AI Is Becoming Ordinary Enough for the Census People to Start Asking About It

There is a point in the life of a technology when it stops being treated mainly as a product launch, a workplace experiment or a source of alarming headlines and starts being treated as a household characteristic. The federal government may be approaching that point with artificial intelligence.

On Sept. 9, the National Telecommunications and Information Administration published a Federal Register notice seeking public comment on the next major revision of its Internet Use Survey, a recurring supplement to the U.S. Census Bureau’s Current Population Survey. Buried among questions about wording, broadband, children’s technology use and the practical limits of a ten-minute survey is a simple question with a larger implication: Should the government collect more data about AI, including its use at work or at home and Americans’ attitudes toward it?

The answer has not been decided. The notice is a request for comment, not an announcement that AI questions are already part of the 2027 questionnaire. But even asking the question marks a subtle change in how AI is being framed. The issue is no longer only whether the technology is impressive, dangerous, overhyped or profitable. It is whether AI has become ordinary enough to be measured as a recurring feature of American life.

That is a different kind of milestone. Technology companies can count users. Polling firms can ask whether people trust chatbots. Employers can survey workers about adoption. Federal statistical programs are built for something else: definitions that can survive repeated measurement, data that can be compared over time and questions designed to tell policymakers not just that a behavior exists, but who is doing it, where it is happening and how it intersects with the rest of household and working life.

The NTIA Internet Use Survey is a particularly revealing place for that transition. NTIA says it has sponsored eighteen editions since 1994, when the federal measurement problem was much more literal: who had a computer, who used the internet and how people connected. The survey has had to follow the internet from dial-up and DSL into a world where connectivity can disappear into the background of smart appliances, streaming devices, phones, games and ordinary transactions. The agency now says the survey needs another serious language check because the technology — and the words people use for it — has changed again.

In the notice, NTIA says AI has already had major effects on how people gather information online and how that information is generated. It also argues that effective longitudinal measurement of AI and other internet-enabled technologies will be necessary to understand future issues. That phrase, “longitudinal measurement,” is the live wire in an otherwise procedural document. It means the agency is not merely interested in taking the temperature of one AI moment. It is considering whether AI belongs in a time series.

And once something enters a time series, the questions get harder in useful ways.

The government has to decide what “using AI” actually means

A household survey cannot simply ask, “Do you use AI?” and assume everyone is answering the same question. Someone who deliberately opens a chatbot to draft an email is making an obvious choice. Someone whose search engine summarizes results with generative AI may be using AI without thinking of the interaction that way. So may a worker whose employer has added an automated assistant to software they were already required to use. A phone may rewrite a message, a photo app may remove an object, a customer-service system may generate an answer, and none of those experiences necessarily register to the person as “AI use.”

That problem is not unique to AI. The new NTIA notice explicitly points out that internet use itself has become difficult to notice because online connections are embedded in everyday devices and services. The survey is being re-examined partly because the older language of “going online” no longer cleanly describes a world in which many things are online by default. AI is arriving at the same measurement problem at remarkable speed: a technology can become more common at the same time that it becomes less visible.

Attitudes are another challenge. A person can distrust generative AI while using it every day at work. Someone can avoid chatbots but rely on products that incorporate machine-learning or generative features. Another person may enthusiastically use AI for personal tasks while objecting to an employer using it to evaluate performance. Measuring “attitude toward AI” therefore forces the survey designer to decide whether AI is one category or several different social experiences hiding behind the same label.

That is exactly why government measurement matters. The boring work of defining a question is part of defining the phenomenon. If the wording is too broad, the data become mush. If it is too narrow, the survey may miss forms of AI use that have already become invisible infrastructure. If the wording changes every cycle, the government loses the ability to tell whether behavior changed or merely the vocabulary did.

Why the Current Population Survey changes the stakes

The proposed setting matters too. The Internet Use Survey is administered by the Census Bureau as a supplement to the Current Population Survey, the monthly survey best known as a foundational source of U.S. labor-force statistics. The NTIA notice itself points to the workforce as a reason some internet-use questions may be particularly valuable. That makes a future AI module potentially more interesting than a standalone poll about whether Americans like or dislike the technology.

CPS supplements can be analyzed alongside the broader demographic and labor information collected through the survey system. In practical terms, that creates the possibility of asking more structural questions: Is AI use concentrated in particular occupations? Does home use look different from workplace use? Are there gaps by age, education, income or other characteristics? Is AI being adopted by workers because they choose it, because their employers require it, or because it has quietly become part of the software around them? The Sept. 9 notice does not specify those questions, but it opens the door to deciding which of them are important enough to measure consistently.

The scale is substantial. NTIA estimates the Internet Use Survey will reach about 50,000 households, with responses collected through personal visits and live telephone interviews. Participation is voluntary, and the estimated response time is ten minutes. That time limit is not a trivial detail. The notice repeatedly acknowledges that survey space is scarce and asks commenters who want new questions to identify what might be removed. AI is not simply being waved into the questionnaire because it is fashionable; it would have to compete for space with other things the government already tracks.

That competition is part of the story. Statistical attention is a finite resource. Deciding that AI deserves a recurring question means deciding that the behavior has enough public-policy value to justify the burden of asking it across tens of thousands of households and, ideally, asking it in a form stable enough to compare across years.

From novelty to background condition

This is not the Census Bureau declaring that AI is good, safe, inevitable or universally adopted. It is not the decennial census, either. It is a Commerce Department survey program, fielded with the Census Bureau through the CPS, trying to decide what should be measured in November 2027. But institutions reveal what they think has become socially durable by what they decide is worth counting.

The distinction is easy to miss because the notice is written in the language of the Paperwork Reduction Act: burden estimates, information collections, OMB approval, public comments. Yet the bureaucracy is doing something culturally recognizable. It is updating the checklist of ordinary life.

In 1994, computer use was a sufficiently distinct behavior that it needed to be measured. Over time, the questions followed internet access, devices, online activities, privacy and the widening number of ways a person can be connected. Now NTIA is asking whether AI has crossed the same threshold — not because the technology has stopped being strange, but because it may be becoming too common to understand through anecdotes and product announcements alone.

There is also a useful humility in the notice. NTIA is not assuming the Internet Use Survey is automatically the right home for AI data. It explicitly asks whether AI use at home, AI use at work and attitudes toward AI belong in this CPS supplement or should be gathered in other ways. That is the right question because a bad national measure can create the appearance of knowledge without actually clarifying behavior.

If AI questions do enter the survey, the important moment will not be the first national percentage that comes out of it. It will be the creation of a baseline. Once there is a baseline, the government can begin asking what changed. Adoption can be compared with occupation, household characteristics and time. Attitudes can be tracked instead of guessed. The language can be tested against what people actually understand. The hype cycle can, at least in part, be replaced with a dataset.

That is what makes this small Federal Register notice worth noticing. AI has spent the last several years demanding to be treated as the future. The federal statistical system is beginning to ask a more mundane and ultimately more consequential question: Is it now part of the present?

Public comments on the proposed 2027 Internet Use Survey revision are due Nov. 9, 2026.

Source correction: The GovInfo URL supplied in the assignment ends in FR Doc. 2026-18266, which is the National Commission on the Future of the Navy notice. That PDF begins on printed page 57322 and therefore happens to include the final lines of the NTIA notice above it. The NTIA Internet Use Survey notice is FR Doc. 2026-18303, published at 91 FR 57320-57322.

SOURCE NOTES

• Federal Register, Sept. 9, 2026 — NTIA Internet Use Survey notice (FR Doc. 2026-18303; 91 FR 57320-57322)
• Public Inspection PDF for FR Doc. 2026-18303
• NTIA Data Central — history and datasets for the Internet Use Survey
• Census Bureau — About the Current Population Survey
• NTIA — 2025 Internet Use Survey Information Collection and prior questionnaire