If you have read two articles about business AI adoption and come away with two completely different pictures, you were not being careless. The US Census Bureau puts the share of American businesses using AI at under a fifth. A large vendor survey of small businesses puts it at more than three quarters. Both numbers are real, both are published by serious organisations, and the gap between them is the single most useful thing on this page — because it tells you which number to use for which question.

Key takeaway

Depending on how you count, between 18% and 77% of US businesses use AI. Government probability samples measuring formal firm-level adoption land near the bottom of that range; vendor surveys of engaged software customers measuring any AI use land near the top. Neither is wrong. They are answering different questions about different populations.

How many businesses actually use AI?

These are the headline adoption figures from the two US federal sources that measure it, plus the largest small-business vendor survey.

17–20%
of US businesses used AI, across the period December 2025 to May 2026.
~18%
of firms had adopted AI as of year-end 2025, on the Census measure.
20–23%
of US businesses expected to be using AI within the following six months.
77%
of US businesses report using AI regularly — up from 48% in July 2024.

That last figure sits roughly four times higher than the federal one. The explanation is in the methodology, not in anybody's honesty, and it is worth understanding before you quote either number.

Adoption rises sharply with company size

This is the most consistent finding across every source, and the most relevant one if you run a small business — because it means the aggregate figures are not describing you.

37%
of firms with 250 or more employees reported using AI.
32%
of firms with 100 to 249 employees reported using AI, as of 3 May 2026.
under 20%
of firms with four or fewer employees reported using AI.
50–60%
use rates among very large firms in Information, Professional Services and Finance — rising to 60–70% when weighted by employment.

The gradient matters practically. A statistic like "most businesses now use AI" is describing firms with hundreds of staff in knowledge-work sectors. For a six-person trades business, the honest reading of this data is that the majority of your direct competitors are not using it yet.

Counting by employees instead of by firms changes everything

Most US businesses are very small, so counting each firm equally produces a low number. Counting by how many people work at AI-adopting firms produces a much higher one.

78%
of the US labor force works at a firm that has adopted AI, on an employment-weighted measure.
54%
of the labor force works at a firm that uses large language models specifically.
41%
of workers reported work-related generative AI use as of November 2025.
32%
firm adoption on an employment-weighted basis for the November 2025–January 2026 reference period, against 18% unweighted.

Both weightings are legitimate and they answer different questions. "What share of businesses use AI?" is a firm-weighted question. "How much of the economy is exposed to AI?" is an employment-weighted one. Publications routinely quote one and headline it as the other.

What businesses report getting from it

These come from Intuit's survey of its own small-business customer base. Read them as what engaged software users report, not as a measured effect across all businesses — the caveat matters and is expanded in the methodology note below.

78%
of surveyed businesses said AI improved their productivity — up from 46% in July 2024.
43%
said AI had increased their revenue.
2%
said AI had reduced their revenue.
as many businesses reported increased hiring as reported workforce reductions.

Why the numbers disagree, and which one you should use

Four differences account for almost all of the gap. The Federal Reserve's own research note exists largely to reconcile them, which should tell you the discrepancy is well recognised rather than a sign that someone is exaggerating.

1. Different populations

The Census surveys a probability sample of US employer businesses — including the very large number of tiny firms that make up most of the American business population. Vendor surveys sample their own customers, who by definition already bought and actively use business software. Those groups are not comparable, and the software-using group will always index higher on technology adoption.

2. Different questions

"Does your firm use AI in producing goods or services?" and "do you use AI regularly?" are not the same question. The first asks about a deliberate business process. The second is satisfied by using a tool that has AI features in it. The Federal Reserve notes that the Census question itself changed in November 2025 — from AI use "in producing goods or services" to use "in any of its business functions" — and broadening that wording moves the number.

3. Firm-weighted versus employment-weighted

Covered above, and worth restating because it alone produces a difference of nearly two to one in the same dataset.

4. Self-selection in who answers

A business that has enthusiastically adopted AI is more likely to complete a survey about AI than one that has not. Government statistical surveys are designed to counter this; commercial ones are generally not.

Which figure to use

For "how many businesses like mine use AI", use the Census figures, filtered to your size band. For "how exposed is the economy", use the Federal Reserve's employment-weighted figures. For "what do businesses that adopted it report", the vendor surveys are legitimate — provided you say out loud that they surveyed adopters.

Cite this page

Using these figures in an article, report or presentation? Please cite the original publishers linked against each statistic. If you are referencing this compilation, either of these works:

Plain citation
HTML link

Methodology and limitations

How this page was built. The figures on this page were read directly from each publisher's own page on August 5, 2026 and are linked individually above. Ranges are reproduced as ranges rather than collapsed to a midpoint. Where a source is a vendor surveying its own customer base — as with the Intuit QuickBooks data — that is stated next to the figure, because it materially affects how the number should be read. We have not calculated, adjusted, extrapolated or combined any of these figures; each is reported as its publisher reported it. Where sources disagree, both are shown rather than one being selected.

Sources