Uneven AI adoption and persistent gender gap in global workforce

New evidence suggests AI adoption remains uneven and its early impact on jobs may be more limited than widely assumed. A large cross-country study led by Golo Henseke finds that while workers in AI-exposed roles are adopting the technology at higher rates, the broader transformation of work itself is still in a transitional phase, with few measurable changes to job tasks so far.

The study, titled "Generative AI at Work: From Exposure to Adoption across 35 European Countries" and published as a working paper on arXiv, is based on survey data from more than 36,000 workers across 35 countries to examine who is adopting AI, under what conditions, and whether this adoption is reshaping work.

While the study focuses on Europe, its findings reflect broader global trends, making the region a useful example of how AI is diffusing unevenly across economies, industries, and worker groups.

Uneven adoption reveals deeper structural divides

The study finds that only about 12 percent of workers reported using generative AI at work in 2024, though adoption varied sharply across countries, occupations, and demographic groups. In some countries, fewer than 3 percent of workers used AI tools, while in others adoption approached 25 percent. This wide variation highlights how national conditions such as digital infrastructure, training systems, and workplace practices shape how quickly new technologies spread.

This uneven diffusion is not simply a matter of access. The research shows that occupational exposure to AI, meaning how much a job's tasks can be supported or automated by AI systems, is a strong predictor of adoption. Workers in highly exposed roles, such as information technology, legal, or professional services, are far more likely to use AI tools than those in routine or manual occupations.

However, exposure alone does not determine adoption. Even within the same occupation, some workers adopt AI while others do not. The study finds that individual characteristics such as education level, skill development, and job complexity significantly influence whether workers take up AI tools.

Workers with higher education levels and those engaged in non-routine cognitive tasks are more likely to adopt AI, suggesting that the technology complements analytical and problem-solving work rather than replacing it outright.

Organizational factors also play a critical role. Employees who have greater influence over workplace decisions and more involvement in how their work is structured are significantly more likely to adopt AI. This indicates that adoption is often driven from the bottom up, rather than imposed solely through top-down corporate strategies.

A persistent gender gap further complicates the adoption landscape. Men are more likely to use AI tools than women, particularly in occupations with high exposure to AI capabilities. The gap is not explained by differences in education or job roles, suggesting deeper behavioral or institutional factors at play.

National systems shape how AI spreads

The study further points to the importance of national-level conditions in shaping AI adoption. Countries with higher levels of digital integration, measured by widespread computer use in workplaces, show significantly higher adoption rates.

Workplace training systems emerge as a key driver. Countries where workers regularly receive training, whether employer-provided or self-funded, show stronger links between AI exposure and actual adoption. This suggests that general learning systems, rather than AI-specific training programs, play a critical role in enabling workers to integrate new technologies.

The cognitive complexity of jobs across an economy also matters. Nations with a higher share of jobs involving analytical, problem-solving, and knowledge-intensive tasks are better positioned to convert AI exposure into real adoption.

Interestingly, traditional indicators such as GDP per worker or overall education levels do not strongly predict AI adoption once digital infrastructure is accounted for. This finding challenges assumptions that wealth alone determines technological uptake, pointing instead to the importance of institutional readiness and workforce capabilities.

The study suggests that AI diffusion follows a layered process. First, digital infrastructure must be in place. Then, workforce skills and organizational practices determine whether exposure to AI translates into actual use. Without these enabling conditions, even highly exposed occupations may show low adoption rates.

Early adoption has not yet reshaped jobs

The study finds little evidence that it has fundamentally changed the structure of jobs so far. While individual users report that technology has both removed and created tasks, these changes do not appear to be directly caused by AI adoption at the aggregate level. When the researchers account for differences across occupations and countries, the apparent link between AI use and task changes largely disappears. This suggests that workers in already evolving roles are more likely to adopt AI, rather than AI itself driving those changes.

The findings point to a transitional phase in which AI is being integrated into existing workflows rather than transforming them. Workers may use AI tools to complete tasks more efficiently, but the overall structure of their jobs remains largely unchanged. This aligns with broader economic evidence showing limited short-term impacts of AI on wages, working hours, or productivity. The study suggests that the benefits of AI may take time to materialize, as organizations and workers adapt to new ways of working.

The concept of a "productivity lag" helps explain this pattern. New technologies often require significant learning, experimentation, and organizational change before their full impact becomes visible. In the case of generative AI, this process is still underway.

A transition phase with long-term implications

The study's findings highlight a critical moment in the evolution of AI in the workplace. Adoption is accelerating, but its effects are not yet fully visible. This creates both opportunities and risks. On one hand, the lack of immediate disruption may provide time for policymakers and organizations to prepare for longer-term changes. Investments in training, digital infrastructure, and inclusive workplace practices could help ensure that AI benefits are widely shared.

On the other hand, the uneven nature of adoption raises concerns about growing inequalities. Workers in highly exposed, high-skill roles are more likely to benefit from AI, while others may be left behind. The persistent gender gap further underscores the risk of unequal outcomes.

The study also points to the importance of workplace design. Organizations that encourage employee involvement and provide opportunities for skill development are better positioned to harness AI effectively. This suggests that the future of work will depend not only on technological innovation but also on how work itself is organized.