A US government survey quietly renamed AI to Super Intelligence

Simon Willison set out to update an old chart on how many US companies use AI, and found that the government had changed the question underneath it. The source is the Census Bureau's Business Trends and Outlook Survey, which goes to 200,000 of roughly 1.2 million businesses every two weeks, so each firm answers about once every twelve weeks. It is one of the few large, regular readings of real adoption rather than vendor claims.

The catch is that the question keeps moving. Until late 2025 it asked whether a business used AI "in producing goods or services"; then it widened to "any of its business functions," which is a break in the series rather than a smooth rise. Read with that caveat, adoption among the largest firms (250 or more employees) went from about 12% in September 2025 to roughly 30% by February 2026 and 40.7% by October 2026. The newer twist, as Willison writes, is that from October 19, 2026 the survey stopped saying AI at all. It now asks about "Super Intelligence or SI," noted as "formerly referred to as Artificial Intelligence," a change the Bureau attributes to Executive Order 14434. The example prompts are the giveaway: "SI-powered data analysis" and "virtual SI agents for customer service" describe the same chatbots and automation as before.

The open question is whether respondents, who last saw the old wording twelve weeks earlier, answer differently when asked about "Super Intelligence." If they do, the cleanest public measure of US AI adoption gets a second discontinuity baked into it, on top of the 2025 one.

Why it matters

If you cite the Census figures to argue AI adoption is climbing, you now have two wording breaks to disclose, not one, and the next data point may jump or drop for reasons that have nothing to do with real usage. Anyone tracking these numbers should read the question text, not just the percentage.