What the Census Bureau's AI-adoption data shows about who is using AI in 2026
On 21 May 2026, the US Census Bureau released a new set of figures on how American businesses are using artificial intelligence. It is one of the clearest pictures yet of who is actually adopting AI, and it is free for anyone to read. For a marketing or product team trying to work out who to talk to about AI and where, it is worth an afternoon.
The numbers come from the Business Trends and Outlook Survey, which the Census Bureau runs across a sample of about 1.2 million businesses and refreshes every two weeks. Late in 2025, it added a set of supplemental questions on AI, fielded between November 2025 and February 2026, which is where most of the detail below comes from. The release and its interactive tables are set out in the Bureau's 21 May announcement.
How many businesses are actually using AI
The headline is more modest than the noise around AI would suggest. Across all businesses, the Census Bureau's summary of the data reports that AI use hovered between 17% and 20% from December 2025 to May 2026, with between 20% and 23% of businesses expecting to be using it within six months.
A companion Census Bureau working paper puts the supplement period at 18% of firms using AI in a business function, rising to 32% once you weight by employment. That second number matters. Weighting by employment counts each business by how many people it employs, so a figure that climbs from 18% to 32% is telling you that the larger an employer is, the more likely it is to be using AI. The average business has not adopted AI. The average employee increasingly works somewhere that has.
Where the differences are
The interesting part for anyone doing segmentation is not the average. It is the spread.
By size, the same Census summary reports 37% of businesses with 250 or more employees using AI, against under 20% for businesses with fewer than 20 employees, a gap that widened over the six months. The working paper goes further at the top end: among very large firms in the most AI-heavy sectors, use runs to between 50% and 60%, and between 60% and 70% on an employment-weighted basis.
By sector, the Census summary shows the Information sector highest at 39.7% current use, Finance and Insurance at 33.9%, and Retail Trade at 14%. So the distance between the most and least AI-adopting industries is close to threefold, before you even look at firm size within them.
What businesses are using it for
The supplement also asks what AI is being used for across fifteen business functions. For a marketing audience, the standout is in the working paper: among the businesses that use AI, the most common function is sales and marketing, at 52%, ahead of strategy and business development at 45% and IT at 41%. The thing the data says AI is most often pointed at is the marketing team's own work.
The paper also notes that adoption is still shallow inside most firms. 57% of users apply AI in three or fewer business functions, and where workers use it for specific tasks, the most common are writing, analysing documents, and searching for information.
Why a marketing team should care
Set the AI angle aside for a moment, because the more useful point is about the data itself. This is a free, regularly updated, government-run dataset that lets you cut adoption by industry, by firm size, and by state. If you sell anything to other businesses, that is a ready-made map of where demand for an AI-related product is concentrated and where it is not.
Three concrete uses:
- Industry targeting. The sector breakdown shows which industries to lead with. An AI-adjacent pitch is received very differently in Information, at about 40% adoption, than in retail, at about 14%.
- Account sizing. The firm-size cuts sharpen an ideal-customer profile. If adoption is concentrated in larger firms, a campaign aimed at the smallest businesses is targeting the group least likely to have adopted.
- Regional rollouts. Because the figures break down by state and major metropolitan area, you can sequence a launch by where adoption is already strongest rather than spreading evenly across the country.
The point about the source
What makes this worth building on is not just that it is detailed, but that anyone can check it. The methodology is published, the sample is enormous, and the tables are open. If a colleague questions a number in your plan, you can point at the survey it came from rather than at a confident guess.
That is the same standard Cambium AI applies to the people in a persona: figures that trace back to public data you can inspect, not to a model's best guess. Statistics agencies, central banks, and large firms make real decisions on data like this. For a marketing campaign, deciding where the budget goes, the same source clears the same bar.
