When Capability Meets Concern

The European workplace evidence leaves one important part of the conversion problem unresolved: workers in the same broad occupation do not turn exposure into adoption at the same rate. This chapter examines the gender pattern and then adds a behavioural condition that capability measures alone miss—how safe, legitimate, and socially acceptable people judge generative AI to be. The aim is not to explain unequal use as a fixed gender preference, but to separate occupational sorting, within-occupation conversion, personal and work use, predictive evidence, and causal claims.

The workplace gap is concentrated where exposure is highest

Across the 35 European countries in the 2024 survey, men were 4.1 percentage points more likely than women to report using AI at work: 13.8% compared with 9.7%. Adjusting for demographics leaves a similar gap. Adding education and seniority makes the difference larger rather than smaller, suggesting that women’s observed human-capital characteristics positively predict use and initially mask part of the residual gap. Adding occupation, computer use, job content, employment status, industry, and workplace size reduces the estimate again, but it remains about four percentage points.

The average is only the beginning of the result. When the analysis interacts gender with occupational exposure and absorbs broad occupation effects, the male advantage in the least exposed quintile is small and statistically indistinguishable from zero. In the most exposed quintile, men are 7.5 percentage points more likely to adopt than women in the same broad occupational setting. The gender difference is therefore not distributed evenly across the exposure scale. It becomes most visible where the potential gains, pressures, and expectations associated with generative AI are likely to be greatest.

This pattern limits a familiar explanation. If men and women simply held different kinds of jobs, occupational sorting could account for the aggregate gap. Occupation fixed effects do not remove every difference between people in a two-digit occupational group, but they make the comparison more demanding: workers share the same broad measure of occupational susceptibility, and the remaining contrast concerns how that potential is converted into reported use. The evidence is consistent with a differential propensity to adopt within exposed work, not merely with men being located in more exposed jobs.

The word “propensity” is important. The data do not reveal a single reason for the within-occupation difference. Education, confidence, previous technology use, employer rules, task allocation, workplace influence, and perceived risk may all be related to it. The European study’s joint moderator model finds a gender penalty even after accounting for observed skill, job-content, and organisational variables. That tells us the measured factors do not explain the whole gap; it does not identify which unmeasured mechanism is responsible.

The concentration in highly exposed work makes the gap consequential even if it later narrows. The most exposed jobs are where experimentation may shape emerging norms and where new AI-related skills could become useful.