· AI

The Smallest Businesses Were Never as Far Behind on AI as the Government's Own Survey Said

The businesses everyone assumed were falling behind on AI mostly just weren't being asked the right question.

The Census Bureau's Business Trends and Outlook Survey, released October 22, found that 23.3 percent of businesses with one to four employees now report using AI in some business function. That is barely behind firms with 20 to 49 employees, at 23.8 percent, and ahead of firms with five to nine employees (21.1 percent) and ten to nineteen (22.1 percent), according to Simon Willison's analysis of the Bureau's own release data. The smallest businesses in America, the ones every "you're falling behind" sales pitch is aimed at, are sitting in the middle of the pack.

That is not what the same survey said six months earlier. The Census Bureau's own write-up from May 2026, covering data through early May, put AI use among four-or-fewer-employee firms under 20 percent, and said that number "did not change significantly" for firms under 20 employees between December 2025 and May 2026. Something shifted in the five months after that.

Part of the shift is a real question change, not a trick. In November 2025, the Bureau broadened its core AI question from use "in producing goods or services," a framing built for factory-floor automation, to use "in any business function," which captures the stuff a four-person shop is actually likely to use: drafting an email, summarizing a call, scheduling. Both the May and October figures I just cited were measured under that same broadened wording, so this is not a comparison contaminated by a definition swap. It is five months of the same question showing a real move, and the most obvious explanation is the ordinary one: cheap, ambient AI, the kind bundled into Gmail, Microsoft 365, and a dozen other subscriptions tiny businesses already pay for, kept getting better and more default-on throughout the year.

Here is the steelman for ignoring this entirely. A yes-or-no question about "any business function" is a low bar. Checking a box because you used autocomplete in Outlook once this month is not the same as a company running AI in its finance or operations stack, and BTOS samples the smallest firms the most thinly, so a few points of movement there carries more noise. That point holds up on its own evidence: Gartner's September 2026 survey of 1,303 companies with at least $50 million in revenue found only 22 percent have scaled AI across multiple business units or gone AI-first, and that high performers who actually track return on investment saw positive returns on 81 percent of initiatives, versus low performers who could not even say the return on 29 percent of their own, according to coverage of that Gartner survey. Usage and value are different numbers, and the gap between them is where most of the enterprise AI story lives.

Grant the skeptics that point: showing up in a usage survey is not the same as capturing value. It does not erase the diffusion story, because diffusion and value capture are separate claims. Nobody ran an outreach program to get a four-person landscaping company onto an AI assistant. Microsoft and Google priced it into tools that business already bought, and competition between them pushed the free tier of "good enough" further than either wanted to go alone. That is the mechanism that moved the smallest-firm number, not a grant, not a mandate, not a "readiness" campaign. It is the same mechanism that cut inference prices by double digits over the past year: rivals racing each other drove the floor down faster than any policy office could have planned it.

There is a tidy footnote to how little government attention went toward actually tracking any of this well. The same month this data shows small firms catching up, Executive Order 14434, signed September 29 and published in the Federal Register October 2, directed federal agencies to replace "Artificial Intelligence" and "AI" with "Super Intelligence" and "SI" across non-statutory communications. The Census Bureau's own BTOS FAQ page now defines the term Super Intelligence as what the survey used to call artificial intelligence. The order does not touch pricing, chips, or any input that would make AI cheaper or more useful. It is the second wording change to this one survey series in a year, and neither change came from anyone trying to lower a cost or ship a product. One was a real methodology fix that surfaced the diffusion story above. The other was a rename.

If you run a small business and keep hearing that you are behind, check what you already have before you buy anything new. The tool you pay for every month probably already does more than it did a year ago, for the same price. The question worth asking is not whether you have AI. It is whether any one task you do every week is worth paying to automate further than what is already bundled in, with a number attached to the hours it costs you now. If you want a second opinion on that specific number before spending anything, that is the whole job at Mojo's AI automation practice, and I'm glad to take a look.

Sources

References used in this article. Links also appear alongside the relevant claims.

Let's talk

Let's make it happen.

You don't need a polished brief. A couple of lines about where you want your company to go is plenty, and we'll come back with what we'd tackle first.

We scope the work around your goals and timeline before quoting anything, so you know exactly what you're getting.

LocationBoca Raton, Florida
CoverageSouth Florida + remote nationwide
Status Now accepting clients