Since the mid-1990s, the US has consistently outpaced Europe in productivity growth. From 1995 until 2025, productivity in the US has grown by almost 90% but only by around 30% in the euro area (Figure 1). This trend has largely been attributed to the faster diffusion of information and communication technologies (ICT) in the US than in Europe (Bloom et al. 2012). The Draghi Report argued that the emergence of artificial intelligence (AI) as a potentially transformative technology, “has opened [a window] for Europe to redress its failings in innovation and productivity” (Draghi 2024).

Figure 1 Output per hour worked in the US and Europe

Notes: Figure shows quarterly output per hour worked for the US (data from Bureau of Labor Statistics) and for the Euro area 20 countries (data from European Central Bank since 1995). European data before 1995 are annual output per hour worked for the EU12 countries, excluding UK (data from the Penn World Tables). EU12 excluding UK: Belgium, Denmark, France, Germany, Greece, Ireland, Italy, Luxembourg, Netherlands, Portugal, Spain. Euro area (20): EU 12 excluding Denmark and the UK, plus Austria, Croatia, Cyprus, Estonia, Finland, Latvia, Lithuania, Malta, Slovakia, and Slovenia.

Two critical questions arise:

1. How rapidly is AI adopted across countries (Bencivelli et al. 2026, Yotzov et al. 2026)?
2. What are the implications for productivity growth and employment (Albanesi et al. 2023, Aldasoro et al. 2026)?

We answer these questions in a new paper (Bick et al. 2026).  We draw on two types of surveys. First, building on the US Real-Time Population Survey (Bick et al. 2024), in May-June 2025 and January-February 2026, we surveyed workers in seven countries (France, Germany, Italy, Netherlands, Sweden, UK, US) to track AI adoption by workers. Second, we use the European Union’s ICT Usage and E-Commerce in Enterprises Survey (EU-ICT-Firm), which asks firms in 32 European countries whether they use any of eight specific AI technologies. 

The US is leading in AI adoption, large variation within Europe

Figure 2 shows that in the US, 43% of workers report using generative AI for work in January-February 2026. Adoption rates in European countries are substantially lower, ranging from 26% (Italy) to 36% (UK). In addition to differences in the share of workers who adopt AI, countries also differ in how intensively workers use AI. The bottom panel shows that 5.2% of work hours in the US are spent using AI, while the share in Germany, France, and Italy is less than one-third of this.

Figure 2 Share of workers and work hours using generative AI in 2026

Notes: The upper panel shows the share of survey respondents who use generative AI for work. The lower panel reports the share of last week’s working hours spent using generative AI. For daily usage, respondents could indicate (i) no usage, (ii) less than 15 minutes, (iii) 15–60 minutes, (iv) 1–4 hours, or (v) more than 4 hours. To obtain the total minutes spent using generative AI, we assume daily usage of 0, 7.5 minutes, 37.5 minutes, 2.5 hours, and 4 hours for each option, respectively. Weekly usage is calculated by combining daily usage with the reported number of days used. If respondents report using generative AI on ‘some days’ (rather than one or all days), we assume they used it on half of their working days. For non-users, the share of work hours using generative AI is mechanically 0.
Data source: Authors’ survey, run in January-February 2026. N=20,916.

A similar picture emerges in the firm data, with a large cross-country correlation between AI adoption by workers and firms in the countries covered by both surveys. The upper panel of Figure 3 shows the share of firms using AI in 2025. In the 32 European countries covered, on average 20% of firms used AI in 2025. The range is wide, spanning from over 35% in Scandinavian countries to less than 10% in Eastern European countries.

Figure 3 Share of firms in Europe using AI in 2025

Notes: The upper panel reports the share of firms using at least one of eight specified AI technologies: (1) text mining, (2) speech recognition, (3) natural language generation, (4) image recognition and processing, (5) machine learning for data analysis, (6) AI-based software robotic process automation, (7) autonomous robots or drones, self-driving vehicles, (8) generating pictures, videos, sound/audio. The lower panel correlates firms’ AI use with 2024 GDP per capita (PPP, current international $). We exclude Ireland and Luxembourg since their GDP per capita exceeds that of Norway, the country with the third highest GDP per capita in our sample, by around 30% and 50%, respectively. The adoption rate is 20% for Ireland and 34% for Luxembourg.
Data sources: 2025 EU-ICT-Firm Survey, World Bank Open Data.

The bottom panel of Figure 3 provides a simple summary of this pattern. Similarly to earlier ICT technologies like computers and the internet, richer countries tend to be the first adopters, with broader diffusion following. However, in the relatively short period that AI use by firms has been elicited in the EU-ICT-Firm Survey (since 2021), AI usage is diverging: the early adopters such as the Netherlands or Sweden experience stronger growth rates than the laggards (Figure 4).

Figure 4 Share of firms in Europe using AI over time

Notes: Figure shows the share of firms using at least one of eight specified AI technologies, except for the UK, where the survey asks generally whether they use AI technologies. 10th and 90th percentiles refer to percentiles among the 32 countries covered in the EU-ICT-Firm Survey.
Data sources: 2021, 2023, 2024, and 2025 EU-ICT-Firm Survey, 2023 UK Management and Expectations Survey.

Management practices explain large share of differences in adoption rates

What can explain the large differences in AI adoption rates? We turn back to our detailed worker surveys to understand why a technology that is available at similar prices across countries and requires only minor, if any, upfront investments is used at such different rates.

In a first step, we assess to what degree differences in worker characteristics (education, age, sex) and differences in industry and occupation composition and firm size explain why European countries are lagging compared to the US. For instance, US workers are on average younger and more educated, and these demographic groups are more likely to use AI. An Oaxaca-Blinder decomposition reveals that these factors do matter: compositional differences can explain about half of the gap between Europe and the US. But this also means that a large share remains unexplained.

We then explore the role of management practices. Research on earlier ICT technologies highlighted the close link between management practices and ICT investments and returns to investments (e.g. Bloom et al. 2012, Bresnahan et al. 2002). Using country-level data from the World Management Survey, we find that management scores are strongly predictive of firm-level adoption. We then leverage questions on management practices, specifically on how worker performance is incentivised, from our worker survey to explore whether management practices help to explain differences across and within countries.

Figure 5 Management practices and AI adoption

Notes: The index of management practices is based on three management-related questions on incentives at work. A higher value indicates that performance is more incentivised. The index is standardised across countries.
Data source: Authors’ survey on generative-AI adoption by workers, in January-February 2026.

Figure 5 shows a robust positive correlation between workers’ personnel management index and AI adoption. At the country-level (solid circles), we see that countries with a higher management index (meaning that performance is more incentivised) have higher AI adoption rates. When we pool all countries (hollow circles), the estimated slope suggests that a one standard deviation increase in a worker’s management index is associated with a 9.6 percentage point higher adoption rate. Higher values of the management index are associated with stronger employer encouragement to use AI and more provision of AI tools and AI-related training.

When we include employer encouragement in the decomposition, more than 80% of the gap in AI adoption between Europe and the US is accounted for. This finding could be a silver lining for Europe: the AI adoption gap between the US and Europe is not an inevitable consequence of Europe’s demographic, industry composition, or culture. It is, to a significant degree, a product of how firms are managed and whether they actively embrace AI as a tool for their workers. That is something that can, in principle, change.

AI adoption is correlated with stronger productivity growth

In a last step, we examine whether differences in AI adoption are already reflected in broader productivity growth.

A rapidly growing body of experimental and quasi-experimental research has provided early evidence that AI can substantially increase worker productivity. In line with this, we find reported time savings in our worker survey of 1–2.3% for all workers and 4.8–6.1% among AI users. Yet, it is not clear whether in the relatively short period since AI has been widely adopted in the labour market, productivity gains are already visible in the aggregate. Some also argue that even in the longer-run, AI will only have modest productivity effects (e.g. Acemoglu 2025).

In contrast to much of the literature which relies on exposure measures to AI, we use detailed sector-level information on AI adoption rates to assess the preliminary evidence of AI’s effects on productivity.

For the analysis of productivity effects in European countries, we regress country-industry productivity growth on 2025 firm adoption rates for 29 European countries, controlling for country and industry fixed effects. Our dependent variable is annualised log productivity growth for a given country-industry. In a placebo exercise, covering the years before recent developments of AI (2015–19), we find no significant relationship between AI adoption in 2025 and productivity. When we instead consider productivity growth from 2019 or 2022 up to 2024, we find larger coefficients which are significant in most specifications. The range of estimates in different specifications spans from annual productivity increases of 0.5 to 2.6 percentage points associated with a 10-percentage-point increase in firm-level AI adoption.

Looking at the US (Figure 6), we similarly find a positive correlation between industry-level AI adoption and productivity growth. We construct an industry-specific productivity pre-trend using annualised log growth in output per worker from 2015 Q4 to 2019 Q4 and subtract this from annualised log productivity growth since 2019 Q4. Industries with higher AI adoption experienced larger deviations from their pre-2019 productivity trends (𝜌 = 0.49). A 10-percentage-point increase in worker AI adoption is associated with approximately 0.6 percentage points of additional annualised productivity growth per year over 2019–2025, or about 3.7 percentage points cumulatively. This estimate is similar to our European estimates, which imply additional cumulative growth of 2.4 to 5.0 percentage points from 2019–2024.

Figure 6 US AI adoption and productivity growth relative to trend

Notes: Figure plots worker AI-adoption rates versus productivity growth in the US. Observations are at the industry level. Productivity growth is annualised growth of log output per worker from 2019 Q4 to 2025 Q3. Trend productivity growth is annualised growth of log output per worker from 2015 Q4 to 2019 Q4. Productivity growth relative to trend is the difference between productivity growth and trend productivity growth. We drop the utilities sector because it has a small sample size, and the value of its output is highly volatile and particularly dependent on commodities prices.

In a final exercise, we use the same strategy as for our analysis of productivity growth to study whether AI has affected sector-level employment. For both Europe and the US, we find no robust evidence that AI adoption is associated with employment expansion or contraction at the industry level in recent years.

Given the brief timeline between recent developments in AI capabilities and our most recent economic indicators, these findings should be interpreted with caution. Revisiting them as more data become available will be essential.

References

Acemoglu, D (2025), “The simple macroeconomics of AI”, Economic Policy 40(121): 13–58.

Albanesi, S, A Dias da Silva, J F Jimeno, A Lamo, and A Wabitsch (2023), “Artificial intelligence and jobs: Evidence from Europe”, VoxEU.org, 29 July.

Aldasoro, I, L Gambarcorta, P Rozália, D Revoltella, C Weiss, and M Wolski (2026), “How AI is affecting productivity and jobs in Europe”, VoxEU.org, 17 February.

Bencivelli, L, S Formai, E Mattevi, and T Padellini (2026), “Embracing AI in Europe: New evidence from harmonised central bank business surveys”, VoxEU.org, 9 January.

Bick, A, A Blandin, and D J Deming (2024), “The rapid adoption of generative AI”, VoxEU.org, 20 October.

Bick, A, A Blandin, D J Deming, N Fuchs-Schündeln, and J Jessen (2026), “Mind the gap: AI adoption in Europe and the US”, CEPR Discussion Paper 21337.

Bloom, N, R Sadun, and J V Reenen (2012), “Americans do IT better: US multinationals and the productivity miracle”, American Economic Review 102(1): 167–201.

Bresnahan, T F, E Brynjolfsson, and L M Hitt (2002), “Information technology, workplace organization, and the demand for skilled labor: Firm-level evidence”, The Quarterly Journal of Economics 117(1): 339–76.

Draghi, M (2024), A competitiveness strategy for Europe (Part A), Publications Office of the European Union.

Yotzov, I, J M Barrero, N Bloom, P Bunn, S J Davis, K M Foster, A Jalca, B H Meyer, P Mizen, M A Navarrete, P Smietanka, G Thwaites, and B Z Wang (2026), “Firms predict an AI productivity boom is coming”, VoxEU.org, 12 March.