5 min read
86% of small businesses use AI. Or fewer than 20% do. Both figures were published in 2026 by serious organizations, and both are accurate for what they measured. The first comes from a Goldman Sachs survey. The second comes from the U.S. Census Bureau, and it describes firms with four or fewer employees, which is to say, businesses like yours.
If you’ve ever read a headline and felt quietly behind, this gap matters. Here are five beliefs about small business AI adoption that fall apart once you look at who was asked and what they were asked.
Myth 1: “Nearly every small business already uses AI”
Verdict: false for the smallest firms, at least on the best public data.
The Goldman Sachs report, as covered by ANI, found 86% adoption, up from 56% in 2024. But its sample was 2,473 alumni of the bank’s 10,000 Small Businesses program, and its own findings apply to those businesses, not all US small firms. These are owners who signed up for a growth program, so they’re hardly a random draw.
Thryv’s 2026 survey of 561 decision-makers put adoption at 66%, up from 55%. Thryv sells an AI-enabled marketing platform, so read it with that in mind, though it did use an independent research panel and excluded its own customers.
The Census Bureau’s Business Trends and Outlook Survey tells a different story. Between December 2025 and May 2026, 17% to 20% of businesses overall reported using AI, and under 20% of firms with four or fewer employees did.
Myth 2: “One of these numbers must be wrong”
Verdict: false. They’re answering different questions.
Three things drive the spread:
- Who was asked. Motivated program alumni, a panel of decision-makers, or a broad government sample.
- What counts as “use.” The Census asks whether a business used AI in any business function in the past two weeks. A survey asking “do you use AI?” with no time limit will catch an owner who tried ChatGPT once in spring.
- Whose use counts. An owner privately drafting an email with a chatbot may say yes to one survey and not think of it as the business “using AI” on another.
The useful reading is a range, not a single fact: plenty of owners have tried AI, far fewer rely on it, and the share depends heavily on how you ask. Treat any single adoption number the way you’d treat a confident answer from a chatbot: check where it came from before you repeat it.
Myth 3: “Adopted means AI is running the business”
Verdict: false, even among the most enthusiastic group.
The same Goldman survey that produced the 86% found that only 15% of surveyed firms have fully integrated AI into core operations, or about 18% of those using it. So in the most AI-forward sample available, more than eight in ten users still treat it as a helper around the edges.
That should be reassuring. The businesses in the headlines are mostly doing what you are doing: drafting, summarizing and experimenting. For a sense of how to go further with one job rather than everything at once, our story of a landscaper moving from dabbling to dependable shows the pattern.
Myth 4: “Comfortable with AI means good at AI”
Verdict: false, and owners say so themselves.
In the Thryv survey, 86% said they were comfortable using AI, and 70% said they need more training to use it well. Both can be true. Comfort is about whether the tool feels easy to open and use. Skill is about knowing when the answer is wrong, what not to paste in, and which jobs it shouldn’t do.
The same survey gives a hint about why: 57% learn from YouTube and social media, and 49% from online resources and webinars. Free learning is fine, but it tends to favor flashy demonstrations over boring checks. If you want a practical standard, write down what you will and won’t use AI for. Our one-page AI policy takes about as long as watching a few videos.
Myth 5: “If you’ve held back, you’re already too late”
Verdict: mostly false. Timing depends on your line of work.
The Census found AI use rose among firms with at least 20 employees but didn’t change significantly among smaller ones. Sector matters even more: about 40% of Information businesses reported using AI, against about 14% in Retail Trade.
And not adopting isn’t always ignorance. A San Francisco Fed brief on the 2024 Small Business Credit Survey found owners who held back cited concerns about accuracy and intellectual property, the importance of human contact to their business, or a belief that AI simply doesn’t apply to them. Those are business judgments, and some of them will be right.
That said, this isn’t an argument for ignoring AI. Thryv’s respondents reported real time savings, and the larger firms are moving. The point is that you have time to do this properly, not that you can skip it.
A better number to track than the national one
National adoption stats tell you almost nothing about your business, and they mostly serve to sell things, including anxiety. Thryv found 53% of respondents spend at least $100 a month on AI tools, a figure that’s easy to read as a target. It isn’t one. Replace all of it with three numbers of your own:
- Hours saved last week, counted honestly, net of checking time.
- Dollars spent on AI each month, across every subscription.
- Mistakes caught before they reached a customer.
Track those for a month. If hours saved are low and spending is high, cut a tool. If mistakes are climbing, slow down. If one job is saving real time, go deeper on that job, instead of adding another app because a headline told you everyone else already has.
Next time you see a big adoption number, ask two questions: who was surveyed, and what did “use” mean? What’s the most surprising AI statistic you’ve seen lately, and did it survive those two questions?
