What the October 2026 Census AI figures tell a large company deciding how fast to move
On 22 October 2026 the US Census Bureau published its newest count of businesses using AI. In the Business Trends and Outlook Survey for 5 to 18 October 2026, 24.6% of US employer businesses said they had used AI in the last two weeks. Among firms with 250 or more employees, the figure was 43.6%.
We read the tables, the question wording and the notes behind them. Our view is that the figures are useful background, and that a large company should set its pace with AI from tests on its own tasks.
What the October figures say
The survey goes to a sample of 1.2 million businesses, by the Census Bureau's description, split into six panels that answer in turn every two weeks. The question respondents answered was: "In the last two weeks, did this business use Artificial Intelligence (AI) in any of its business functions?"
| Group, 5 to 18 October 2026 | Used AI in the last two weeks |
|---|---|
| All US employer businesses | 24.6% |
| Firms with 250 or more employees | 43.6% |
| Firms with 100 to 249 employees | 35.3% |
| Firms with 1 to 4 employees | 24.3% |
| Information sector | 46.4% |
| Finance and insurance | 40.5% |
| Retail trade | 18.5% |
The same release says 28.6% of businesses expect to use AI in the next six months. Under the current wording, which started on 17 November 2025, the all-business figure has risen from 17.3% in the first period to 24.6% now. In Europe, Eurostat's 2025 survey of enterprises with 10 or more employees, run in early 2025 with a different question, puts large enterprises at 55.03%.
Two notes from the Census Bureau come with these figures. The first concerns a name. Since 22 October 2026 the Bureau labels the series "Super Intelligence or SI (collected as Artificial Intelligence or AI)," because Executive Order 14434 tells federal agencies to use those words. The respondents were asked about AI. We covered what the order changes in an earlier post on Super Intelligence and AI.
The second note matters more for anyone tracking a trend. The Bureau changed the question wording "Beginning with the October 19 2026 collection," so the October figures above are the last ones collected under the old wording. When it last changed the wording, in November 2025, it saw what it called "a level shift" and started a new series. The next releases may show another break.
Where the figures stop
They count any use in a two-week window. The question sets no minimum. A firm where one team used a chatbot to draft an email answers yes, and so does a firm that runs AI in every customer call. The UK's Office for National Statistics (ONS) asked about depth in June 2026 and found that only 10% of adopting UK businesses with 10 or more employees use AI "extensively." The ONS says answers to that new question are early indications.
The large-firm figure moves between periods. The Bureau publishes a standard error of 1.51 points for firms with 250 or more employees in the October period. A standard error measures how far an estimate can move from sample to sample. In the three periods before October the large-firm figure read 36.5%, 44.0% and 38.4%. So the 43.6% is one reading in a range, and a single jump between two periods can come from the sample alone.
Many respondents at large firms do not know the answer. In the Bureau's AI supplement, collected from 17 November 2025 to 8 February 2026, 22.9% of respondents at firms with 250 or more employees said they did not know whether their firm used AI.
Results need their own measure. In the same supplement, 64.3% of firms using AI said they had made no changes at all to use it, and 95.7% said AI had not changed their total employment. The figures count firms that tried AI, and whether it worked has to be measured separately.
Other surveys give other numbers. A Federal Reserve note of April 2026 found 16 US surveys whose estimates of work-related AI adoption ranged from 5 to 40 percent in mid-2024. The Stanford AI Index 2026 reports that 88% of surveyed organizations used AI in 2025. The ONS explains the gap: "Higher estimates reflect a broader, top-down view of whether firms use any AI technology across their operations," while lower ones capture use by workers or specific business functions.
The guide page on how many companies use AI sets out every official count side by side, with its method.
What the research says about results
If the use figures cannot tell a company how fast to move, results can. The published research shows results that depend on the task.
In a field study of an AI assistant at work, Brynjolfsson, Li and Raymond followed 5,172 customer support agents at one Fortune 500 software firm and found that AI assistance raised issues resolved per hour "by 15% on average." The authors say their findings cover effects over the medium term in a single firm.
A Harvard Business School working paper by Dell'Acqua and colleagues tested 758 consultants. On tasks inside what the AI did well, those using GPT-4 finished 12.2% more tasks, 25.1% faster, with more than 40% higher quality. On a task chosen to sit outside it, "consultants using AI were 19 percentage points less likely to produce correct solutions compared to those without AI." Two tasks that look equally hard can fall on different sides of that line.
At the level of a whole economy, Humlum and Vestergaard studied 25,000 Danish workers in 7,000 workplaces and found no measurable effect on earnings or recorded hours two years after ChatGPT's launch, "ruling out effects larger than 2%."
Read together, the studies show gains on some tasks, losses on others, and no measurable effect on earnings or hours across one country's workers so far. A company finds out which of its own tasks gain by testing them.
How we would set the pace
Our view is that a large company should set its AI pace in three steps, each with its own record.
- Find out what you already use. If 22.9% of respondents at large US firms cannot say whether their firm uses AI, an internal inventory comes first. List each tool, the team that uses it, the data that goes into it and the person who owns it.
- Pick two or three tasks and test them on your own cases. Write down the task, a set of real examples and the correct result for each, before anyone tries the AI on them. Score the AI against those results. The Dell'Acqua study is the reason: a task can look like a good fit and still be one where AI makes the work worse.
- Expand where the tests pass, at the speed the tests allow. A task that passes moves to wider use with the same test running after every change. A task that fails stays where it is until something changes and the test is run again.
The guide page on choosing your first AI use cases covers step 2 in detail, and the page on measuring what AI returns covers what to record once a use is live.
The case for moving faster
A board member reading these figures has a fair point. The Census count puts large US firms at 43.6%, Eurostat counts 55.03% of large EU enterprises, and a company that waits for perfect evidence will wait a long time. Speed has a value of its own.
We agree that speed matters, and none of the three steps needs to be long. An inventory is a list. A test of one task needs a set of real cases and the correct result for each. What the steps change is the reason for each expansion. A company that expands because a test passed knows what it expanded. A company that expands because a survey figure went up knows only that other firms answered yes to one question.
The guide on enterprise AI adoption covers the rest of the program: training, governance, security, data and older systems.
Reveneau is an AI software development consultancy with offices in San Francisco and Vancouver. Reveneau says AI writes all of its code and every change must pass an eval suite written from the specification. It says Jev, TypeSafe AI's decision model, grades that suite, and that the suite runs ten times faster than it did with its previous grader. Reveneau says it uses AI instead of hiring more engineers and puts that saving into clients' prices.
Credit for this piece goes to the statisticians at the US Census Bureau, Eurostat and the ONS, who publish their tables, standard errors and wording changes for anyone to check, and to the researchers whose field studies we quote.
A survey counts the firms that tried AI. A test on your own work shows whether more of it will help.
Sources
- US Census Bureau, Business Trends and Outlook Survey data, National, Employment Size Class and Sector tables, period 5 to 18 October 2026, published 22 October 2026
- US Census Bureau, BTOS AI Supplement Table 2026, collected 17 November 2025 to 8 February 2026, published 23 April 2026
- US Census Bureau, BTOS AI Core Question Updates, 3 December 2025
- The White House, Executive Order 14434, Inaugurating the Era of Super Intelligence, 29 September 2026
- Federal Reserve Board, Monitoring AI Adoption in the US Economy, FEDS Notes, 3 April 2026
- Eurostat, Use of artificial intelligence in enterprises, data extracted December 2025
- Office for National Statistics, Artificial intelligence in UK businesses: 2023 to 2026, 20 July 2026
- Stanford HAI, The 2026 AI Index Report: Economy
- Erik Brynjolfsson, Danielle Li and Lindsey Raymond, Generative AI at Work, Quarterly Journal of Economics, published 4 February 2025
- Fabrizio Dell'Acqua and colleagues, Navigating the Jagged Technological Frontier, Harvard Business School Working Paper 24-013, 22 September 2023
- Anders Humlum and Emilie Vestergaard, Still Waters, Rapid Currents: Early Labor Market Transformation under Generative AI, NBER Working Paper 33777, revised March 2026
