The headline number belongs to the largest companies
Simon Willison's November 2 analysis of the Census Bureau's Business Trends and Outlook Survey (BTOS) tracks AI use across seven business-size categories. His chart shows the largest businesses — those with 250 or more employees — moving from roughly 12% reporting AI use in September 2025 to around 30% by February 2026, and approximately 40% by October 2026.
Those figures are not an overall US business adoption rate. They describe the largest employer businesses in the survey, and they are Willison's six-survey trailing averages calculated from Census Bureau data. The smoothing is his choice, not Census methodology.
The apparent jump also crosses a material change in the question. Before November 17, 2025, businesses were asked whether they used AI in producing goods or services. From that date, the question asked whether they used AI in any of their business functions. The broader definition captures uses that the earlier wording could miss.
The result is a chart that looks like a rapid acceleration in adoption, but cannot be read as a continuous measurement of behavioral change. The question changed, and the two series need to be treated separately.
What Census changed, and why
The Census Bureau's published documentation identifies November 17, 2025 as the effective date of the major change. The original question asked whether a business had used AI in "producing goods or services." The replacement asked whether it had used AI in "any of its business functions."
That difference matters. During cognitive testing, Census found that some respondents did not consider their businesses producers of goods or services and therefore thought the original question did not apply. Others initially answered no but, when questioned further, described using AI for hiring, project management, accounting, research and development, or functions embedded in their software.
Census said the broader wording was based on expert feedback and cognitive testing. It also acknowledged a level shift in the results and created a new AI time series beginning with data released on December 4, 2025.
This is not evidence that Census manufactured a higher adoption rate. It is evidence that the old and new questions measure different scopes of business activity. The new measure may better capture how businesses actually use AI, but the two series cannot establish how much of the apparent increase came from new behavior.
Another terminology change is already in place
There is a further change to watch, but it should not be confused with the November 2025 expansion.
Beginning with the October 19, 2026 collection, Census changed the instrument's terminology from Artificial Intelligence (AI) to Super Intelligence (SI), following Executive Order 14434. The new question explicitly describes SI as formerly referred to as AI and updates the examples provided to respondents.
Willison's October 22 data release and the chart in his November 2 analysis still reflect the earlier AI wording. The first results using the new SI wording had not yet been published when he wrote.
Whether that terminology change affects responses remains an open question. It is a new measurement risk, not an explanation for the 40.7% figure already reported.
What the survey measures, and what it does not
BTOS is a business survey, not a worker survey. It asks businesses whether they used AI in the preceding two weeks; it does not directly measure employee-level use, AI spending, output or productivity.
The percentages discussed here are estimates from businesses, with sampling and response uncertainty. Willison calculates an arithmetic average across six successive survey estimates, separately for each question wording. That smoothing reduces individual-period volatility but is not an official Census methodology and does not make the old and new series comparable.
His chart also separates businesses by employee count. The 40.7% figure applies to businesses with 250 or more employees, not the entire US business population. Readers should not quote it as a national business adoption rate.
Sources
- [1] Simon Willison, "The U.S. Census Bureau and 'Super Intelligence'", November 2, 2026 (primary analysis; read first-hand in full — the seven-size-class chart built from the Census "Employment size class" XLSX, the 12% / ~30% / 40.7% six-survey trailing averages for businesses with 250+ employees, the old and new question text, and the note that the first Super Intelligence-wording results are due for publication on Thursday)Read source
- [2] U.S. Census Bureau, "BTOS AI Core Question Updates", December 3, 2025 (primary; read first-hand — the November 17, 2025 wording change based on expert feedback and cognitive testing, the original and updated question text, the level shift and the new time series beginning with data released on December 4, 2025, and the FAQ's cognitive-testing findings)Read source
- [3] U.S. Census Bureau, "BTOS — About", updated October 22, 2026 (primary; read first-hand — the survey design and panel structure, the November 17, 2025 broadening to "any of its business functions", and the October 19, 2026 instrument wording update due to Executive Order 14434)Read source