ECB: 52% of Euro Area Workers Use AI, Economy-Wide Time Saved 3.8%

Half of euro area workers now say they use AI on the job. That number, from the ECB's own monthly survey of about 20,000 people, is what Christine Lagarde reached for on 14 September 2026 when she argued that Europe's productivity problem has a way out. Then read the next line in the underlying survey write-up: the economy-wide saving in working hours that all this use adds up to is about 3.8 percent. The gap between those two figures is the story.

Where the numbers come from

[Fact] The primary source is an ECB Blog post of 26 August 2026 by António Dias da Silva, Laura Lebastard and David Sondermann, three economists in the ECB's supply-side and labour division. It draws on the ECB Consumer Expectations Survey, which polls roughly 20,000 people across 11 euro area countries every month. The survey question is direct: "Do you personally use artificial intelligence (AI) in your work (including tools such as ChatGPT, Claude, Gemini, etc.)?"

[Fact] Lagarde's speech, "A new age of capital: growth, sovereignty and AI," delivered on 14 September 2026, cites that blog for its claim that the share of euro area workers using AI at work has doubled in two years and now exceeds 50 percent, "a level it took the internet about a decade to reach." The speech adds that US workers spend two to three times as much of their working week using AI as workers in the largest euro area economies, and that ECB estimates put the productivity gain from fast adoption at up to 4 percent over a decade. The speech is the occasion; the blog is the evidence, so this post stays close to the blog.

Adoption: 26, 41, 52

[Fact] The share of employed respondents using AI at work went from 26 percent in 2024 to 41 percent in 2025 to 52 percent in 2026. Users report using it around three days a week on average.

[Estimate] Two things stand out when you line those up. First, 26 to 52 is exactly a doubling, so "doubled in two years" is not a rounding flourish. Second, the annual increments are 15 points and then 11 points. The curve is still rising but it is already bending. If the next step is 11 points or less, the euro area is looking at something like 60 to 63 percent in 2027, not another doubling.

[Fact] Adoption is stratified by education and age. Among the highly educated, 61 percent use AI at work; among those with lower education, 37 percent. Younger workers are around 20 percentage points more likely to use it than older colleagues. Men report slightly higher use than women, but the authors say age and education are the primary drivers, and the pattern has held for three years. [Estimate] The education gap is 24 points, wider than the age gap.

The detail I found most useful: [Fact] once people start using AI, they use it at very similar intensity regardless of group, with weekly averages between 2.5 and 2.9 days. The divide is at the door, not inside the room.

Time saved: three hours, 7.7 percent, then 3.8 percent

[Fact] The median AI user reports saving three hours a week, which the authors put at about 7.7 percent of median working time. The distribution is highly skewed: most users report moderate savings and a small number report very large ones.

[Estimate] Three hours being 7.7 percent implies a median working week of about 39 hours, which is a sensible check that the two figures come from the same population.

The authors then do the arithmetic that most coverage skips. [Fact] Only 48.8 percent of workers both use AI and report saving time with it. Multiplied through, the whole-economy efficiency gain, the share of working hours saved because of AI, is closer to 3.8 percent. And they add two conditions on top: the saved hours only become productivity if workers turn them into more output rather than into less directly productive activity, and only if the employer is in a position to use the extra capacity.

[Fact] For scale, the blog notes that estimates of additional annual productivity growth from AI over ten years range from 0.1 to 3.4 percent, and that separate ECB work puts the euro area figure at around 0.35 percentage points a year. [Estimate] Ten years of 0.35 is 3.5 percent, which is close to the "up to 4 percent over a decade" in Lagarde's speech. The speech and the blog are consistent, but the speech quotes the ceiling.

Which tasks, and which workers

[Fact] The biggest reported time savings come from tasks few people use AI for. Generating or debugging code saves nearly eight hours a week for those who do it, but only about 8 percent of workers use AI that way. Data analysis, automating routine tasks and creating audio or visual content follow the same pattern. The most common uses, research, information gathering, writing and text editing, save considerably less. The authors flag a mechanical caveat: respondents who report saving a lot of time also tend to tick more tasks, which inflates the per-task averages.

[Estimate] Weighting by usage, the coding task contributes roughly 0.6 hours per worker per week to the aggregate (8 hours times 8 percent). The headline "eight hours" is real for the software developers and data scientists who report it; it is a rounding error for the median office worker, whose most common AI task is closer to what writers and authors do.

[Fact] Managers use AI the most, save the most time, and hold the most positive views of it. The authors read this as putting managers in a good position to steer adoption. There is a less comfortable reading: the group with the most power over whether AI is rolled out is also the group that personally benefits most from it.

Sentiment is falling while use is rising

[Fact] The share of workers who view AI positively fell from 43 percent to 41 percent over the past year. Workers with negative views of AI are also more pessimistic about the economy, expect higher unemployment a year ahead, fear job loss more, and expect weaker income growth. The authors note that fear of being replaced by AI rises when labour market prospects are generally worse.

That correlation runs both ways, and the blog does not claim otherwise. A worker who expects to be laid off may blame AI; a worker who has watched AI take over parts of their job may expect to be laid off. The survey cannot separate the two. This site's coverage of the ECB's earlier analysis of AI-intensive firms and the ONS's parallel UK business survey found the same tension: adoption figures rising faster than any measurable change in employment.

Why the other half does not use it

[Fact] One-third of non-users say AI is irrelevant to their tasks. Others prefer traditional methods, worry about accuracy and reliability, or say their employer does not provide it. And 41 percent of workers say they are simply not interested in using AI, including 34 percent of managers.

[Fact] Asked what would change their minds, about half point to better training and a clearer understanding of what AI is useful for. About half of firms plan to invest in AI training in the next 12 months, according to the ECB's SAFE survey of firms, which the authors note means about half do not, possibly connected to the third of managers who are not interested.

Training is the answer the survey respondents give, and the answer the authors endorse. It is worth setting that against a finding this site covered from KPMG's internal telemetry, where the effect of AI training on usage sophistication faded within a month. What workers say would help and what measurably helps are not the same evidence.

The limits of a self-report

Everything above is self-reported. "Hours you would have needed without AI" is a counterfactual the respondent is asked to imagine, and the authors themselves note their time-savings figures run higher than a comparable Federal Reserve Bank of St. Louis study, partly, they suggest, because of question design and geography. The "use AI at work" question counts a worker who asked ChatGPT one question last month the same as one who runs a coding agent all day. And the survey covers 11 of the 20 euro area countries.

None of that makes the numbers wrong. It makes them what they are: a measure of how many people have picked up the tool, and how much time they believe it gives back. The 3.8 percent is the honest version of the 52 percent, and it is the number the productivity argument has to be built on.

Sources

• Dias da Silva, A., Lebastard, L. and Sondermann, D. (2026), "AI adoption and the productivity promise: what workers report," The ECB Blog, 26 August 2026. https://www.ecb.europa.eu/press/blog/date/2026/html/ecb.blog20260826~e1c1a89999.en.html
• Lagarde, C. (2026), "A new age of capital: growth, sovereignty and AI," speech, 14 September 2026. https://www.ecb.europa.eu/press/key/date/2026/html/ecb.sp260914_2~a3f0efbee4.en.html
• Data: ECB Consumer Expectations Survey (about 20,000 respondents, 11 euro area countries, monthly); ECB SAFE survey of firms, as cited in the blog.

Quotations are brief and used for commentary. The blog states that the views expressed are the authors' and not necessarily those of the ECB or the Eurosystem.

AI-assisted analysis: this post was drafted with AI assistance and reviewed by a human editor before publication. Figures marked [Estimate] are this site's own arithmetic on the published numbers, not results the ECB reports.