Census Bureau Puts a Number on AI at Work: 55% Use It, a Third Save 1–2 Hours per Task

The U.S. Census Bureau has published what may be the most consequential dataset of the AI era so far: a representative, government-run measurement of how American workers actually use artificial intelligence on the job — and how much time it really saves them. The answer, drawn from the March 2026 wave of the Household Trends and Outlook Pulse Survey (HTOPS), is more grounded than either the boosterism or the doom narrative suggests.

About 55% of U.S. workers said they have used AI on the job for at least one of 11 tasks the survey asked about. That is a striking adoption number for a technology that barely registered in official statistics two years ago. But the productivity dividend is smaller and lumpier than the industry’s marketing would have you believe: among those who used AI at work in the previous week, a quarter said it saved them less than an hour, and 31% said it saved them one to two hours. Only 15% reported three-to-four-hour savings, and another 15% more than four hours. Meanwhile, 10% said AI saved them no time at all — and 3% said using AI actually added time to their work.

What workers actually do with AI

The survey’s 11 task categories read like a map of everyday knowledge work. The top five uses:

• 37% — searching for information or technical help
• 32% — writing communications, documentation, or instructions
• 32% — generating ideas
• 31% — interpreting, translating, or summarizing information
• 27% — administrative tasks

Notably, the tasks at the center of the AI hype cycle — coding and modeling, customer support, logistics and supply-chain management — drew far smaller shares. The modal AI user at work is not an agent orchestrating a software pipeline; it is someone asking a chatbot how to fix an Excel formula or drafting an email. The revolution, so far, is largely a reading-and-writing revolution.

Usage frequency reinforces the picture of a technology still settling in. Among AI users at work, 24% used it every day in the previous week, 46% used it at least one day (but not daily), and 30% didn’t use it at all that week. Daily-deep integration is the exception, not the rule.

The divides: education, sex, and age

The data reveals sharp demographic divides. Workers with a bachelor’s degree or higher were far more likely to use AI at work: 46% of them used AI to interpret, translate, or summarize information, versus 25% of those with some college and just 16% of those with a high school diploma or less. On frequency, 75% of college-educated AI users used it at least one day in the previous week, against 62% and 59% for the lower education bands.

Gender differences showed up in both task mix and intensity. Male workers were more likely to use AI for information search (42% vs. 31%) and for interpreting or summarizing (34% vs. 28%), and 30% of male AI users reported daily use against 17% of female users. Age mattered less than expected: younger workers (18–49) were more likely to use AI for four of the top five tasks, but there was no statistical difference in information search by age group, and the daily-use gap (26% vs. 20%) was modest.

Why a government number matters

Private surveys on AI adoption have proliferated — Gallup, Pew, and a cottage industry of vendor studies — but they vary widely in methodology, sampling, and definitions. The Census Bureau’s figure carries a different kind of weight. HTOPS is a bi-monthly, nationally representative household survey, its comparative statements are statistically tested at the 90% confidence level, and its author, Renee Stepler, is a supervisory survey statistician at the Bureau. When historians write the economic history of this period, this is the class of data they will cite.

The finding also lands amid an intensifying debate about “circular” AI financing and hidden debt-fueled infrastructure spending. Roughly $690 billion in annualized AI capital expenditure is flowing into data centers and chips on the promise of productivity gains. The Census numbers don’t refute that promise — 55% adoption with meaningful time savings for about half of users is genuinely significant — but they quantify how far the average workday still is from the agentic future being sold. Nearly a quarter of AI users report no time savings; fewer than half save more than two hours.

For enterprises, the survey is a reminder that AI ROI currently concentrates in language-heavy, information-retrieval tasks performed by highly educated workers — and that deployment strategies should be honest about that distribution rather than assuming uniform gains.

What to watch

The March 2026 table package is a baseline, not a finale. HTOPS repeats bi-monthly, meaning the Bureau has effectively built a longitudinal instrument for tracking whether workplace AI use deepens (daily-use rates climbing, coding and agent tasks growing) or plateaus. Paired with its May 2026 companion release on AI adoption by businesses — which found AI use growing across firm sizes and sectors between December 2025 and May 2026 — the federal government now has both sides of the ledger: what firms deploy and what workers experience.

The gap between those two ledgers — soaring corporate capex and modest individual time savings — is the defining tension of this phase of the AI economy. The Census Bureau, characteristically, has no opinion on it. But it has now given everyone the ruler.