What Percent of Small Businesses Actually Use AI?
Revised August 22, 2026
What percentage of businesses have adopted AI?
About 17 to 20 percent of U.S. businesses report using AI, according to the Census Bureau’s Business Trends and Outlook Survey covering December 14, 2025 through May 3, 2026, published May 26, 2026. A Federal Reserve FEDS Note dated April 3, 2026 put it at roughly 18 percent of firms at the end of 2025. Those are the two federal readings, and they agree.
Keep reading ↓Imagine it’s a Wednesday morning in Ballwin and you’re standing at the back counter with coffee going cold, reading an email that says 77 percent of small businesses now use AI. You put the phone down. An hour later a trade newsletter lands in the same inbox and says the real figure is about 18 percent. Same country, same year, same question, and a gap of nearly sixty points.
Maybe you’ve had the same whiplash. A contractor in Herculaneum gets a sales call that opens with “most of your competitors are already using this.” A shop owner in Waterloo reads a state-level roundup that says adoption is still in the single digits for firms her size. Both people were told a number. Neither was told what the number counted.
So here’s the honest version, with the sources and the dates attached, and then the part that actually matters: how to tell within about thirty seconds whether an AI adoption figure means anything at all.
What percentage of businesses have adopted AI?
About 17 to 20 percent of U.S. businesses report using AI, according to the Census Bureau’s Business Trends and Outlook Survey covering December 14, 2025 through May 3, 2026, published May 26, 2026. A Federal Reserve FEDS Note dated April 3, 2026 put it at roughly 18 percent of firms at the end of 2025. Those are the two federal readings, and they agree.
Both are firm-weighted. That word does a lot of work. Firm-weighted means every business counts once — the sole-proprietor lawn service in Jefferson County counts exactly as much as a company with 4,000 employees. Since the overwhelming majority of American businesses are tiny, a firm-weighted number is mostly a measurement of what tiny businesses are doing.
The Census survey is not a small poll. It goes to a rotating panel drawn from about 1.2 million U.S. businesses, with cohorts of roughly 200,000 firms surveyed every 12 weeks. That is the largest recurring read on this question anyone has. When a vendor tells you a number based on 500 respondents, this is the thing it should be measured against.
Why does every source give a different number?
Because they are counting different things. The Federal Reserve note from April 3, 2026 lined up three federal or academic surveys side by side and got 18 percent, 41 percent, and 78 percent — all defensible, all measuring something else. That spread is not a scandal. It is the answer.
The Business Trends and Outlook Survey asked firms whether the business used AI, and returned about 18 percent as of December 2025. The Real-Time Population Survey asked individual workers whether they use generative AI for their job, and returned 41 percent as of November 2025. The Survey of Business Uncertainty asked around 1,032 business executives and returned 78 percent — but that figure is employment-weighted, meaning it approximates the share of the workforce employed at firms that have adopted AI, not the share of firms.
Read those three again and notice that none of them contradicts the others. Most firms have not adopted AI. A large minority of workers use it anyway. And most employed people work somewhere that has adopted it, because big employers adopt first and big employers hold most of the jobs. The Fed’s own explanation names the cause plainly: the biggest driver of variation relates to differences in sampling distributions and units of analysis, with question framing, information gaps between respondents, and social desirability bias also in play.
That last one matters more than people admit. The Fed note observed that 10 to 11 percent of Census respondents answered that they did not know whether their firm used AI. If a tenth of respondents cannot answer the question, the confident single number you saw in a headline is doing a lot of pretending. There is a related and uncomfortable version of this problem covered in what happens when your staff adopts AI before the company does.
Does AI use rise with the size of the company?
Yes, steadily and without exception in the federal data. In the Census reading published May 26, 2026, less than 20 percent of firms with four or fewer employees reported using AI, 32 percent of firms with 100 to 249 employees did, and 37 percent of firms with 250 or more did. The article’s own headline says it: large firms with at least 20 employees are the biggest AI users.
The gradient is the most reliable finding in this whole subject. Every credible survey that breaks results out by headcount finds the same slope, and the slope is steep at the bottom. A 250-person company is roughly twice as likely to report AI use as a four-person one.
There is a second detail in that Census release worth more than the headline percentages. Between December 2025 and May 2026, AI use increased among firms with at least 20 employees but did not change significantly among firms with fewer than 20. The growth everyone is describing as explosive is happening above a headcount line that most St. Louis businesses sit well below.
By industry the same survey put Information at 39.7 percent, Finance and Insurance at 33.9 percent, and Retail Trade at roughly 14 percent as of May 3, 2026. If someone quotes you a national average and your work is retail or trades, the national average is not describing your industry.
Why did the AI numbers jump in November 2025?
Because the Census Bureau changed the question, not because businesses changed their behavior. Before November 2025 the survey asked whether the business used AI in producing goods or services. After November 2025 it asked whether the business used AI in any of its business functions. That single edit raised reported adoption by 47 percent to 159 percent depending on industry, per the Federal Reserve note of April 3, 2026.
The two ends of that range are worth sitting with. Professional services rose about 47 percent, an increase of 10.6 percentage points. Manufacturing rose about 159 percent, an increase of 7.5 percentage points. Nothing was installed. Nobody bought anything. The wording moved, and drafting an email with a chatbot went from not counting to counting.
This is why any AI trend line that crosses November 2025 is broken, and why a chart showing adoption “doubling” over the last two years should be treated as a chart of two different questions stapled together. Under the older, stricter wording, the U.S. Small Business Administration’s Office of Advocacy reported on September 24, 2025 that firms under 250 employees were at 8.8 percent as of August 2025, up from 6.3 percent six months earlier. Real growth, in a real series, at a scale nobody puts in a headline.
Hold those two facts next to each other. Genuine adoption among small firms moved about two and a half points in half a year. A question rewrite moved the reported rate by seven to eleven points overnight. The wording change did more to the number than the businesses did.
Wildly off-topic — your roots need doing too. Find a salon nearby.
Why do vendor surveys say 58 percent or 77 percent?
Because they ask a much easier question, of a self-selected group, and they count everything. The Intuit QuickBooks 2026 AI Impact Report, published May 12, 2026, found that 77 percent of U.S. businesses use AI regularly, up from 48 percent in July 2024. Read its own definition and the gap resolves: the report states that this includes any kind of AI use, from free tools and built-in features to dedicated paid subscriptions.
Built-in features. If your accounting software added an AI categorization button and you clicked it once, you are inside that 77 percent. That is a defensible thing to measure and it is not the same thing the Census measures. Intuit is also a company that sells AI features, which does not make the survey false but does belong in the sentence whenever the figure is quoted.
For scale, that report draws on responses from more than 34,000 business owners across the U.S., Canada, the U.K., and Australia, plus anonymized data from more than 5.3 million QuickBooks businesses, developed with economists at the University of Chicago. It is a serious piece of work. It simply answers a different question than “what share of American firms have adopted AI.”
The trade-association version lands in the same territory. The U.S. Chamber of Commerce Technology Engagement Center, working with Teneo Research, reported on August 18, 2025 that 58 percent of small businesses self-identified as using generative AI. Now put that beside the federal figure from the same month: 8.8 percent under the production-based wording. One month, one country, 58 percent versus 8.8 percent. Neither side lied. They asked different questions of different samples and let the press round the difference into a trend.
How do you read an AI statistic before you repeat it?
Ask five questions, in this order, and most AI statistics fall apart or become useful within about thirty seconds. Who ran the survey and do they sell AI. What exactly was asked. Who counts as one unit — a firm, a worker, or a job. What date does the figure describe. And does the series cross November 2025.
The unit question is the one people skip and it is the one that swings results hardest. Firm-weighted numbers are dominated by tiny businesses, so they read low. Employment-weighted numbers are dominated by large employers, so they read high. Worker-level numbers describe people, not companies, and a company can show up as a non-adopter while half its staff quietly uses a chatbot at home.
Apply the same test to the pessimistic statistics, not only the flattering ones. The widely repeated claim that 95 percent of generative AI pilots fail comes from a July 2025 MIT Media Lab Project NANDA report titled The GenAI Divide: State of AI in Business 2025, built on roughly 150 leader interviews, 350 employee surveys, and 300 public deployments. That is a preprint from a research lab, not a peer-reviewed census, and what it actually measured was pilots with no documented profit-and-loss impact — which frequently means no baseline was recorded before the pilot started. It is a real finding and a poor headline.
The same discipline works on the rules of thumb people quote at each other, which is its own small industry — there is a longer look at which AI rules of thumb survive contact with evidence. And if the number you actually need is a dollar figure rather than a percentage, the honest ranges live in what AI actually costs a small business.
What should a metro of mostly small firms conclude?
That the national headline number is not about you, and the low number is not permission to ignore the subject. Nearly every business in the St. Louis metro sits in the exact size bucket the Census reports at under 20 percent — and most sit below even that, in the no-employee tier.
The SBA Office of Advocacy 2025 Small Business Profiles, drawing on Census Nonemployer Statistics and Statistics of U.S. Businesses, count 590,131 small businesses in Missouri, which is 99.4 percent of Missouri businesses. Of those, 475,437 have no employees at all, 100,524 have 1 to 19 employees, and 14,170 have 20 to 499. Run the arithmetic and 97.6 percent of Missouri small businesses have fewer than 20 people.
Across the river the shape is identical. Illinois has 1,353,957 small businesses, 99.6 percent of Illinois businesses, of which 1,100,655 have no employees, 224,490 have 1 to 19, and 28,812 have 20 to 499. That is 97.9 percent under 20 employees. Small businesses employ 44.4 percent of Missouri workers and 43.7 percent of Illinois workers, so this is not a rounding error in the regional economy. It is the regional economy.
Two practical consequences. First, when a salesperson tells you that most businesses like yours already use AI, the federal data says otherwise for firms your size, and you can say so out loud. Second, the segment where adoption stalled between December 2025 and May 2026 is firms under 20 employees — so the competitive pressure on a four-person operation in Herculaneum or Waterloo is genuinely lighter than the trade press implies. That buys time to learn properly rather than to buy quickly. The workshop material behind the AI workshops for St. Louis business owners exists for that reason, and the broader overview in AI for St. Louis covers where a first hour is usually best spent.
From the owner’s side of the counter
If you run a small local operation, you already know why the adoption numbers look the way they do. Your constraint is not curiosity. It is that the person who would have to learn the tool is also the person answering the phone, running the truck, closing out the drawer, and doing payroll on a Sunday. Evaluation time is the scarcest input in the building, and it does not appear on any survey.
The seasonality makes it worse. Most local trades have two or three weeks that carry the year and one dead stretch where the thinking could actually happen — and the dead stretch is usually when cash is tightest, which is exactly when nobody signs up for anything. That is why owners in this position tend to complain to each other about the same thing: not that AI is useless, but that every pitch assumes a spare afternoon and a spare budget in the same month. The operators who get somewhere start with one repeating task, usually written follow-up or quoting, and give it four weeks before deciding.
One connected point, and it costs nothing. People search this metro by category and neighborhood constantly, and assistants now summarize whatever they can read about you. A complete, accurate listing is the cheapest version of being findable there, which is what a directory listing is for.
Tired of being sold a statistic instead of a skill? Browse AI workshops across the St. Louis metro on St Louis Near Me Directory, then bring the one repeating task in your week that eats the most hours — that is the task worth testing first.
Frequently asked questions
What is the 30% rule in AI?
There isn’t a standard one. No government agency or peer-reviewed body defines a 30 percent rule in AI, and the phrase is used for at least three unrelated ideas online: a limit on how much of a task to hand over, a training-data split, and a productivity claim. Treat any number-plus-rule phrasing as a slogan until someone shows you the study behind it.
Which 5 jobs will survive AI?
No credible source publishes a list of five. The Bureau of Labor Statistics projects occupational growth but does not rank jobs by AI survival, and anyone offering exactly five is writing a listicle, not research. What the data does support is that hands-on, licensed, on-site work in a specific location is the hardest category to automate, which describes most local trades.
What is the biggest user of AI?
By industry, Information leads at 39.7 percent, followed by Finance and Insurance at 33.9 percent, per the Census Bureau’s Business Trends and Outlook Survey as of May 3, 2026. Retail Trade sits near 14 percent. By company size, firms with 250 or more employees lead at 37 percent, roughly double the rate of firms with four or fewer.
What percentage of businesses have adopted AI?
Roughly 17 to 20 percent of U.S. firms, per Census data covering December 14, 2025 to May 3, 2026 and published May 26, 2026, with the Federal Reserve reporting about 18 percent at year-end 2025 in an April 3, 2026 note. Surveys reporting 58 to 77 percent are measuring workers, executives, or any use of any built-in feature.
Is AI being adopted by small businesses?
Yes, but slowly and from a low base. The SBA Office of Advocacy reported on September 24, 2025 that firms under 250 employees reached 8.8 percent as of August 2025, up from 6.3 percent six months earlier under the older survey wording. Census data shows firms with fewer than 20 employees saw no significant change between December 2025 and May 2026.
What is the failure rate of AI adoption projects?
The widely quoted 95 percent comes from a July 2025 MIT Media Lab Project NANDA report built on about 150 leader interviews, 350 employee surveys, and 300 public deployments. It counted pilots showing no measurable profit-and-loss impact, which often reflects missing baseline measurement rather than technical failure. It is a preprint from one lab, not a federal statistic.
How many companies are actually using AI?
In firm-weighted federal terms, about one in five or fewer. The Census Business Trends and Outlook Survey samples from roughly 1.2 million U.S. businesses in rotating cohorts of about 200,000 every 12 weeks, and returned 17 to 20 percent through May 3, 2026. Note that 10 to 11 percent of respondents answered that they did not know.
Who are the big 4 of AI?
There is no official designation, and the list changes depending on who is writing it — some group by consumer chatbots, some by cloud platforms, some by chip suppliers. For a small business the question rarely matters, because the practical choice is which tool fits one repeating task, at what monthly cost, with what happens to your customer data.
How many Fortune 500 companies use AI?
Nobody publishes an authoritative count of use, only counts of disclosure. The Conference Board and ESGAUGE reported on October 6, 2025 that 72 percent of S&P 500 companies flagged AI as a material risk in 10-K filings available through August 15, 2025, up from 12 percent in 2023. That measures what lawyers wrote, not what the company deployed.
Written by a dedicated team of St. Louis locals who live, work, and play right here in the St. Louis metro. Founder Lane Forman and team are committed to building the region’s most trusted directory by verifying listings and connecting local businesses with loyal customers across Missouri and Illinois.