On this pageJump to a section
Two numbers describe the same thing, and they are 41 points apart.
The U.S. Chamber of Commerce says 58% of small businesses use generative AI. The Census Bureau's survey, read through the Economic Innovation Group, puts the national rate of businesses using AI at 17.3%. If you only saw one, you would walk away with a very different idea of whether you are late. The honest answer is that both numbers are real, they answer different questions, and neither one tells you whether AI will pay off in your business.
Why do the numbers disagree?
Because "use AI" is a slippery phrase.
The Chamber's number
The Chamber's 2025 Empowering Small Business Report, based on a survey by Teneo Research and published August 18, 2025, found that 58% of small businesses use generative AI, up from 40% in 2024 and 23% in 2023. It also reported that 82% of small businesses using AI increased their workforce over the past year, and that 77% of AI-using businesses say limits on the technology would hurt their growth.1 Those last figures are what businesses told a survey, not outcomes anyone audited.
58%
| Label | Value |
|---|---|
| 2023 | 23% |
| 2024 | 40% |
| 2025 | 58% |
Source: U.S. Chamber of Commerce, Small Business Use of AI Surges, Driving Daily Efficiency
The Census number
The Census Bureau's Business Trends and Outlook Survey has measured AI use for longer. In an early paper, researchers found the share of firms using AI rose from 3.7% in September 2023 to 5.4% in February 2024.3 Economist Nathan Goldschlag wrote in December 2025 that under the survey's original wording, AI use "in the production of goods or services," the rate rose from 4.6% at the start of 2024 to 10% in September 2025. When the question was broadened in early November 2025 to AI use "for any business function," the national rate was 17.3%.2
17.3%
Same year, different rulers
Goldschlag notes that the narrower wording helps explain why the survey's rates are lower than other business and individual surveys.2 Add differences in who was asked and how, and the gap makes sense. One survey asks small businesses whether they use generative AI. The other asks businesses of all sizes whether AI is part of any function.
| U.S. Chamber survey | Census BTOS |
|---|---|
| Small businesses, generative AI | All businesses, AI in any business function (since November 2025) |
| 58% in 2025 | 17.3% when the wording was broadened |
| Conducted by Teneo Research | Run by the federal statistical agency |
| Self-reported use | Self-reported use, with a consistent series |
What are small businesses actually using it for?
Mostly the unglamorous things.
What the Census data shows
In the Census Bureau's early data, the most common uses were marketing automation, virtual agents and data and text analytics. The paper also found that AI users showed better overall performance and more frequent employment expansion than other businesses, that many firms made organizational changes such as training and new workflows, and that few reported employment reductions. The most common reason for not adopting AI was that it did not apply to the business.3
What Anthropic's usage data adds
Anthropic's September 2025 Economic Index looks at how its own Claude product is used, not the whole economy, and says so. Among its findings: coding was the largest category on Claude.ai at 36%, and in enterprise API traffic, 77% of transcripts showed automation patterns versus 12% augmentation, with marketing materials accounting for about 4.7% of API use.4 It is a window into one company's customers, not a census of small business.
Where does adoption pay, and where is it experimentation?
I have to be straight about the limits first: none of these sources measures return on investment by use case, which is part of why so many pilots stall (see the 95% pilot failure headline). They tell you what people do and what they report. What I can offer is an operator's reading.
Where it tends to pay
Work that is repeated, checkable and low-risk if slightly off. Drafting a first version of a routine email. Summarizing a call. Cleaning a spreadsheet. Pulling together a weekly report. The pattern is that a person can verify the output in seconds, and mistakes cost little.
Dan Martell made this point about follow-up in a video he posted on November 29, 2025. He is an investor in some of the tools he names there, so weigh the pitch accordingly, but the observation holds up on its own.5
Humans will 100% forget to follow up, but AI doesn't.
Follow-up is repeated, easy to check and cheap to get slightly wrong, which is exactly the profile above. I would still keep a person approving the messages, and so does he.
Where it is still mostly experimentation
Work where the output is hard to check, the stakes are high or the process is fuzzy. Writing claims in a regulated category. Making pricing decisions. Handling upset customers without review. Those can get faster too, but they need more design, review and rules (see where AI stops and a human starts) before you can call them a gain.
What should an operator do with this?
Ignore the headline and run your own numbers.
- List your five most repeated tasks and how long each takes you per week. Writing them down as SOPs first makes this step easy.
- Mark which ones have output you can check in under a minute.
- Pick one. Define success before you start, such as hours saved or response time.
- Run it for four weeks and log what you saved and what you had to fix.
- Keep, change or drop it. Then pick the next one.
The adoption race is not the point
Being early does not matter if the thing you adopted does not help. Being late does not matter either if you pick the right two or three tasks and do them well. A small business is not competing with a survey percentage. It is competing for customers, and the question is whether AI makes you faster, cheaper or better at serving them.
My read
I use AI agents heavily in my own solo work and in the system I build for clients, so I am not speaking as a skeptic. I am saying that adoption is not the same as advantage, and a survey cannot tell you which one you have.
The question is never "are we using AI." It is "what did it make cheaper or better, and how do we know."
Keep building.Ekai
Questions people ask
What percentage of small businesses use AI?
It depends on the survey. The U.S. Chamber of Commerce's August 2025 report, conducted by Teneo Research, found 58% of small businesses use generative AI. The Census Bureau's survey, under a broader wording introduced in November 2025, showed a national business AI use rate of 17.3%, according to the Economic Innovation Group.
Why do AI adoption surveys disagree?
They ask different questions of different groups. The Census Bureau's earlier question asked about AI used in producing goods or services, which the Economic Innovation Group describes as restrictive, and which helps explain why its rates are lower than other surveys.
What do small businesses use AI for?
In the Census Bureau's early BTOS data, the most common uses were marketing automation, virtual agents and data and text analytics. The most common reason for not adopting was that AI did not apply to the business.
Does AI adoption increase profits for small businesses?
The data cited here shows association, not proof. The Census paper found AI users showed better overall performance and more employment expansion, but it does not establish that AI caused those results.
Sources
- 1
- 2How Many Businesses Are Using AI?
Economic Innovation Group (Nathan Goldschlag), Dec 19, 2025123
- 3
- 4
- 5
