BLS wants to measure AI use in everyday work

What this is about

The U.S. Bureau of Labor Statistics scheduled a Federal Register notice for July 10, 2026 on new artificial intelligence questions in the American Time Use Survey. ATUS is the U.S. survey that measures what people actually spend time doing: work, care, household tasks, commuting and leisure.

This is not a shiny product launch. That is why it is interesting. Once AI appears in official time-use data, the debate moves from opinions toward measurable habits: Who uses AI? For which tasks? How often? And where is adoption more claim than routine?

What the survey actually does

The American Time Use Survey asks people about a daily timeline. Adding AI questions would not automatically prove that AI raises productivity or replaces jobs. It can, however, show whether AI has entered real routines.

The key difference is the vantage point. Companies often report whether they adopted AI. Workers can show whether they actually use it, whether it comes from the employer or from personal tools, and whether it replaces, speeds up or merely supports work steps.

Why it matters

The measurement gap is large. A Federal Reserve FEDS Note from April 3, 2026 compared several U.S. data sources and found very different perspectives: about 18 percent of firms had adopted AI by the end of 2025, while work-related GenAI usage in an individual survey stood at about 41 percent. Employment-weighted estimates from the Survey of Business Uncertainty were higher because many people work at large firms that had adopted AI.

Those numbers do not necessarily contradict each other. They measure different things. That is exactly why better questions matter. For wages, training, work time, regulation and productivity debates, it is not enough to know whether a company bought an AI tool. We need to know whether people use it in daily work.

In plain language

Imagine a city wants to know whether new bike lanes work. Asking bike shops how many bicycles they sold is not enough. The city has to count whether people actually ride on those lanes in the morning, when they ride and which routes they avoid. AI is similar: tool sales are not the same as changed work.

A practical example

A case worker handles 42 customer requests in one day. In 12 cases she uses an AI system to draft a first summary. In 8 cases it saves about 5 minutes each, in 2 cases she has to rewrite the summary entirely, and in 2 cases she uses AI only to phrase a more polite reply.

For the employer, this may sound like automation. For statistics, it is more nuanced: AI sometimes saves time, sometimes creates rework, and changes work even when it does not fully replace a task. Those patterns only become visible when measurement gets close to actual work.

Scope and limits

First: a new question is not a causal study. It shows usage, but it does not automatically prove whether AI raises wages, displaces jobs or improves productivity.

Second: self-reports are imperfect. People forget tools, overestimate time savings or fail to recognize AI when it is embedded inside ordinary software.

Third: the data will build slowly. It will not answer fast product questions. For long-term labor policy, however, a consistent time series can be more valuable than loud one-off studies.

SEO & GEO keywords

BLS, American Time Use Survey, AI workforce data, generative AI adoption, labor statistics, productivity measurement, Federal Register, Federal Reserve, AI jobs, work automation