The U.S. Census Bureau has changed how it asks American businesses about artificial intelligence. Starting with the October 19, 2026 collection, its Business Trends and Outlook Survey (BTOS) calls the technology "Super Intelligence or SI." The Bureau's own about page says the instrument wording was updated "due to Executive Order 14434," according to a November 2 post by Simon Willison, who read the Bureau's documents.
That matters because BTOS is the series most people quote when they say "X percent of US firms use AI." The wording of a survey question changes the answers, and this is the second wording change in about a year. This post explains what changed, what is confirmed, what is not, and how to read the numbers while the data settles.
TL;DR: what changed and what to do #
| Question | Answer |
|---|---|
| What changed? | From the October 19, 2026 collection, BTOS says "Super Intelligence or SI" in place of "AI" |
| Why? | The Bureau cites Executive Order 14434, per its about page as read by Willison |
| Is this the first break? | No. The question was reworded on November 17, 2025 |
| Are old and new numbers comparable? | Not directly. The Bureau says so for the 2025 change; the 2026 effect is unknown |
| When do we see the first SI results? | Scheduled for the Thursday after November 2, per Willison |
| What should I do? | Quote adoption figures with the wording period attached, and avoid stitching series together |
What did the Census Bureau actually change? #
BTOS is a recurring survey. According to Willison's write-up, the Bureau sends it to about 200,000 of roughly 1.2 million businesses every two weeks, and each business is surveyed once every twelve weeks. One of its questions asks whether firms use AI. That question is the source of many headline adoption figures in the press and in investor notes.
From the October 19, 2026 collection, the instrument was updated. The Bureau's BTOS about page states, in the words Willison quotes: "Beginning with the October 19, 2026 collection, instrument wording was updated due to Executive Order 14434." The new question describes the term as "formerly referred to as Artificial Intelligence or AI," and the examples inside it now use "SI" throughout. The most recent published figures, from the October 22, 2026 release, still use the old AI wording, because they cover earlier collection periods. The first results with the Super Intelligence wording are due shortly after, per Willison.
Where does the rename come from? #
Executive Order 14434, titled "Inaugurating the Era of Super Intelligence," was signed on September 29, 2026. We covered it in detail in our explainer on the order. In short, it directs executive-branch agencies to use "Super Intelligence" and "SI" in non-statutory documents, and it states that the new term carries the same meaning as AI under existing law. It also asked for proposed legal definitions within 60 days.
The Census Bureau is an executive-branch agency, so the order reaches its public communications and, it turns out, its survey instruments. The order does not mention BTOS. The Bureau applied it to a questionnaire sent to businesses, which is a more consequential place to change a word than a press release.
Private companies were not ordered to follow. Some did anyway, and we tracked the early adopters in who followed the Super Intelligence rebrand. The domain-name side of the same trend is in our report on ICANN applications for agent, AGI and ASI domains.
Why a one-word change can move survey numbers #
Survey methodologists treat question wording as a measurement instrument. A change in the words can change who says yes, even when nothing in the real world changed. Willison points to a clear example from this same survey.
On November 17, 2025, BTOS changed its AI question. Before that, it asked whether the business used AI "in producing goods or services." After it, the question asked about use "in any of its business functions." Willison quotes the Bureau's own note, published in an AI question wording update document: answers before and after that date "are not directly comparable."
The change in reported adoption was large. In Willison's chart of the biggest firms, those with 250 or more employees, reported use rose from about 12 percent in September 2025 to about 30 percent in February 2026 and about 40 percent in October 2026. Part of that rise is real adoption. Part is a broader question. The data alone cannot separate the two.
Horizontal survey result bars of different lengths, one highlighted, illustrating how Census Bureau survey wording shifts AI adoption answers
This is the same trap we flagged when a viral chart claimed a lack of AI payoff outside tech. Our write-up of the Apollo chart claim walks through how a series built on BTOS data can mislead if you ignore the wording break. A line that runs flat, then jumps, then flattens again may reflect questionnaires rather than companies.
What could the "Super Intelligence" wording do to responses? #
Nobody knows yet, and that is the honest answer. There are a few plausible mechanisms, and we label them as hypotheses, not findings.
- Confusion or hesitation. A respondent who last saw the AI wording twelve weeks ago may at "Super Intelligence." Willison raises this question himself. The Bureau's note that SI was formerly called AI may soften the effect.
- Narrower reading. "Super Intelligence" sounds like something more advanced than a chatbot. Some firms that use ChatGPT for emails may answer no. That would push reported adoption down.
- No effect. The note may be enough, and respondents may map SI to AI without thinking.
The two plausible effects point in opposite directions from the 2025 break, which pushed numbers up. If the next release shows a drop among large firms, the wording is a suspect. If numbers hold, the rename was cosmetic. Until the Bureau publishes, any claim in either direction is speculation.
Who this affects #
| Reader | Why it matters |
|---|---|
| Analysts and journalists | Charts built on BTOS now have two wording breaks. Annotate them |
| Investors reading adoption trends | A drop or jump near late 2026 may be a questionnaire artifact |
| Policy staff | The rename shows how an executive order's vocabulary rule reaches data collection |
| Founders selling AI to businesses | Market-size slides that cite BTOS should state the wording period |
| Researchers | Use the Bureau's change documents to place a break flag in your regressions |
For a broader sense of how thin the adoption evidence is, see the PNC finding that only 2.2 percent of US households pay for an AI subscription and Gallup's teacher survey. Both show that the answer depends on who is asked and how.
How to quote BTOS numbers responsibly #
A short checklist works well here.
- Name the period. Say "under the old production wording," "under the business functions wording (from November 17, 2025)," or "under the Super Intelligence wording (from October 19, 2026)."
- Do not splice. Do not draw one trend line through two wording regimes without a visible break.
- Prefer within-period changes. Comparing two releases that use the same wording is safe. Comparing across the break is not.
- Cite the Bureau's documents. Link thedata page and the wording-change note, not only a secondary chart.
- Check the next release. The first SI-wording numbers will show how big the shift is.
If you work with survey data inside an AI workflow, the same discipline applies to model-generated summaries. We wrote about the cost of checking and redoing machine output in our piece on the AI verification tax. A figure that looks clean in a summary may sit on a broken series.
The wider pattern: language as policy #
This is not the first time a federal agency touched AI data in an unexpected way. In September we covered OpenAI's disclosure that its agents had touched SEC, Census and Investor.gov data, a very different story about the same agency. Together they show how deeply AI now interacts with official statistics, both as a topic and as a tool.
The rename also shows a tension. The executive order says SI means the same as AI under existing law. A statistical agency cannot make that promise about respondents. Law can declare two words equal, but a farmer, a dentist or a logistics manager reads each word with their own associations. Measurement lives on the respondent's side of that gap.
What we know, and what we do not #
Confirmed from the sources we read: the Bureau's about page says the BTOS wording changed from the October 19, 2026 collection due to Executive Order 14434, according to Willison's quotation. The previous change on November 17, 2025 is documented by the Bureau, and the Bureau says results before and after it are not directly comparable. The survey scale, about 200,000 businesses every two weeks, is Willison's description of the Bureau's design.
Not confirmed: whether the rename changes answers, by how much, and in which direction. We did not independently load the Census pages beyond what Willison quotes and what the Bureau's published pages state, so check the primary pages before you cite exact wording. The first SI-wording results will be the first real evidence.
Our read: treat the next release as a natural experiment. If you publish adoption numbers, add a footnote now. It costs one sentence and protects you if the series jumps.
Related reading #
Details are accurate as of November 2, 2026. The Bureau may revise its documents and publish new results.