Most businesses have not switched it on.

The U.S. Census Bureau’s Business Trends and Outlook Survey asked, between 7 and 20 September 2026, whether the business had used artificial intelligence in any business function during the previous two weeks. 23.8 percent of employer businesses said yes. The standard error is 0.26. 66.5 percent said no. Another 9.7 percent did not know.

Those are three answers, and they do not say the same thing. “Yes” means recent use, in some function, not in every function. “No” is the majority. “I don’t know” is nearly one business in ten, unable to say whether their own business had used it. The survey does not explain why that group did not know. The reason should not be invented.

Looking six months ahead, 27.6 percent expected to use it. That figure is above 23.8 percent, but not by much, and it is still under a third. It looks forward. It does not measure use that has already happened. It should not be mixed with the 23.8 percent, because one is an expectation and the other is what they said they had done in the previous two weeks.

Scope matters too, and it is easy to skip. This is the national figure for employer businesses, excluding farms, from a sample of about 1.2 million. It is not the figure for one industry. Hotels may sit above it or below it, and this survey does not say which. It is not an Ebensense figure either. We did not collect it. The 23.8 percent is in the survey data, not in the 24 September 2026 tip sheet.

The technology can be available and still not be in use. The survey asked the second question, not the first. For most employer businesses in that window, it had not been switched on. The distance between having a model at hand and having a system the company uses is the work: deciding what it is for, building the system it sits in, writing the software that connects it, and putting the data into a form a team can query. None of that comes with the model.

That is why it is a mistake to assume the company already uses the tool just because the model exists. In that window, 66.5 percent said they had not used it, and 9.7 percent did not know. Ask whether it is used, in which function, and who knows, not whether a product is available. The work starts with that answer: what it is for, which system it sits in, and which data the team can query.