What the Adoption Data Actually Show

The question for the vertical AI firm is not simply how many businesses say they use AI. It is where that use sits: in a formal firm decision, inside a particular business function, in a worker’s task, or in the organizational investments that make a new workflow durable. The evidence from the 2026 Business Trends and Outlook Survey helps answer that narrower question. It gives a more useful picture of the boundary decision than a single adoption rate, while also showing why the results should not be treated as a forecast of total automation.

A firm can have AI running in the background without workers directly operating an AI tool. A worker can use a general-purpose system without the firm having formally adopted it. A business can report use in one function and make no change to the surrounding organization. These are different states of integration. They carry different implications for whether a vertical capability is merely easy to invoke or is becoming part of a redesigned system of responsibility.

Adoption has three layers, not one

The second AI supplement to the U.S. Census Bureau’s Business Trends and Outlook Survey measures AI at three connected but non-equivalent layers. The first is firm-level adoption: whether the business reports using AI in a recent period. The second is functional deployment: whether AI is used across particular business functions over a longer period. The third is worker-task use: whether employees use AI or generative AI to assist with work-related tasks and, if so, which tasks.

The distinction addresses a visibility problem. A firm may install an AI-enabled system that processes information with little individual intervention. Workers may receive its outputs without considering themselves AI users. In another firm, employees may adopt external tools independently while formal systems remain unchanged. If a survey asks only whether the business has adopted AI, it can miss the first and third arrangements or conflate them with the second.

The supplement therefore also asks about task effects, capital replacement, organizational changes, employment, and barriers. These additions matter for the vertical AI question because an invocation is not the same as integration. A business-function deployment can be a sign of formal redesign. A worker’s use may be a sign of local experimentation or immediate assistance. An investment in data practices or new workflows may show that the firm is changing the conditions under which AI operates. None of these alone proves that authority over a consequential decision has moved from a governed vertical platform to an agent.

The reference periods are part of the meaning. The core firm question asks about use in the last two weeks. Functional and worker-task questions generally ask about the last six months, while some questions ask about the next six months. A firm can therefore be counted as a recent non-user at the firm level while reporting a task-level use over the longer period. The apparent contradiction may be a difference in time window rather than a disagreement about what happened.

The instrument itself also changed. Earlier BTOS questions asked about AI in producing goods or services. Beginning in late 2025, the revised core wording asked about AI in any business function. The authors attribute the resulting level shift to the wording change alongside continued adoption and the context of the new supplement. Pre-revision and post-revision headline rates should not be compared as if the question had remained constant.

flowchart LR M["Connected but non-equivalent measurement layers"] -.-> F["Firm-level adoption"] M -.-> B["Business-function deployment"] M -.-> T["Worker-task use"] I["Top-down organizational investment"] --> B W["Bottom-up worker experimentation"] --> T F -. "May align or diverge" .-> T B -. "May align or diverge" .-> T B --> O["Reported task, capital, employment, and performance outcomes"] T --> O I --> O

This architecture changes how adoption rates should be read. A firm-weighted rate gives each firm the relevant weight, so the many small firms in the business population matter greatly. An employment-weighted rate gives more weight to firms with more employees and therefore describes the share of employment located in AI-using firms. It does not describe the share of tasks automated or the share of workers using an AI system every day.

The headline numbers are concentrated, and the instrument matters

For the November 2025–January 2026 reference period, the BTOS reports firm-level AI use at about 18% of firms, or about 32% when employment-weighted. Worker-task AI use appears in about 23% of firms, representing about 41% of employment. The survey’s headline expectation is that firm-level use will reach about 22% within six months. These numbers describe different populations and questions; they should not be combined into one measure of “the AI economy.”

The employment-weighted difference is strategically important.