The US Census Bureau has changed the AI question in its business survey twice, so the figures across several years cannot simply be compared.
Why this is useful to know
The US statistics agency Census Bureau measures through a fortnightly survey how many businesses use AI. Since 19 October 2026, AI in that survey is called 'Super Intelligence', following a presidential decree (Executive Order 14434). Simon Willison shows why that matters. The question had already changed in November 2025, and as a result AI use at large companies appears to have gone from about 12 percent to almost 40 percent. Part of that rise therefore comes from the question and not only from reality. Remember: adoption figures only mean something if you know which question was asked.
When you need this
Do you use adoption figures from government sources in a quote, presentation or business case? Then first check the exact question, the measurement date and the definition of AI that was used. The US figure for large companies (250 or more employees) rose from about 12 percent in September 2025 to about 30 percent in February 2026, after the question change of 17 November 2025, but those two points cannot be compared with each other. If you run your own AI measurement in your company, write down your question and keep it the same over time. If you change the terminology, such as a term that has a different name in a source, explain which words you use, so colleagues and customers do not get confused.
How it works under the hood
The Census Bureau's BTOS survey runs every two weeks among about 200,000 of the 1.2 million businesses. Each business is surveyed again every twelve weeks, so panel members see the same question again. The time series contains at least two breaks: the change from 'producing goods or services' to 'any of its business functions' on 17 November 2025, and the terminology change per collection from 19 October 2026. The publication of 22 October 2026 still uses the old AI wording; the first results with 'Super Intelligence' were, according to Willison, scheduled for early November. For your data processing this means: add a variable for the question version and never compare series across a break without marking it. Comparable estimates call for recalibration or a parallel measurement with the old and new wording.
This news touches chapter 25 and 4 of the course AI(l) or nothing by Newway. If you follow the course, this subject returns there at a calmer pace.
What you can do with this
Want to know what this means for your business? Plan an introduction: half an hour, no sales pitch, but an honest answer to the question where to begin. Rather read on first? The course AI(l) or nothing is free and starts with the basics.
Written in response to a publication by Simon Willison: https://simonwillison.net/2026/Nov/2/census-bureau-super-intelligence/
