On October 22, 2026, two credible sources answered the same question on the same day. Goldman Sachs said 86% of the small businesses it surveyed use AI. The Census Bureau said 24.6% of US employer businesses used AI in the last two weeks.
Both numbers are right. They count different businesses and different kinds of use. The number I care about is smaller than either: how many companies have handed AI an actual job. I run hundreds of AI agents across my own companies and client companies, and from where I sit that gap is the whole story.
How many small businesses use AI in 2026?
Between a quarter and nearly nine in ten US small businesses use AI in 2026, depending on who asks. Census puts it at 24.6% of employer businesses (October 2026). The Federal Reserve found 46% of small employer firms, the US Chamber of Commerce 66%, and Goldman Sachs 86% among alumni of its own small-business program.
Here are the four small business AI adoption surveys side by side, newest data first.
| Survey | Who was asked | When | What counted as “using AI” | Result |
|---|---|---|---|---|
| Census Bureau BTOS | A sample of US employer businesses of every size, weighted to the whole population | Collected Oct 5–18, 2026 | The business used AI “in any of its business functions” in the last two weeks | 24.6%; 28.6% expect to within six months |
| Goldman Sachs 10,000 Small Businesses | 2,473 alumni of Goldman’s small-business program | Aug 10–24, 2026 | Owners “report using AI in their business” | 86% |
| US Chamber of Commerce | US businesses with fewer than 250 employees, online | Jul 23–Aug 6, 2026 | Ticked any AI tool on a list of technologies “you use at work” | 66% |
| Federal Reserve Small Business Credit Survey | 5,248 employer firms with 1–499 employees answered the AI questions | Sep 3–Nov 14, 2025 | “The business or any of its employees” using AI for work, including AI features in software they already use | 46% |
Every one of these is a real survey run by a serious organization. None of them is lying. They answer four different questions.1
Why do small business AI statistics disagree so much?
Small business AI statistics disagree because each survey asks different businesses a different question. Goldman surveyed graduates of its own application-only growth program. The Fed counted a firm if any employee used ChatGPT. Census asks whether the business used AI in its operations in the last two weeks. Change any of those and the number moves.
The Federal Reserve Board’s Jeffrey S. Allen compared three major AI adoption surveys, the Census one among them, in an April 2026 FEDS Note. His summary: “The biggest driver of variation in these estimates likely relates to differences in sampling distributions and units of analysis, but question framing, the materiality of reported usage, information asymmetries between different target respondents, and social desirability bias may play a role as well.”
In plain terms, four things move the number.
Who was asked. Goldman’s 86% comes from its October 2026 report, “The AI Economy Is a Small Business Economy,” a survey of alumni of 10,000 Small Businesses. That’s an application-only program for owners Goldman says are “poised to grow”. The survey went to 14,967 alumni and 16.5% answered. Goldman’s own report says its graduates are “outpacing the average U.S. small business owner,” and the source it cites for that average is the Census survey. Census, meanwhile, weights its sample to every employer business in the country, and most of those are tiny.
What counts as use. The Fed counts the business if “the business or any of its employees” use AI for work, and its definition includes “features of software you already use.” One bookkeeper drafting emails in ChatGPT puts the whole firm in the 46%. Census asks whether the business used AI “in any of its business functions,” which, Allen suggests, respondents may read as something more deliberate.
Wording alone can nearly double a result. In November 2025, Census changed its question from AI used “in producing goods or services” to AI used “in any of its business functions.” The national rate went from 10.0% on the old wording to 17.3% on the new one, and Census started a new series rather than pretend the two connected.2
The time window. Census asks about the last two weeks. The Fed and the Chamber ask about use now, with no fixed window, which catches the owner who tried ChatGPT last month. Goldman doesn’t publish its question.
Who answers. A Census working paper found owners and top executives report more AI use than finance or admin staff answering for the same kind of firm. Allen also flags social desirability: senior leaders “may face pressure to report AI usage as an efficiency initiative.”
Which small business AI statistic should I trust?
Trust the Census Bureau’s number for the share of US businesses using AI, about one in four in October 2026, and the Fed’s for how many small employer firms have staff using it at all, about half. Read Goldman’s 86% as what growth-focused owners look like, not as the national average.
I didn’t come up with the Census half of that ranking. Allen’s note says it directly: “The BTOS is the best source for an estimate of the percentage of U.S. businesses that have adopted AI.” The Census figure also climbs with company size. In the same October release, 24.3% of businesses with one to four employees used AI, against 29.0% at 50 to 99 employees and 43.6% at 250 or more.
Even at 250 employees and up, most businesses don’t report using AI in their operations yet.
How are small businesses actually using AI?
Small businesses mostly use AI for writing. In the Fed’s survey, 83% of small employer firms that use AI use it for writing or marketing, and 21% use it for process automation. Census found 85.4% of firms whose employees use generative AI use it for writing or editing documents, emails or communications.
The Fed’s full list, as a share of firms using AI: writing or marketing 83%, individual productivity 61%, planning or analysis 51%, administrative business functions 38%, customer service 29%, process automation 21%, coding 19%.
Goldman’s alumni look similar. The most common uses were productivity at 66% and written marketing content at 65%. The tools are the ones your staff already have open: 86% of Goldman’s AI users rely on chatbots such as ChatGPT, Claude or Gemini, and 54% on AI built into software they already use. Only 18% use agentic AI, the kind that carries out tasks on its own.
One more number from that report worries me more than any adoption rate. Of owners who use AI, 74% have no formal policy on how their employees use it. If your staff paste customer data into a personal ChatGPT account, that’s where I’d start; I wrote up how I set agent permissions in AI agent security.
How much do small businesses spend on AI?
Most small businesses spend very little on AI. Goldman found 73% of its AI-using owners use free tools or spend under $100 a month. Chase’s bank data puts the median small business paying for AI at about $30 a month in 2025. The Chamber found a median of $250 a month among businesses that pay.
The JPMorganChase Institute number is the most objective of the three, because it comes from actual payments rather than survey answers. By December 2025, 17.7% of the small businesses banking with Chase, sole proprietors included, were paying for an AI service. Median spending “peaked at approximately $80 per month in 2022 before declining to roughly $30 per month by 2025.” Payment data can’t see free tools, which is part of why it lands near the Census number. In Goldman’s survey, 16% of AI users use only free tools.
For comparison, my AR (accounts receivable) agent costs me less than $200 a month to run, and it does one job: it applies client payments every Monday and drafts the collection reminders I send. Most of Goldman’s owners spend less than $100 a month on all of their AI put together. I broke down what each of my agents cost to run in October in How much does an AI agent cost to run per month?
A $20 chatbot subscription buys a tool. A job costs more to run, and the running cost is the small part. The setup is where the money goes: writing down what the job is, connecting the systems it touches, and deciding what waits for a person.
How many small businesses have fully integrated AI?
Very few small businesses have fully integrated AI. Just 7% of small employer firms using AI told the Fed it is fully integrated. Goldman found 15% of its alumni have fully integrated AI into core operations. Census found 10.1% of AI-using firms had AI doing a task an employee used to do: under 3% of all US employer businesses.3
Goldman names the gap itself. Its report says “The gap between adoption and integration is now the defining frontier for small businesses,” and its own breakdown of AI users shows it: 18% fully integrated, 50% “actively using AI in some areas, not yet central to the business,” and 32% “experimenting or dabbling.”
The Census supplement is blunter. Of firms using AI, 64.3% said they “did not make any changes to use AI.” Only 15.4% developed new workflows, and 3.8% used vendors or consultants to install or integrate it.
The businesses I meet with fit that picture. Of all of them, only one had AI agents in production. In the rest, the team already uses ChatGPT, often every day, and nobody has handed it a job. By a job I mean a named piece of recurring work, with access to the systems it needs, written rules for what waits for a person, and a check that it got done. The difference between those two states is the distance between Goldman’s 86% who use AI and the 15% of its alumni who have it fully integrated.
When it works, it looks like work. Goldman quotes a medical clinic owner in Georgia: “We built AI agents that audit our medical billing every night, flag denials and underpayments, build worklists for our biller, and call insurance payers about stuck claims using an AI voice agent.” At a 53-person travel company, 13 agents run 17 jobs that start on their own, and in one month they finished 290 tasks.
What should a small business owner do with these numbers?
Ignore the headline adoption rate; your staff almost certainly use AI already. The question worth asking is the smaller one: has anyone handed it a job? Pick one recurring piece of desk work, write down how it’s done and what should wait for you, and give it to an agent.
A few things I’d take from the data.
- You aren’t behind on adoption. By Census’s measure, three in four US businesses didn’t report using AI in their operations in the last two weeks. By the Fed’s, half of small employers have nobody using it. If your team uses ChatGPT daily, you’re already ahead of the average.
- You are probably early on integration. In the two surveys that ask, fewer than one in five AI-using businesses has fully integrated AI (18% at Goldman, 7% at the Fed). That’s the useful place to be early.
- Write a policy before you write a prompt. Three in four of Goldman’s AI-using owners have none.
- Keep the job yours. Goldman’s new Small Business AI Institute partners with Anthropic, Google, Meta and OpenAI. Learn from all of them, and keep your job descriptions in plain files you own. Sometimes Claude doesn’t do a good job and I move an agent to an OpenAI Codex model; my setup allows it because the job isn’t locked inside one vendor’s app.
- Some jobs can stay in ChatGPT. A daily news digest needs no company context and no company tools. It doesn’t need an agent.
If you want to see what that first job could be, I wrote about how to hire your first AI agent and when ChatGPT is enough. The next Census reading, due November 5, will be the first collected under its new wording, and Census hasn’t said whether that will move the 24.6%. The 7% won’t move until owners hand AI real work.
If you’d rather work through it with your team, Fifth Nine runs a free session for that: Put AI to Work.
Sources
How this was written: drafted with AI from my notes and Fifth Nine’s operating records, then edited and approved by me.
Footnotes
-
Goldman’s figures are from its October 2026 report “The AI Economy Is a Small Business Economy,” fielded by Babson College; Goldman doesn’t publish the question wording, and the report says its figures “may not be representative of all program alumni.” The 86% is a different survey from Goldman’s March 2026 10,000 Small Businesses Voices poll (1,256 participants, January 27 to February 4, 2026), which found 76%, so the two don’t form a trend line. Goldman’s own alumni series runs 56% (July 2024), 77% (2025) and 86% (2026). The Chamber’s survey was run by Teneo Research; the report cites 3,732 small businesses but doesn’t give the 2026 wave’s size. The Fed’s survey is a convenience sample weighted to Census data, and its report says “The SBCS is not a random sample.” ↩
-
Census documented the change in AI Question Wording Updates (December 3, 2025). The 10.0% is the last old-wording reading (collected September 22 to October 5, 2025) in Census’s original AI series; the 17.3% is the first new-series reading (collected November 17 to 30, 2025). Since October 22, 2026, Census also publishes the series under the label “Super Intelligence or SI (formerly referred to as Artificial Intelligence or AI),” following Executive Order 14434. The 24.6% was collected with the “Artificial Intelligence (AI)” wording; the relabeled question went into the field on October 19, 2026, and its first results are scheduled for November 5. ↩
-
The Census figures come from the 2026 BTOS AI supplement, fielded November 17, 2025 to February 8, 2026, in which 17.9% of firms used AI in the last two weeks. The follow-up questions went to firms that used AI in at least one of 15 business functions in the last six months: 27.7% of all firms, per Census’s working paper on the supplement. The “under 3%” is my arithmetic: 10.1% of that 27.7% is about 2.8% of all employer businesses. The Fed’s 7% is a share of AI-using firms; Goldman’s 15% is a share of all alumni surveyed, which is 18% of its AI users. ↩
Put AI to work with your team.
A free session using one real workflow from your business. Start with a 30-minute planning call.