Why Does AI Adoption Differ So Much across Countries?

In a previous post, we documented a sizeable gap in artificial intelligence (AI) adoption between the U.S. and Europe. In early 2026, 43% of U.S. workers reported using generative AI for their jobs, compared with usage that ranged between 26% and 36% in the six European countries we surveyed.They were Europe’s four largest economies—Germany, the U.K., France and Italy—and its two leading digital economies—Sweden and the Netherlands. Data on firm adoption among 32 European nations tell a similar story: U.S. firms are more likely to use AI for production than those in all but a handful of European countries.

But documenting a gap is only the first step. The more interesting question: Why does a gap exist? AI tools like ChatGPT are available at similar prices on both sides of the Atlantic. So what is holding back European adoption? In this post, we dig into two potential explanations.This blog post and the previously mentioned post are based on a working draft (PDF). A version of that draft was presented at the Brookings Papers on Economic Activity (BPEA) Spring 2026 Conference. The final article will be published in the spring 2026 volume of the BPEA journal. Conference drafts, slides and presentation recordings are available on the Brookings website.

Who Works Where: The Role of Demographics and Firm Composition

A natural starting point is to ask whether Europeans and Americans are simply different in ways that would predict divergent adoption rates. We know, for instance, that younger workers, college-educated workers and workers in larger firms or in industries like information and communication technology adopt AI at much higher rates than others. If European countries have proportionally more older workers, fewer college graduates or a smaller tech sector, that alone might explain part of the gap.

To test this, we use a Oaxaca-Blinder decomposition, a standard statistical technique for breaking down differences between two groups into a part explained by observable characteristics and a part that remains unexplained. Think of it this way: If we gave European workers the same age, gender, education, occupation, industry and firm-size distribution as U.S. workers, how much of the adoption gap would disappear?

The answer is: a meaningful share, but far from all of it. On average across the six European countries we surveyed, compositional differences in age, education, occupation, industry and firm size account for about 55% of the U.S.-Europe adoption gap. Industry, occupation and firm size matter more than demographics. In Sweden, compositional differences statistically account for the entire gap: If Sweden had the same workforce composition as the U.S., it would have nearly identical AI adoption rates. But in Germany, France and Italy, a large share of the gap remains unexplained even after accounting for all these factors.

So composition matters, but it is not the whole story. Something else is going on.

Management Practices: A Surprisingly Powerful Predictor

A literature on the spread of information and communication technologies in the 1990s and 2000s found that differences in firm management practices were a key driver of why U.S. firms invested more in computers and software than European firms invested, and why U.S. firms got more out of those investments. Could the same be true for AI?

The first figure plots country-level AI adoption rates by firms for production processes (to be precise, for the production of goods and services) against management scores from the World Management Survey, an international study that measures management quality through structured interviews with firm managers (these particular scores are for those in the manufacturing sector).For details, see Nicholas Bloom, Renata Lemos, Raffaella Sadun, Daniela Scur and John Van Reenen’s 2021 manufacturing dataset, “World Management Survey,” at Harvard Dataverse. The AI usage data are derived from the European Union’s Information and Communication Technologies Usage and E-Commerce in Enterprises survey and the U.S. Census Bureau’s Business Trends and Outlook Survey.

The relationship is striking: Countries with higher-quality management practices have significantly higher rates of AI adoption among firms, with a correlation of 0.83 for AI adoption in production processes. The U.S. has among the highest management scores in the sample, and some of the lowest-scoring countries, such as Poland and Greece, have the lowest AI adoption rates.

Of course, a cross-country correlation based on a handful of observations can tell us only so much. To dig deeper, we turn to our worker surveys.

What Do Worker Surveys Reveal about Management Practices and AI Usage?

In our January-February 2026 survey, we asked each worker three questions motivated by questions from the World Management Survey about whether their employer rewards performance, promotes based on merit and addresses poor performance. We combined these into a worker-level management index and then asked: Do workers at firms that provide more incentives for performance, as reflected by a higher score, adopt AI at higher rates?

They do, strongly and consistently. The second figure shows that AI adoption rises steadily across the management index, with a one-standard-deviation increase associated with a 9.6 percentage point increase in the adoption rate. The relationship holds within every country we studied, not just across them.

In the figure, the colored dots represent the average national values for the U.S. and six European countries—France, Germany, Italy, the Netherlands, Sweden and the U.K.—and the gray dots represent values using the pooled data for all seven nations.

Encouragement, Tool Provisioning Strongly Predictive of AI Usage at Work

Why would management practices matter so much for AI adoption? Our surveys shed some light on the mechanism. Workers at higher-scoring firms are significantly more likely to report that their employers actively encourage them to use AI and provide them with access to AI tools. And it turns out that encouragement and tool provision are among the strongest predictors of whether a worker adopts AI, far more so than, say, formal AI training. In fact, once we account for whether employers encourage and equip their workers to use AI, the management index itself no longer adds much additional predictive power. Better-managed firms do not just set performance incentives and wait for workers to figure things out on their own. They actively promote AI adoption.

This has meaningful implications for the cross-country gap. When we redo our Oaxaca-Blinder decomposition and add employer AI encouragement to the mix, the unexplained gap shrinks dramatically. In five of the six European countries we surveyed, more than 95% of the U.S.-Europe adoption gap is statistically accounted for by differences in workforce composition and employer encouragement combined. U.S. employers are significantly more likely to encourage their workers to use AI, to provide access to AI tools and to create a workplace environment in which adoption is the norm rather than the exception.

What Does This Mean?

Our findings suggest that the AI adoption gap between the U.S. and Europe may not simply be an inevitable consequence of geography or culture. Management practices and whether firms actively encourage AI use among their workers are strongly associated with adoption rates. Understanding why these patterns exist and whether they reflect deeper organizational differences between U.S. and European firms are important questions for future research.

If management practices do turn out to be a key driver of adoption, then narrowing the U.S.-Europe adoption gap may require first narrowing the management gap.

Notes

1. They were Europe’s four largest economies—Germany, the U.K., France and Italy—and its two leading digital economies—Sweden and the Netherlands.
2. This blog post and the previously mentioned post are based on a working draft (PDF). A version of that draft was presented at the Brookings Papers on Economic Activity (BPEA) Spring 2026 Conference. The final article will be published in the spring 2026 volume of the BPEA journal. Conference drafts, slides and presentation recordings are available on the Brookings website.
3. For details, see Nicholas Bloom, Renata Lemos, Raffaella Sadun, Daniela Scur and John Van Reenen’s 2021 manufacturing dataset, “World Management Survey,” at Harvard Dataverse.