Small-business AI adoption: the real number is under 20 percent

Ask a room of owners whether they use AI and most hands go up. Ask the Census Bureau and the answer is different. Both are right; they are measuring different things, and the difference is the whole story.

What the Census measures

The Census Bureau's Business Trends and Outlook Survey asks whether a business used AI in producing goods or services in the last two weeks. On 3 May 2026 the national figure was 19.8 percent. For firms with four or fewer employees it was under 20 percent and had barely moved since December 2025. For firms with 250 or more employees it was 37 percent.

What owners report

The Federal Reserve's Small Business Credit Survey of 6,525 employer firms, run in late 2025, found 46 percent using AI and another 15 percent planning to within a year. The uses were mostly writing and marketing, at 83 percent, then productivity tasks and planning. The problems named were accuracy, by 46 percent, and adapting tools to the business, by 43 percent. Goldman Sachs surveyed 1,256 owners in its 10,000 Small Businesses programme in early 2026: 76 percent used AI in some form, but only 14 percent had it fully integrated into core operations. The barriers were data privacy and security for 50 percent, lack of technical expertise for 49 percent, and difficulty choosing tools for 48 percent. The SBE Council's survey of 517 employers with two to 99 staff found 82 percent had adopted at least one AI tool, with a median annual spend of $2,200.

Washington: a quarter using, four in ten exploring

The Association of Washington Business surveyed 378 employers in July 2026. 28 percent said their business had begun using AI, up from 22 percent the quarter before. 39 percent were curious and exploring. 28 percent had not considered it. The sample skews to manufacturers and larger employers, so treat it as a direction rather than a small-business figure.

The gap is the work

Between the 76 percent who have tried a tool and the 14 percent who have one doing a job inside their operations sits most of the value, and most of the risk. Embedding AI in a booking system, an inbox or a phone line is not a purchase; it is a design task with a boundary to draw: what the assistant may do on its own, what it hands to a person, and how you would know if it went wrong. That boundary, written down first, is what turns a trial into a tool.

• ✦Start from a workflow with a measurable cost: unanswered enquiries, quotes that need chasing, repetitive data entry between two systems.
• ✦Write the hand-off rule before choosing a model: which cases go to a person, always.
• ✦Log every action the assistant takes, so the first month can be reviewed line by line.
• ✦Train the people who work next to it. Goldman's respondents named training and support as the thing they needed most.