Demand
U.S. Business AI Use by Employment Size
Census BTOS weighted percentage of businesses reporting current AI use, shown for four exact official employment-size bands: 1–4, 20–49, 100–249, and 250+ employees. The bands are representative anchors, not a reconstructed all-firm-size aggregate.
U.S. Business AI Use by Employment Size
- 1–4 employees
- 20–49 employees
- 100–249 employees
- 250+ employees
In the latest BTOS period ending 2026-08-09, 250+ employees reported 41.5% current AI use. Compare size classes only with their sampling uncertainty and within the same question-wording era.
Methodology
The metric is the Census BTOS weighted percentage answering “Yes” to current AI use in the prior two weeks. Exact published estimates and standard errors are selected from the official period-by-EMPSIZE API; no averaging, interpolation, smoothing, or imputation is applied.
Periods 31–84 ask whether AI was used “in producing goods or services.” From period 88, the broader question asks about AI use “in any of its business functions.” Census observed a level shift and created a new time series; comparisons across the November 2025 boundary do not measure organic adoption growth.
BTOS is a voluntary probability survey of U.S. nonfarm employer businesses. Its roughly 1.2 million-business annual sample is split into six panels, with each panel surveyed once every 12 weeks. Published percentages use nonresponse-adjusted business weights and remain subject to sampling error, nonresponse bias, coverage error, measurement error, and processing error.
Each CSV note preserves Census’s point-specific standard error. A 90% margin of error is 1.645 times that standard error; uncertainty is typically larger for the largest size classes. Suppressed cells are omitted. Periods 85–87 are structural gaps because BTOS did not collect during the 2025 federal funding lapse.
The chart shows four representative, uncombined Census EMPSIZE strata: 1–4, 20–49, 100–249, and 250+ employees. The official 5–9, 10–19, and 50–99 employee bands are intentionally omitted; no broader band is constructed because the required survey weights are not available in the packaged series.
Frequently asked questions
What exactly does this chart measure?
Within each of seven employee-count brackets, it shows the weighted percentage of U.S. nonfarm employer businesses answering Yes to the BTOS current-AI-use question for the prior two weeks. The unit is businesses, not employees, and it does not measure spending, use intensity, productivity, or task share.
Why does adoption by business size matter?
Size gaps show whether AI use is diffusing beyond larger organizations that may have more capital, data, technical staff, and procurement capacity. The pattern is relevant to competition and labor exposure, but the survey does not identify use cases, affected occupations, employee access, or causal barriers. The chart uses only four exact official bands—1–4, 20–49, 100–249, and 250+—and omits 5–9, 10–19, and 50–99 rather than aggregating them without survey weights.
How should size-class differences and the November 2025 break be interpreted?
Compare classes within the same wording era and account for uncertainty in both estimates. Do not read the shift into period 88 as organic growth: Census broadened the question from producing goods or services to any business function, observed a level shift, and designated the later estimates as a new series. The chart uses only four exact official bands—1–4, 20–49, 100–249, and 250+—and omits 5–9, 10–19, and 50–99 rather than aggregating them without survey weights.
Is this “firm size” or employment size, and how are values sourced?
Census calls the visualization firm size, but the API stratum is employment size class: 1–4, 5–9, 10–19, 20–49, 50–99, 100–249, and 250+ employees. Exact Yes estimates and standard errors come from the official period-by-EMPSIZE API, with no averaging, smoothing, interpolation, or imputation. The chart uses only four exact official bands—1–4, 20–49, 100–249, and 250+—and omits 5–9, 10–19, and 50–99 rather than aggregating them without survey weights.
What survey uncertainty and classification limits apply?
BTOS is a voluntary probability survey of employer businesses, subject to sampling, nonresponse, coverage, measurement, and processing error; it excludes nonemployers. Standard errors, often larger for the biggest classes, are in CSV notes but not plotted. Adjacent classes are not automatically statistically different.
Why are October and early November 2025 missing?
The Census Bureau states that periods 85–87 were not collected during the October 1–November 13, 2025 federal funding lapse and were not backfilled.