From Exposure to Adoption
The question now is whether occupational exposure becomes actual workplace use. This chapter examines Henseke’s evidence from the 2024 European Working Conditions Survey: how strongly a task-based exposure measure predicts generative-AI adoption, why that relationship differs among workers facing similar occupational potential, and what country-level conditions help turn potential into uptake. The result is a more realistic account of adoption than either “AI will spread everywhere” or “exposure determines who uses it.”
The distinction between the OECD capability map and this European study matters. The OECD compares occupational capability profiles with a broad AI frontier. Henseke uses a task-based occupational susceptibility measure to study whether workers report using AI at work. The measures address related questions, but they are not the same instrument. One maps potential capability proximity; the other observes a behavioural outcome and asks which conditions mediate it.
Exposure predicts use, but does not dictate it
Henseke’s data come from approximately 36,600 employed people surveyed face to face across 35 European countries in 2024. The survey’s adoption question asked whether respondents used artificial intelligence at work that simplified complex mental tasks or made recommendations, naming tools such as ChatGPT, LLaMA, DALL-E, Midjourney, and Jasper. Across the pooled workforce, 12% reported using AI for their job. The country range was wide: under 3% in Bosnia and Herzegovina and approximately a quarter in Luxembourg.
That range is already evidence against a purely technological explanation. If occupational exposure were enough to determine use, countries with similar occupational structures would show roughly similar adoption. Instead, the same general opportunity is embedded in different digital infrastructures, training systems, workplaces, and labour-market institutions. The survey also found a geographical gradient, with higher use concentrated in Northern and Western Europe than in Southern and Eastern Europe. The study is not claiming that geography itself causes adoption; it is showing that the environment in which exposure is encountered matters.
Occupational exposure nevertheless has a strong relationship with individual use. Adoption rises from 1.5% in the least exposed occupational quintile to nearly one quarter in the most exposed. In a model controlling for demographics and country fixed effects, a one-standard-deviation increase in occupational susceptibility is associated with a 10.1-percentage-point increase in the probability of reported adoption. That is a large gradient relative to the sample’s mean adoption rate.
The association weakens as the model accounts for more of the setting. Adding individual skills and computer use reduces the estimate to 6.6 percentage points. Adding employment relations, workplace characteristics, and industry fixed effects leaves 4.9 points. The reduction is informative: part of the raw relationship reflects who sorts into which jobs, industries, and digitally ready workplaces. The remaining association is still positive and precisely estimated, so exposure retains explanatory value after those observable differences are taken into account.
The gradient is also broad rather than confined to a handful of countries. It is positive and statistically significant in 32 of the 35 countries studied. In low-adoption countries, even workers in susceptible occupations rarely report use; in high-adoption countries, workers in exposed occupations drive much of the higher aggregate rate. That pattern means adoption is not merely a uniform increase in AI use across every job. It is more concentrated where the technology appears relevant, but the strength of that concentration depends on conditions that exposure alone does not capture.
The survey measure itself needs careful interpretation.