The deck says everyone. Census says 18%. The denominator is the story.

A current vendor deck can say AI is nearly universal. The Census Bureau can say 18% of U.S. firms use it in a business function. Neither number is automatically false. The first question is not “which one wins?” It is “who was counted, what were they asked, and what does a yes mean?”

McKinsey’s 2026 global survey says nearly nine in ten respondents report regular AI use in at least one function at their organization. Its online sample includes 1,719 people across 97 nations, and 36% work at organizations with more than $1 billion in annual revenue. Census surveys U.S. businesses and weights the result to represent firms, including the small ones that dominate the count. Different denominator. Different respondent. Different answer.

Four honest ways to read it

• The board-deck read: among people willing and able to answer a global management survey, encountering AI somewhere in the organization is now ordinary.
• The economy read: across the full U.S. business population, most firms still report no AI use in a business function.
• The worker read: one employee using a model can truthfully make an organization a yes without making AI broad, governed, or valuable.
• The scale read: adoption and enterprise-wide deployment are different. McKinsey reports 44% scaling across the enterprise, while Census finds 57% of adopting firms use AI in three or fewer functions.

What is real

Large organizations adopt earlier and appear more often in executive and technology surveys. Census shows the same size effect inside its own sample: 37% of firms with at least 250 employees reported AI use, compared with less than 20% of firms with four or fewer employees in the December-to-May data. McKinsey also finds larger organizations ahead on scaling. The disagreement shrinks when you compare like with like.

The skeptical read

Both measures are self-reports. McKinsey’s respondent may know one visible pilot and miss shadow use elsewhere. A Census respondent may miss AI embedded in purchased software or define “use” more conservatively. Neither headline measures active seats, workflow depth, accepted outputs, revenue, cost, or risk. Adoption is the start of the question, not the ROI answer.

What other evidence is saying

• McKinsey, The state of AI in 2026 — The report supplies the nearly-nine-in-ten result, enterprise-scaling rate, sample composition, and survey weighting.
• Census, 2026 AI supplement working paper — The nationally representative firm survey reports 18% firm-weighted use and 32% employment-weighted use.
• Census, May 2026 business update — The later waves show use between 17% and 20% and break results out by firm size and sector.

What to watch next

For market adoption, watch repeated representative surveys with unchanged wording. For your company, ignore the market headline and count active users, recurring workflows, accepted outputs, time after review, failure cost, and dollars retained. The denominator you can operate is your own.

Skip this brief if

Skip it if a big adoption number was your reason to buy. A peer statistic cannot name your workflow or approve your risk. Run one measured job. Keep the tool only if the result survives review and changes the economics.