other structural differences between them and therefore does not prove causality.
In a model simultaneously controlling for potential AI exposure, telework suitability (Homeoffice), and an occupation’s vulnerability to economic downturns, the negative correlation between AI exposure and the employment of 20–29-year-olds remains statistically insignificant. Put simply, the model cannot reliably distinguish the observed relationship from random noise. However, this finding also does not prove that AI has no impact whatsoever.
Data and Study Limitations
The analysis relies on monthly Federal Employment Agency (Bundesagentur für Arbeit) data from January 2021 through December 2025. It covers an average of approximately 22.3 million full-time employees subject to mandatory social security contributions aged 20–64. Civil servants (Beamte), the self-employed, individuals working solely in marginal employment (Minijobs), vocational trainees (Auszubildende), military personnel, and the unemployed were excluded. The authors use ages 20–29 as a proxy for early career stages rather than an exact marker of first-time job market entry.
The model also accounts for occupational and industry differences, as well as temporal and seasonal fluctuations. The correlation between telework capability and employment was slightly positive but statistically insignificant. A clearer negative correlation was observed for cyclical vulnerability: young worker employment in occupations tied to economically struggling industries dropped by 4.2% over 2022–2025.
The authors highlight a statistically significant negative association between higher AI exposure and employment trends specifically in manufacturing and non-academic occupations. No such correlation was observed in services or academic fields. These are findings from specific occupational subsets, not a sweeping conclusion regarding all young workers. The authors characterize the study as descriptive, noting that confirming causal relationships requires linking administrative employment records with directly observed workplace AI utilization.