Paloren Research
US business AI adoption, 2026
Census data shows business AI use climbing through 2026. The newest reading uses different wording, so we measure the trend on the series before it.
Key numbers
24.6%
We measured this in cycle 202621. The 90% interval runs from 24.2 to 25.0.
Census BTOS, reference period 21 Sep to 4 Oct 2026; new wording, see the note on wording
+6.5 points
Steady climb on the consistent series, from 17.3% to 23.8% across the run.
Census BTOS, 90% interval 5.8 to 7.2 points
9 states
Nine states.
Each state lower bound is above the national upper bound of 25.0%; Florida is borderline at 25.1%
+10.6 points
Finance and insurance is one of the largest sector jumps. It took about 11 months.
Census BTOS sector file, 90% interval 6.8 to 14.4 points, about 11 months
30.6%
We found this in the AI supplement. Firms under 50 employees land between 17.0% and 18.5%.
Census AI supplement table, 90% interval 29.2 to 32.0 for 250 or more; reference period not stated
What it means
What the headline figure says
Roughly one US business in four used AI in the two weeks before cycle 202621 closed. The estimate is 24.6%, with a 90% interval of 24.2 to 25.0%. The cycle before it read 23.8%.
One cycle is short.
A longer run tells you more about direction, and it is the series we trust most, so most of this study leans on it.
One in four US employers now uses AI somewhere, and the share keeps climbing. Use is not the same as value. The number I want next is how much of that AI output a person signs off without fixing it.
Aaron Agius, Co-founder, PalorenThe trend since late 2025
Census kept one wording for the run from cycle 202524 to 202620. Across it, business AI use rose 6.5 points, from 17.3% to 23.8%. The 90% interval around that rise is 5.8 to 7.2 points.
It wasn't a straight line. A few cycles slipped back. Still, the last one sits well above the first. Whether the newest cycle keeps the climb is harder to say, since its wording changed.
Direction matters more than any single reading.
Where use is higher
Nine states clearly sit above the national rate, with Utah and Colorado leading our estimates. Florida only just makes the cut: its lower bound is 25.1%, against a national upper bound of 25.0%.
We didn't adjust for testing 52 geographies, so treat the state list as a lead to check. Finance and insurance rose 10.6 points (90% interval 6.8 to 14.4), and manufacturing and health care moved almost as far. Information reports the highest use level among sectors with reliable estimates, at 46.4%.
Bigger businesses use more AI
Small firms lag.
In the AI supplement, businesses with fewer than 50 employees land between 17.0% and 18.5%. Firms with 250 or more employees reach 30.6%.
The supplement asks something close to the main question, but its reference period isn't stated. Use it to compare sizes, not to track change.
Plans and use
Six months before cycle 202621, 23.0% of businesses expected to use AI within six months. By cycle 202621, 24.6% were already using it.
That compares populations, not firms, so we can't say the same businesses planned and then adopted. The gap looks more like AI arriving through everyday software than through planned projects, and data permissions and review time may well be going uncosted.
More firms were using AI than had said they expected to six months earlier. That tells you AI is arriving through everyday tools, not planned projects, and that is exactly when data permissions and review time go uncosted.
Aaron Agius, Co-founder, PalorenWhy the UK figure is not compared
The UK survey asks a different question, so we keep it apart. In Wave 147 (15 to 28 December 2025), 67.4% of UK businesses said they weren't using any AI technology. Another 7.9% weren't sure.
The published file gives no UK adoption rate, and we haven't derived one. The two surveys differ in questions, weighting and timing, so we don't set their numbers side by side.
Charts
US businesses using AI in the last two weeks, cycles 202524 to 202620
- 17.3%Nov 2025
- 18.2%Jan 2026
- 18.9%Mar 2026
- 20.6%May 2026
- 21.5%Jul 2026
- 23.8%Sep 2026
Data table for US businesses using AI in the last two weeks, cycles 202524 to 202620
| Category | Value (%) | Low | High |
|---|---|---|---|
| Nov 2025 | 17.3% | 17% | 17.6% |
| Jan 2026 | 18.2% | 17.81% | 18.59% |
| Mar 2026 | 18.9% | 18.37% | 19.43% |
| May 2026 | 20.6% | 20.12% | 21.08% |
| Jul 2026 | 21.5% | 20.81% | 22.19% |
| Sep 2026 | 23.8% | 23.13% | 24.47% |
Six cycles, one wording. We pulled the full series into the downloads, and cycle 202621 (24.6%) stands apart because the wording changed under it. Source: Census BTOS national file, Response Estimates and Response Standard Errors; 90% intervals
States where business AI use is clearly above the US rate, cycle 202621
- Utah32.1%
- Colorado31.9%
- Arizona31.5%
- North Carolina29.6%
- Minnesota29.5%
- Maryland29.4%
- Virginia28.9%
- Texas28%
- Florida26.8%
Data table for States where business AI use is clearly above the US rate, cycle 202621
| Category | Value (%) | Low | High |
|---|---|---|---|
| Utah | 32.1% | 26.3% | 37.9% |
| Colorado | 31.9% | 29.3% | 34.5% |
| Arizona | 31.5% | 27.4% | 35.6% |
| North Carolina | 29.6% | 26.5% | 32.7% |
| Minnesota | 29.5% | 26.7% | 32.3% |
| Maryland | 29.4% | 27.2% | 31.6% |
| Virginia | 28.9% | 27.3% | 30.5% |
| Texas | 28% | 26.1% | 29.9% |
| Florida | 26.8% | 25.1% | 28.5% |
Order is for reading only. Bars run by estimate, but the intervals overlap heavily, so we don't claim a state ranking and we didn't adjust for testing 52 geographies. Source: Census BTOS state file, cycle 202621; 90% intervals
Change in business AI use by sector, cycle 202524 to cycle 202621
- Finance and insurance10.6 pts
- Manufacturing10.2 pts
- Health care9.5 pts
- Professional services9.1 pts
- Wholesale trade8.8 pts
- Construction8.7 pts
- Real estate8 pts
- Administrative support7.4 pts
- Retail trade6 pts
- Transportation5.6 pts
- Information5.4 pts
- Other services4.4 pts
- Arts and recreation3.4 pts
Data table for Change in business AI use by sector, cycle 202524 to cycle 202621
| Category | Value (pts) |
|---|---|
| Finance and insurance | 10.6 pts |
| Manufacturing | 10.2 pts |
| Health care | 9.5 pts |
| Professional services | 9.1 pts |
| Wholesale trade | 8.8 pts |
| Construction | 8.7 pts |
| Real estate | 8 pts |
| Administrative support | 7.4 pts |
| Retail trade | 6 pts |
| Transportation | 5.6 pts |
| Information | 5.4 pts |
| Other services | 4.4 pts |
| Arts and recreation | 3.4 pts |
Thirteen sectors clear zero. We show every sector whose 90% interval sits above zero, and the smallest rises, Arts and recreation and Information, only just clear it. Source: Census BTOS sector file, NAICS 2-digit, change over 12 months as published (about 11 months to cycle 202621)
Business AI use by employment size, AI supplement
- 1 to 4 employees17.6%
- 5 to 917%
- 10 to 1917.2%
- 20 to 4918.5%
- 50 to 9921.3%
- 100 to 24925.3%
- 250 or more30.6%
Data table for Business AI use by employment size, AI supplement
| Category | Value (%) | Low | High |
|---|---|---|---|
| 1 to 4 employees | 17.6% | 17.3% | 17.9% |
| 5 to 9 | 17% | 16.5% | 17.5% |
| 10 to 19 | 17.2% | 16.6% | 17.8% |
| 20 to 49 | 18.5% | 18% | 19% |
| 50 to 99 | 21.3% | 20.7% | 21.9% |
| 100 to 249 | 25.3% | 24.1% | 26.5% |
| 250 or more | 30.6% | 29.2% | 32% |
Close, but not the same. The supplement question is near the biweekly one, the file gives no reference period, and we have only one snapshot, so it shows size gaps rather than change over time. Source: Census BTOS AI supplement table 2026, employment response estimates; 90% intervals
Expected and actual AI use, US businesses
| Measure | Cycle | Share of businesses |
|---|---|---|
| Expected to use AI in the next six months | 202608 | 23.0% |
| Used AI in the last two weeks | 202621 | 24.6% |
| Difference (used minus expected) | +1.6 points |
Method and verification
We wanted to know how many US businesses use AI, who they are by state, sector and size, whether stated plans turned into use, and what the UK data shows.
US figures come from the Census Bureau Business Trends and Outlook Survey (BTOS): the national, state and sector biweekly files, plus the AI supplement table. "Use" is the Yes answer to the AI use question (Q7), and "expectation" is the Yes answer to the six-month question (Q24). Each estimate carries a 90% interval from the published standard errors.
The interval on the change between cycles 202524 and 202620 assumes independent samples. We dropped any cut with a coefficient of variation above 30%.
UK figures come from ONS BICS ad hoc table 3223, waves 92 to 147, and we report the published shares as they stand.
Sources
- Census BTOS national file (National.xlsx)
- Census BTOS state file (State.xlsx)
- Census BTOS sector file (Sector.xlsx)
- Census BTOS AI supplement table 2026
- Census BTOS methodology (V6)
- Census BTOS data downloads
- ONS BICS ad hoc table 3223 (AI adoption, waves 92 to 147)
Definitions and exclusions
Each figure is a published Census estimate with its standard error. Where Census flags a cell as unreliable, or its coefficient of variation tops 30%, we leave that cut out. We never fill a gap with an estimate of our own.
Limitations
- Wording break: cycle 202621 may use the new "Super Intelligence or SI" wording. Its level may not be strictly comparable with the earlier series.
- Multiple comparisons: 52 state and territory estimates are compared with the national rate without adjustment, so borderline results such as Florida may not hold.
- Expectation and use are population rates from different questions and windows. They don't track the same firms.
- The employment size cut comes from a supplement table with a similar question and no stated reference period. It isn't longitudinal.
- The UK and US figures aren't comparable. The UK file doesn't publish an adoption rate.
How we verified
An independent script recomputed every figure on this page from the cached Census and ONS files. A separate reviewer checked the results.
The cache manifest logs the URL, SHA-256 hash and download date of each file. We show figures to the precision the source publishes.
Data
Press kit
Media contact: aaron@research.paloren.ai
Interview: Aaron Agius, co-founder of Paloren, is available to talk through this study. Request a slot at aaron@research.paloren.ai.
About this study: Paloren Research is a small team that measures how firms adopt and describe AI, using public records. We publish the data and the limits with every number.
How to cite this study
Aaron Agius (2026). US business AI adoption, 2026. Paloren Research. Analysis of US Census Bureau Business Trends and Outlook Survey data, cycles 202524 to 202621. https://paloren.ai/research/us-business-ai-adoption-2026
FAQ
How many US employers use AI?
In cycle 202621, 24.6% of US businesses said they reported using AI in the last two weeks. The 90% interval runs from 24.2 to 25.0%.
Is AI use rising, or is the new wording the cause?
On the consistent series, use rose 6.5 points, from 17.3% in cycle 202524 to 23.8% in cycle 202620. Cycle 202621 uses new wording that may not compare, so we report it on its own.
Which states are clearly above the national rate?
Nine states. Florida is the closest call: its lower bound is 25.1%, just above the national upper bound of 25.0%.
Do bigger operators use more AI?
Yes. In the supplement table, 30.6% of firms with 250 or more employees used AI, while firms with fewer than 50 staff land between 17.0% and 18.5%.
Does this show that planned AI projects became real?
Not directly. At population level, 24.6% used AI during cycle 202621, and 23.0% expected to use it six months earlier. The survey doesn't follow the same employers, so it can't prove the plans came true.
Can I compare the UK figure with the US?
No. The UK file reports 67.4% not using AI in Wave 147 and gives no adoption rate. The surveys differ, so we don't compare them.