Disparities in AI-Related Tasks? Evidence from Job Ads in Germany

AI applications are expected to transform work primarily through the automation of non-routine cognitive tasks (Felten et al., 2023; Eloundou et al., 2024). In the US, firms using AI increased hiring in AI-exposed positions at the expense of non-AI-exposed positions (Acemoglu et al. 2022). A lack of skilled workers is the biggest barrier to AI adoption according to a survey among firms in Germany (Krzywdzinski, 2024). However, little is known about the heterogeneity within AI-exposed jobs. Which job candidates are expected to i) follow instructions provided by AI, ii) use AI as an efficiency tool, or iii) program and modify machine learning models? Within occupations, do AI-related positions differ structurally from non-AI-related positions?

This study analyses how AI technologies are embedded in job advertisements using data from over 80 million online ads in Germany from 2020–2024. Based on the job descriptions in the ads, the study examines how AI-related skills are distributed across occupations and requirement levels. We compare ads with and without AI content within the same occupation to understand how AI might change the composition of tasks.

We build on the task and skill framework of Rodrigues et al. (2021) and model skills and tasks through relational annotation. We then apply advanced NLP methods to automatically extract contextualised job ad content. This method allows us to distinguish where AI is embedded as a tool from where it is embedded as a work content, and to classify tasks by complexity or standardisation.

Our findings provide a nuanced picture of AI-related jobs. By drawing attention to the heterogeneity within the differential application of AI at work, the study contributes to the debates about the transformative potential of new digital technologies and their role in perpetuating or mitigating inequalities in job content or work standardisation.