Quick answer
In the Census survey of 24 August to 6 September 2026, national AI use runs from 24% (95% CI 23% to 24%) for 1-4 employees to 44% (95% CI 40% to 48%) for 250 or more, nearly flat below 50 (exploratory reading). Industry matters more: all-sizes rates run from 10% in accommodation and food services to 49% in information.
Key takeaways
- National rates by size, Census, 24 August to 6 September 2026: 24%, 21%, 23% and 24% for 1-4, 5-9, 10-19 and 20-49 employees, then 29%, 34% and 44% for 50-99, 100-249 and 250+.
- Industry differs widely: among the 16 industries with published size cells, the all-sizes rate runs from 10% in accommodation and food services to 49% in information.
- National rates are nearly unchanged from 1-4 to 20-49 employees and rise above 50 (an exploratory reading); inside one industry, larger small businesses tend to report a few points more.
- The numbers describe groups of businesses, not the chance that your business uses AI, and 30 of 133 industry-by-size cells are not published.
What share of small businesses use AI?
In the Census Bureau's survey for 24 August to 6 September 2026, 24% (95% CI 23% to 24%) of US employer businesses reported using AI, and the all-sizes rate runs from 10% in accommodation and food services to 49% in information. That range is among the 16 industries with published size cells, and the national rate by size, an exploratory reading, is nearly flat below 50 employees and rises above. The 1-4 to 250+ gap is 20 percentage points (95% CI 17 to 24). Inside one industry, larger small firms tend to report somewhat more.
A "95% CI" is a confidence interval: the range the true figure very likely falls in, since the Census surveyed a sample. The source is the Census Business Trends and Outlook Survey, which asks whether the business used AI "in any of its business functions" in the last two weeks.
Does AI use depend on how many employees a business has?
Yes, as groups, but not in a smooth line: the national rate is nearly unchanged from 1-4 to 20-49 employees and rises above 50, an exploratory reading. These are the Census's own national rates by size class for that two-week period.
| Employees | Share reporting AI use (national) |
|---|---|
| 1-4 | 24% (95% CI 23% to 24%) |
| 5-9 | 21% (95% CI 20% to 22%) |
| 10-19 | 23% (95% CI 22% to 24%) |
| 20-49 | 24% (95% CI 23% to 26%) |
| 50-99 | 29% (95% CI 27% to 32%) |
| 100-249 | 34% (95% CI 31% to 37%) |
| 250+ | 44% (95% CI 40% to 48%) |
The 20-49 class is 0.7 percentage points (95% CI -0.7 to 2.1) above the 1-4 class, while 250+ is 20 percentage points (95% CI 16 to 23) above 20-49. The split at 50 employees was chosen after seeing the results, so treat it as exploratory. Inside one industry the picture differs: larger small businesses tend to report a few points more. The unweighted class means rise from 22% at 1-4 to 29% at 20-49, and the difference is positive in 14 of 15 industries that publish both cells.
What does AI use look like in my industry and size class?
It depends heavily on the industry: even within one size class the published cells differ by tens of points. The tables show five common small-business industries. Each cell reads "in the group of businesses with this many employees, in this industry, x% reported AI use." To read it, find your industry row, read across to your size, and look at the interval before the point figure. Wide intervals, mostly in the larger size classes, mean a rough guide only.
| Industry | 1-4 | 5-9 | 10-19 | 20-49 |
|---|---|---|---|---|
| Construction | 13% (95% CI 11% to 16%) | 14% (95% CI 12% to 16%) | 23% (95% CI 20% to 27%) | 22% (95% CI 18% to 26%) |
| Health care and social assistance | 24% (95% CI 21% to 28%) | 25% (95% CI 21% to 29%) | 29% (95% CI 26% to 32%) | 27% (95% CI 24% to 31%) |
| Accommodation and food services | 10% (95% CI 8.2% to 13%) | 5.9% (95% CI 4.2% to 7.6%) | 10% (95% CI 7.7% to 13%) | 11% (95% CI 8.2% to 13%) |
| Professional, scientific and technical services | 43% (95% CI 41% to 45%) | 44% (95% CI 40% to 47%) | 47% (95% CI 43% to 50%) | 50% (95% CI 41% to 59%) |
| Other services (repair, personal care) | 14% (95% CI 12% to 16%) | 13% (95% CI 11% to 16%) | 18% (95% CI 14% to 23%) | 26% (95% CI 17% to 35%) |
| Industry | 50-99 | 100-249 | 250+ |
|---|---|---|---|
| Construction | 24% (95% CI 9.8% to 39%) | 33% (95% CI 15% to 51%) | 46% (95% CI 8.1% to 83%) |
| Health care and social assistance | 26% (95% CI 19% to 34%) | 28% (95% CI 18% to 39%) | 38% (95% CI 27% to 50%) |
| Accommodation and food services | 13% (95% CI 11% to 15%) | 23% (95% CI 9.9% to 36%) | 30% (95% CI 15% to 44%) |
| Professional, scientific and technical services | 55% (95% CI 44% to 65%) | 59% (95% CI 50% to 68%) | 56% (95% CI 43% to 70%) |
| Other services (repair, personal care) | 32% (95% CI 15% to 50%) | not published | not published |
The largest standard error among published cells is 19 points. Among 1-4 employee businesses the lowest and highest published cells are utilities at 10% (95% CI 0.9% to 19%) and information at 46% (95% CI 40% to 52%).
Which cells does the Census not publish?
The Census published 103 of 133 possible industry-by-size cells, so 30 are missing, and you will find no estimate filled in for them. There are 19 industries and 7 size classes. Agriculture, forestry and fishing; mining, oil and gas; and management of companies have no published cell.
The missing cells are mostly in small industries: in Statistics of U.S. Businesses 2022 the median published cell holds 14,444 firms, against 1,279 for a suppressed one (counts for the two largest size classes are undercounts). If your trade is small, your cell may not exist.
Is it a straight line from small to large?
No single straight line describes it, and no one slope is the right gradient for a business under 50 employees. A slope here is how many percentage points AI use rises each time business size grows tenfold, for example from 5 to 50 employees. Weighting means giving more say to the groups the survey measured more precisely.
Fitted across all 103 cells with one line, AI use differs by 11.13 percentage points (95% CI 8.23 to 14.04) per tenfold difference in size; within one industry it is 10.74 (95% CI 7.71 to 13.78). Weighting by precision changes this a lot: 1.08 across cells (95% CI -5.63 to 7.80) and 5.94 within industries (95% CI 4.10 to 7.79).
The exploratory split at 50 gives a within-industry slope of 5.50 (95% CI 2.95 to 8.04, n = 61 cells) for 1-4 to 20-49 employees and 17.61 (95% CI 6.71 to 28.51, n = 42 cells) from 50-99 up. The upper slope depends on the value 400 given to the open-ended 250+ class.
Takeaway: the straight-line figures overstate the difference for businesses under 50 employees, where the national rates for 1-4 and 20-49 differ by about a point (0.7 percentage points, 95% CI -0.7 to 2.1).
Does the size gap differ by industry?
You cannot tell from this data: it neither shows nor rules out that the gap between 250+ and 1-4 employees differs by industry. The test is weak: industry gaps have standard errors of 6 to 19 points. The gaps themselves run from 12 percentage points (administrative and support services) to 46 percentage points (transportation and warehousing).
Is pay in an industry associated with its AI use?
Only as a secondary, fragile industry-level association: among 16 industries, those with higher payroll per employee tend to report more AI use at a given size. It rests on 16 pay values (Statistics of U.S. Businesses, 2022, against survey data from 2026), a straight-line model that fits imperfectly, and a pattern driven largely by information, finance and professional services.
The estimate is 49.93 percentage points (95% CI 29.72 to 70.14) per tenfold difference in payroll per employee, and the interval is probably too narrow. Dropping those three industries leaves a much smaller estimate.
What can this data not show?
It cannot show how likely your business is to use AI, whether AI helps, or what happens to a business as it grows.
- Each figure is a share of a group, not a probability for one firm.
- "Used AI" is a self-reported yes for the last two weeks. It says nothing about how much, for which task or with what result, and "do not know" answers count in the denominator.
- Only employer businesses are covered; a one-person business without payroll is not in the survey. The response rate is low and the Census weights for it, but non-response bias cannot be checked from public files. No Census page read for this article states a response rate for this period, so none is printed.
- It is one two-week period, 24 August to 6 September 2026, on the current question wording only.
- The data are observational, so these are associations only.
- Cells in one industry resemble each other, so the 103 cells carry about as much independent information as 25 (effective n = 25). The residuals are also not normally distributed, which the intervals ignore.
How can I check these figures?
Download the Census files and read the cell for your industry and size. They are the sector by employment size class, employment size class and sector workbooks, plus Statistics of U.S. Businesses 2022 for payroll and firm counts.
Analysis by Deivy Hernandez, from the Census Bureau's published files. Method: intervals use the Census's published standard errors; slopes are fitted with errors clustered by industry.
Tests, stated once. The first two hypotheses (size, then size within industry) were confirmatory in form, but the direction was visible beforehand: the analyst had read the 2026 cell values before freezing the protocol. Both gave p < 0.001; the estimates and intervals are unaffected.
Averaged over six survey periods, the 20-49 minus 1-4 difference is 1.1 percentage points (95% CI 0.3 to 1.9). A squared term added to the within-industry model is positive (4.64, 95% CI 1.44 to 7.83), descriptively consistent with a steeper climb among larger classes. For the 250+ minus 1-4 gap across 14 industries there is no evidence it differs by industry (Q = 10.49, p = 0.654), but the test has little power. A six-period version gave p < 0.001, a more cautious one p = 0.215. The split at 50 was not corrected for multiple tests.
What should I do with these figures?
Treat your cell as a group figure, not a forecast for your business. Once you have your row, AI receptionist cost and AI agents for small businesses cover what a first tool costs and does.
FAQ
Is my business behind if it does not use AI?
This evidence cannot say. In the Census survey, 24% (95% CI 23% to 24%) of all US employer businesses reported using AI in the two weeks before the survey, so most did not. The data measure use, not results.
Does it matter that the survey only covers businesses with employees?
Only employer businesses are in it, so a one-person business without payroll is not covered. The figures do not describe them.
What if my industry has no published cell?
The Census published 103 of 133 industry-by-size cells and filled in nothing for the other 30. Agriculture, forestry and fishing; mining, oil and gas; and management of companies have no published cell at all. Missing cells are mostly in small industries, so a small trade may have no cell.
Are these figures comparable with older surveys of AI use?
No. The Census changed the wording of its AI question in late 2025. This article uses only the current wording.