Quick answer
On average across non-tech sectors, 13.4% of all businesses used AI in sales and marketing in the last six months. Among non-tech businesses that use AI, 8.1% (8.3% unweighted) say it did a task an employee used to do. Both show use, not value; the survey cannot say whether AI paid off.
Key takeaways
- Find your industry in the table: sales and marketing use runs from 2.6% (Mining) to 27.4% (Education), as a share of all businesses in each sector.
- Among AI-using businesses outside tech, 8.1% say AI did a task an employee used to do (8.3% unweighted). That is a share of AI users, not of all businesses.
- 14 of 16 non-tech sectors are at or above our 5% bar on this item. We set the bar after seeing the national values, so the test is confirmatory in form only.
- The survey measures AI use in general. It does not measure AI agents, and it does not show whether AI paid off.
What do businesses outside tech use AI for?
On average across non-tech sectors, 13.4% of all businesses used AI in sales and marketing in the last six months (13.0% unweighted). Among non-tech businesses that use AI, 8.1% (8.3% unweighted) say it did a task an employee used to do. The 8.1% is a share of AI users only. Both figures are averages across non-tech sectors, weighted toward the more precisely measured ones.
"Task substitution" is the survey's idea in everyday words: AI did something an employee used to do, as reported by the business. We use the plain phrase "task substitution" for it from here on.
The Census Bureau survey measures AI use in general, not AI agents, and it measures use, not value. It cannot tell you whether AI paid off for anyone. "Tech" here means two Census sectors: Information, and Professional, scientific and technical services.
What do businesses in your industry use AI for?
Find your row. Of the two functions in this table, sales and marketing is higher than customer service in every sector, and the range is wide. The first two columns are shares of all businesses in the sector over the last six months. The third is a share of businesses that use AI. The two rows marked (tech) are there for comparison and are not part of the non-tech averages.
| Sector | Sales and marketing (share of all businesses) | Customer service (share of all businesses) | AI did a task an employee used to do (share of businesses that use AI) |
|---|---|---|---|
| Education | 27.4% | 13.0% | 12.8% |
| Information (tech) | 25.5% | 12.2% | 16.8% |
| Real estate | 22.9% | 9.8% | 11.4% |
| Finance/insurance | 20.6% | 16.2% | 13.7% |
| Professional/scientific/technical (tech) | 17.7% | 10.0% | 15.5% |
| Arts/entertainment | 17.2% | 6.4% | 8.6% |
| Retail | 15.8% | 7.2% | 8.7% |
| Admin/support/waste | 15.0% | 8.0% | 9.8% |
| Wholesale | 14.9% | 6.2% | 7.7% |
| Manufacturing | 12.8% | 4.9% | 7.5% |
| Health care | 12.6% | 8.1% | 9.5% |
| Accommodation/food | 12.0% | 5.4% | 5.6% |
| Management of companies | 10.5% | 10.3% | no published figure |
| Other services | 10.1% | 6.6% | 6.5% |
| Construction | 7.7% | 4.4% | 5.0% |
| Utilities | 7.2% | 6.9% | 10.5% |
| Transportation/warehousing | 6.7% | 5.1% | 7.6% |
| Agriculture | 4.9% | 2.3% | 3.7% |
| Mining | 2.6% | 1.8% | 3.9% |
Each figure has a margin of error and describes a group of businesses, not one firm. The third column has a different base from the first two, so do not compare it with them. Census suppressed the third column for Management of companies as too noisy, and we did not fill it in. Every sector has a published figure in the first two columns.
How to read it: in the construction sector, 7.7% of businesses used AI in sales and marketing and 4.4% in customer service. In the health care sector, the figures are 12.6% and 8.1%. In retail trade, they are 15.8% and 7.2%. The figures are for each whole sector; the published table does not split by business size.
What does the national picture look like by function?
Nationally, of the three functions shown here, sales and marketing is the highest. Of all US businesses, 14.3% used AI in sales and marketing in the last six months, 11.4% in information technology and 7.6% in customer service. Non-tech sectors are at 13.4% for sales and marketing and 6.5% for customer service (95% CI 5.6% to 7.6%), weighted by how precisely each sector was estimated. For size effects, see our earlier piece on AI use by business size and industry.
Do businesses report that AI changed their total employment?
Among AI-using businesses, 2.3% reported an increase in total employment and 2.0% a decrease. Nationally, both are self-reported over the last six months, and both are shares of AI-using businesses. The answers of the remaining AI-using businesses (no change, or do not know) are not in our figures. The survey does not measure jobs lost.
Is California different?
California is somewhat higher than other states on task substitution, but not unusual among states. Among AI-using businesses, 11.5% in California report it, against 9.7% for other states, a gap of 1.8 percentage points (2.1 percentage points against an unweighted average of other states), and California ranks 7 of 47 states with a published estimate. That falls just short of the 3 percentage points we had set as the mark for a premium. The margin is narrow, and a different way of averaging the other states gives the slightly larger gap, so we do not treat a premium as settled either way. A state table cannot say anything about Silicon Valley or tech firms, because it covers single-state firms of every industry and has no state-by-sector cut.
Do sectors with more AI use also report more task substitution?
Exploratory only: across 18 sectors the two track each other (r = 0.95), but the link may be partly built in. "AI use" here is any AI use over the last two weeks as a share of all businesses, not a column in the table above. Sectors where that share is higher also report more task substitution among their AI users, and the task question is asked only of AI users. It is not evidence that more AI use is followed by more substitution.
What did we test, and how?
We tested one question: is AI "not really used" outside tech? We expected the answer to be yes. The direction of the result was visible before we fixed our method, so the tests are confirmatory in form only. We used the Census Bureau's Business Trends and Outlook Survey (BTOS) AI supplement, one cross-section collected from 17 November 2025 to 8 February 2026. No file measures agents or value, so we tested "not really used" and nothing broader.
We set two thresholds after national, sector and California values were visible: 5% for a sector to count as using AI for something, and 3 percentage points for California to count as a premium. The 5% bar is half the national task-substitution value of 10.1% that we had already seen, so clearing it is not a strong test. By that bar, 14 of 16 non-tech sectors clear it on task substitution and 15 of 17 non-tech sectors on sales and marketing. A margin of error (95% CI) is the range sampling error alone would give if the survey were repeated many times.
Technical detail: sector averages are weighted by precision, with the unweighted mean (8.3%) and median (8.2%) shown beside the weighted 8.1% (95% CI 6.7% to 10.0%); for sales and marketing the unweighted mean is 13.0% and the median 12.6%. The one-sample Wilcoxon test and the normal-theory z for California come from the scipy library. The California comparator is the weighted one; against an unweighted mean of other states (9.4%) the gap is 2.1 percentage points. The non-tech to tech ratio of 0.52 is exploratory. The California gap has 95% CI 0.7 to 2.9 percentage points from sampling error alone, which sits just below the 3-point bar, and p = 0.002 reflects sampling error, not importance.
Across n = 18 sectors r = 0.95 (95% CI 0.88 to 0.99), and r = 0.92 without the tech sectors (n = 16, 95% CI 0.80 to 0.97); the smallest detectable correlation is 0.62 and 0.65. Across n = 47 states r = 0.71 (95% CI 0.54 to 0.83), with low estimated reliability for state task shares. The tech sectors' weighted task substitution is 15.6%.
What can this data not show?
It cannot show whether AI paid off, and it measures AI use in general. The limits:
- It is BTOS data only, one cross-section, so it shows no change over time.
- Answers are self-reported by one respondent per business. Any use covers two weeks; functions and task substitution cover six months.
- Task substitution is asked only of AI-using businesses.
- Multi-state and multi-sector firms, typically large, are excluded from the state and sector tables.
- Suppressed cells (the task item for Management of companies, and the states Alaska, North Dakota, West Virginia and Wyoming) are not imputed.
- Estimates describe sector or state cells, not individual firms. An association across cells is not an association across firms.
- There is no measure of value, cost or quality. We tested "not really used" and cannot test whether AI adds value to a business.
- Our thresholds were set after values were visible, and results are associations only.
- An earlier 2023-24 supplement asked about "virtual agents or chat bots". It used different wording and a different survey round, so it is not comparable to 2026 and is not a trend.
How can you check it?
The inputs are public Census files. The 2026 values come from the 2026 AI supplement workbook. The earlier-round item is in the earlier supplement workbook, and response rates are in the unit response rate workbook. All survey files are listed on the BTOS data page.
What can you do with this?
Find your sector's row. The figure is the share of businesses in that sector that used AI in that function in the last six months. The survey says nothing about whether it paid off, or about businesses your size. For a plain guide to the tools themselves, see AI agents for small business.
Analysis by Deivy Hernandez, from public data files.
FAQ
Is AI replacing jobs at businesses?
This survey does not measure jobs lost. It asks about tasks and, separately, about total employment. Among AI-using businesses nationally, 2.3% reported AI increased total employment and 2.0% reported it decreased it, both self-reported over the last six months.
Can I multiply the sales figure by the task figure?
No. Function use is a share of all businesses. The task question is asked only of businesses that already use AI, so it is a share of AI users. The two have different bases and cannot be combined.
Does a low figure for my industry mean AI would not help me?
No. A low share means few businesses in that sector reported using AI there. The survey says nothing about results, cost or quality, so it cannot tell you whether it would help.
How current is this data?
The 2026 supplement was collected from 17 November 2025 to 8 February 2026. It is one snapshot, so it shows no trend.