AI at Work Statistics 2026: Adoption, Time Savings, and the Education Gap

AI at work is spreading, but not evenly. In the Federal Reserve’s survey, 25% of workers used generative AI at work in the prior month. Adoption ranged from 10% among workers with high school education or less to 43% among workers with graduate education. Users reported much stronger practical benefits, while concern about replacement barely changed.

What the survey measured

The Federal Reserve’s Survey of Household Economics and Decisionmaking, or SHED, was fielded in October 2025. The overall report surveyed nearly 13,000 U.S. adults.

The AI employment questions used more specific populations. The work-use and attitude questions had 7,488 unweighted worker respondents. The follow-up question about usage frequency had 1,850 AI-user respondents.

That distinction matters. The 25% adoption figure uses workers as its denominator. The user-benefit figures describe people already using AI, so they do not describe the average worker. The survey asked about use during the prior month. This makes the result a recent snapshot, not a full-year adoption rate. Fielding in October may also capture conditions specific to that period, but the survey does not by itself establish a seasonal pattern.

Adoption is unequal across education, age, and autonomy

Education shows the clearest divide. Adoption rose from 10% for workers with high school education or less to 17% for workers with some college or technical education, 20% for workers with an associate degree, 34% for workers with a bachelor’s degree, and 43% for workers with graduate education.

Age also matters. Use was 20% among workers ages 18–29, 31% among ages 30–44, 28% among ages 45–59, and 13% among workers ages 60 and older. Workers with a disability reported 19% adoption, compared with 25% for workers without a disability.

Control over daily work is associated with use as well. 30% of employees with frequent task autonomy used AI, compared with 20% of employees with less control.

These are relationships, not proof of cause. The survey cannot show that education, age, disability status, or autonomy directly causes AI adoption. Work roles, access to tools, employer policy, and other factors may also differ across these groups.

How often users turn to AI

Among workers who used generative AI at work during the prior month, usage ranged from intensive to occasional.

The denominator here is AI users, not all workers. That means the table describes intensity after adoption. It does not imply that 15% of all workers use AI multiple times daily.

The distribution also cautions against treating “AI use” as a single behavior. Someone using a tool weekly may have a different workflow, training need, and productivity outcome from someone using it several times a day.

Users report more utility than the average worker

The survey shows a large difference between general worker sentiment and the experience of people who already use AI.

The pattern is clear: current users are much more likely to report time savings, better quality, new capabilities, employer encouragement, and career benefits. Their replacement concern is only slightly higher.

That does not prove AI caused the positive experience. Workers who choose to use AI may already have better access, more supportive managers, or tasks that fit these tools. Still, the contrast suggests that direct experience changes perceived usefulness more than it changes perceived replacement risk.

If saved minutes are the practical benefit, the next question is what to do with them. A short priority system can help turn recovered time into deliberate action through Do Only 3 Things a Day. Related guidance on how to prioritize tasks and digital minimalism can help keep additional tools from creating additional noise.

PlainReads calculations from the SHED data

The following table is a PlainReads calculation table created from the supplied SHED adoption shares. The ratios divide the higher share by the lower share; percentage-point differences subtract the lower share from the higher share.

The same calculation applies to the attitude table. Relative to all workers, AI users were 37 percentage points more likely to say AI saves time, 28 points more likely to cite quality improvement, 27 points more likely to cite new tasks, 32 points more likely to report employer encouragement, and 28 points more likely to expect career improvement. Replacement worry differed by only 2 percentage points.

What labor projections add

The Bureau of Labor Statistics provides a different view: projected employment change across occupations, not reported workplace adoption. Its 2024–34 projections identify several occupations with strong expected growth.

These are projections, not promises. They should not be read as proof that AI adoption will create every projected job, or that every worker using AI will move into these occupations. BLS estimates can be revised in later releases, so the linked publication is the relevant version for these figures.

Methodology and limitations

The SHED results are survey estimates, not a census of every worker. The reported worker count is unweighted for the AI question, while the overall survey covers a broader adult population. The different denominators mean adoption, frequency, and attitude results should not be combined as though they came from identical groups.

The prior-month wording limits what the adoption rate can tell us. It measures recent use during the survey window; it does not establish how many workers used AI earlier in the year, how many will continue using it, or whether use is growing steadily. A single field period also cannot separate a durable trend from temporary conditions.

The subgroup comparisons are descriptive. Higher adoption among more educated workers does not establish that education alone produces adoption. The autonomy comparison similarly identifies an association. It does not show whether autonomy makes AI easier to use, whether AI creates more autonomy, or whether another workplace factor influences both.

The attitude comparisons have a related limitation. AI users may report more benefits because they have had successful experiences, but people with favorable expectations may also be more likely to become users. The data support a difference in reported experience; they do not establish a causal effect.

Sources

• Federal Reserve SHED employment chapter
• Federal Reserve SHED Appendix B
• BLS artificial intelligence, information technology, and employment projections

Frequently asked questions

Is 25% an annual AI adoption rate?

No. It is the share of workers who reported using generative AI at work during the prior month when the survey was fielded.

Why do AI users report more benefits than all workers?

The AI-user group has direct experience with the tools, while the all-worker group includes people who did not use them. The groups also use different denominators.

Does education cause higher AI adoption?

The survey shows an association, from 10% among workers with high school education or less to 43% among workers with graduate education. It does not prove that education alone causes the difference.

Do BLS projections prove AI will create these jobs?

No. The projections are forward-looking estimates for occupations. They describe expected employment change, not guaranteed outcomes for individual workers or proof of a direct causal effect from AI.