How much time AI saves at work: survey figures from the US Census Bureau and the ECB

Ask someone who uses AI every day how much time it saves and the answer usually comes quickly. The catch is that the answer is mostly an estimate.

On 11 August 2026 the U.S. Census Bureau published an article by Renee Stepler. It uses March 2026 data from the Household Trends and Outlook Pulse Survey (HTOPS) to examine how American workers use AI at work and how much time they report saving. The figures were later revised on 24 September 2026 using updated weights. On 26 August 2026 the European Central Bank (ECB) also published similar survey results for euro area workers on its official blog. The figures in this article reflect the versions checked in October 2026.

United States: most users save two hours or less

The Census Bureau survey shows that 56% of American workers said they had used AI at work for at least one of 11 listed tasks. The most common use was searching for information or technical help (37%), followed by writing communications, documents or instructions, and generating ideas (32% each), then interpreting, translating or summarising information (31%) and administrative work (27%).

The survey asked workers who had used AI in the last week how many extra hours the same work would have taken without AI. The answers were:

• Less than 1 hour: 25%
• 1 to 2 hours: 31%
• 3 to 4 hours: 15%
• More than 4 hours: 15%
• No time saved: 10%
• AI made the work take longer: 4%

In total, about 14% of users said they saved no time or even spent more, roughly one in seven.

Among workers who had used AI in the last week, frequency of use also differed clearly: the share using AI every day was 30% for men and 17% for women, and 25% for those aged 18 to 49 against 20% for those aged 50 or over.

When reading these figures, bear in mind the Census Bureau's definition: AI means computing systems that perform tasks normally associated with human intelligence, with examples including machine learning, natural language processing, virtual agents and speech recognition.

Euro area: a median of about 3 hours a week

The ECB blog post "AI adoption and the productivity promise: what workers report" was written by António Dias da Silva, Laura Lebastard and David Sondermann. Its data come from the ECB Consumer Expectations Survey, which interviews about 20,000 people a month across 11 euro area countries.

The share of workers using AI at work rose from 26% in 2024 to 41% in 2025 and 52% in 2026. Users make use of AI on about three days a week on average. The share was 61% among those with a university education and 37% among those with less education.

The median weekly time saved by users is about 3 hours, equivalent to 7.7% of median working time, and the distribution is highly skewed. By task, coding or debugging saves the most, almost 8 hours a week, but only about 8% of workers use AI for it; research, information gathering, writing and text editing are more common but save less time.

Counting all workers, only 48.8% report both using AI and saving time, and on that basis the ECB estimates an economy-wide efficiency gain of about 3.8%.

Saved time does not automatically become productivity

The ECB cautions that saved time becomes productivity only if two conditions are met: workers turn the freed-up time into output, and employers are able to make use of the extra capacity.

At the macroeconomic level, the literature cited by the ECB estimates additional annual productivity growth from AI over ten years of between 0.1% and 3.4%, while a separate ECB study estimates about 0.35 percentage points a year for the euro area. The range of estimates is wide.

Where the limits of self-reported figures come from

In both surveys, "time saved" is self-reported by respondents. The ECB asks how many extra hours respondents think they would need without AI, and they can enter an exact number or choose a range of up to 15 hours; the Census Bureau likewise asks how many extra hours would be needed without AI. Questions of this kind measure respondents' estimates, not timed records.

Questionnaire design affects the results. The ECB authors note that their figures are higher than those of a St. Louis Fed study, which reported time savings capped at four hours, partly because of differences in questionnaire design and regional coverage. They also note that people who save more time tend to select more tasks, which mechanically raises the average time saved per task.

There is also the time spent checking and correcting AI output. We could not find verified research that measures this time specifically, so we give no figure; all that can be confirmed is that 4% of users in the Census Bureau survey said AI made the work take longer.

When you see any figure for time saved, three questions are worth asking first: was it measured or self-reported, who was in the sample, and was the time spent checking and editing deducted?

Limits of the two studies

• The Census Bureau states that comparisons in its article are statistically significant at the 90% confidence level unless otherwise noted, and that all data are subject to sampling error, nonsampling error, modelling error and other factors.
• In September 2026 the Census Bureau revised its figures using updated weights, which changed some estimates but not the conclusions of the statistical tests.
• HTOPS uses a random sample of addresses, and invited households respond online through a link in a mailed letter; participation is voluntary, only the selected addresses can respond, the number of invitations is limited, and the average response time is about 20 minutes.
• The Census Bureau's time-saving question covers only workers who used AI in the last week.
• The ECB's time savings are also self-reported and are affected by questionnaire design, so they may not be directly comparable with figures from other studies.
• The two surveys cover the United States and 11 euro area countries respectively, and cannot be applied directly to other markets.

A two-week self-test: use records instead of impressions

Rather than adopting other people's averages, measure your own work. The method below is our suggestion and is not the conclusion of any study.

1. Choose three to five tasks you repeat often, such as writing follow-up emails, tidying meeting notes or gathering information.
2. In the first week, do these tasks without AI and note the start and finish time each time.
3. In the second week, do the same kinds of tasks with AI, time them in the same way, and record the time spent checking and editing the AI output in a separate column.
4. For each task, note whether the result needed rework and how long the rework took.
5. After two weeks, compare total hours, deduct the checking and rework time, and then look at what the time saved was actually used for.

Two weeks of records is of course not rigorous research, and task difficulty and workload will differ from week to week. But it at least turns "it feels faster" into a figure you can examine.

What is really worth knowing is not only how much time AI saves you, but where that time ends up.

Prices and details correct as of October 2026

Compiled by the Focus50 Editorial Team from public sources, for general reference only and not professional advice. Details are correct as of the publication date and may change.