News
Two honest numbers: what an AI adoption headline counts
November 21, 2026 · PenDraco
The Census Bureau's survey of October 19 to November 1, 2026 put AI use at 24.2% of employer businesses. McKinsey reports regular AI use by nearly nine in ten respondents. Both are honest because they count different things, and a mid-size company needs a third number.
On November 5, 2026, the Census Bureau published its two-week survey of employer businesses. It put the share that use AI at 24.2%. "Nearly nine in ten respondents report regular use of AI in at least one business function", McKinsey wrote in August. Both numbers are honest. They count different things.
An owner or a chief financial officer who sets next year's plan meets both numbers. A company that uses AI in one function looks early against the first. Against the second it looks ordinary. The useful question is what each headline counts.
One number counts businesses
The Bureau says its Business Trends and Outlook Survey gives "nationally representative data on U.S. employer businesses every 2 weeks". Each collection goes to about 200,000 businesses. A business answers once, for "all domestic operations of this consolidated business". In the result, a firm of 4 people counts as much as a firm of 4,000.
The question asks whether the business used AI "in any of its business functions" in the last 2 weeks. In the collection of October 19 to November 1, 2026, the share answering yes was 24.2%. The first collection of this series, in November 2025, gave 17.3%.
A Federal Reserve Board staff note from April 2026 calls the survey "the best source" for the share of businesses. The same note counts firms with 1 to 49 employees at 95% of all firms. So the low number is a fair estimate for a mostly small population.
The other counts respondents
McKinsey's number comes from an online survey of people. It "garnered responses from 1,719 participants in 97 nations", the report says. The survey ran from May 4 to June 8, 2026. Each participant answers about the organization where that person works. McKinsey weights the data "by the contribution of each respondent's nation to global GDP".
One subject shows the same split inside the United States. A question new to that Census collection asks whether the business's employees used generative AI for work tasks. There, 26.8% of businesses said yes. The GenAI Adoption Tracker asks workers. In August 2026, 40.5% "of the employed respondents used genAI for work". The Census question covers 2 weeks, so the two are not a like-for-like pair.
Size explains part of the gap
The staff note gives the arithmetic. Firms with 250 or more employees are 0.9% of firms. They hold 56.2% of employment. So a survey of people reaches large firms far more often than a count of firms does.
A Census Bureau working paper from October 2026 shows the effect inside one data set. "18% of firms used AI in a business function, rising to 32% on an employment-weighted basis", the authors write. The period is November 2025 to January 2026. The answers are the same. Only the weight changes.
Here is our reading of the Census file against the McKinsey report. If size explained the whole gap, the largest businesses in the Census file would answer like McKinsey's respondents. They do not. In the collection of October 19 to November 1, businesses with 250 or more employees answered yes at 44.1%. That is about half of nine in ten. No size class in the file reaches one half.
The working paper reports the same limit. Its firm-weighted rates rise with size, "reaching to 35% for 1000+ employee firms". The rate comes near nine in ten only after 2 more filters. The first keeps 3 sectors where AI use is high, Finance among them. The second keeps answers from top executives and owners. Then "the rates reach 80%", the authors write.
McKinsey's report gives its own counts on size and on title. "Thirty-six percent of respondents say they work for organizations with more than $1 billion in annual revenue", it says. Among respondents who use AI in their own roles, it counts 654 at C-level and 195 individual contributors.
So size is only part of the gap. Sector matters, and so does the title of the person who answers. In the Census file, larger businesses answer "Do not know" more often. At businesses with 250 or more employees it was 24.5% in the same collection. One respondent cannot always see the whole business.
For a company with 20 to 249 employees, the closer benchmark is its own row in the Census file. In the same collection, the yes share was 29.2% for 20 to 49 employees. It was 33.2% for 50 to 99 and 40.0% for 100 to 249.
What the sources do not show
The 24.2% comes from the first collection under a new term. From October 19, the form asks about "Super Intelligence or SI (formerly referred to as Artificial Intelligence or AI)". In that collection the answer "Do not know" rose from 9.6% to 13.9%. One period cannot separate the new term from the trend.
Responses fell from about 24,000 to about 17,500. In that collection, 9% of the businesses asked gave an answer. The Bureau says its estimates "may be subject to nonresponse bias".
The size rows move more. In the collection before, the 3 rows above ran from 26.0% to 35.3%. The Bureau's standard errors for them run from 1.1 to 2.1 points. The row for 250 or more employees rests on about 400 answers.
Until October 5, 2025, the Bureau asked about AI "in producing goods or services". The last share under that wording was 10.0%. The first share under the current wording was 17.3%. The Bureau collected no data in the 6 weeks between, so the files cannot split wording from adoption.
One headline survey covers employer businesses in the United States. The other covers participants in 97 nations, and we found no figure for the United States alone in its report. The questions differ too: regular use, against any use in the last 2 weeks.
None of these numbers measures results. A yes means some use in at least one function in a short period. It does not say how much work the tool does.
Five steps before the plan
Next year's plan needs a benchmark that counts companies like yours, in a unit you can count at home.
- Ask what the headline counts. A count of businesses and a count of people are different units.
- Read the question and its period.
- Find the row for your size class in the Census file, and read it as a range.
- Count your own company the same way. List each business function where AI does production work today.
- Pick one workflow from that list, and set the plan on it.
For a workflow that writes to the books, An AI agent on a live ERP sets out the controls.
If one workflow on your list is ready, start there. The embedded sprint puts one workflow into production in 30 to 60 days. Get in touch.
Sources
- U.S. Census Bureau, Business Trends and Outlook Survey, National data download, period published November 5, 2026, read November 16, 2026: the question, the 24.2%, the 17.3%, the 26.8%, the collection dates and the "Do not know" shares of 9.6% and 13.9%.
- U.S. Census Bureau, Business Trends and Outlook Survey, Employment size class data download, same period: the 44.1%, the 24.5%, the rows for 20 to 249 employees in 2 collections and their standard errors.
- U.S. Census Bureau, Business Trends and Outlook Survey, response counts and response rates, same period: about 24,000 and about 17,500 responses, the 9% and the 400 answers.
- U.S. Census Bureau, AI Core Questions (Original), collections of September 11, 2023 to October 5, 2025: the earlier wording and the 10.0%.
- U.S. Census Bureau, Business Trends and Outlook Survey: About the Data, updated November 5, 2026: the survey's description of itself, the change of term from the October 19, 2026 collection and the words on nonresponse.
- U.S. Census Bureau, Business Trends and Outlook Survey Methodology, last updated October 2026: about 200,000 businesses in each collection and the periods with no data in 2025.
- U.S. Census Bureau, BTOS questionnaire, Content Year 5, Cycle 2, in the field from October 19, 2026: one answer for all domestic operations.
- Bonney, Breaux, Dinlersoz, Foster, Haltiwanger and Pande, The Microstructure of AI Diffusion: Evidence from Firms, Business Functions, and Worker Tasks, Census Bureau Center for Economic Studies working paper CES 26-25R, October 2026: the 18% and 32%, the 35% and the 80%. The paper gives its authors' views, not the Bureau's.
- Jeffrey S. Allen, Monitoring AI Adoption in the U.S. Economy, FEDS Notes, Board of Governors of the Federal Reserve System, April 3, 2026: the shares of firms and of employment by firm size, and the words on the best source. The note gives its author's views, not the Board's.
- McKinsey & Company, The state of AI in 2026: On the road to ROI, August 2026, archived copy of September 15, 2026: the headline sentence, the 1,719 participants in 97 nations, the survey dates, the weighting, the revenue share and the counts by role.
- GenAI Adoption Tracker, home page, wave of August 2026, read November 12, 2026: the 40.5% of employed respondents.
- The comparison of 44.1% with nine in ten and the count of 6 weeks without data are PenDraco's arithmetic from the published figures.