Reviewed October 31, 2026. Each estimate retains its own observation period.
There is no single AI adoption rate. Asking whether a company uses AI, observing a payment, and measuring a financial return answer different questions. A useful comparison starts with who was measured, what counted as adoption, and when.
Three estimates, three different denominators
These figures are separate observations, not a league table or a time series. On a narrow screen, scroll the table horizontally.
Selected AI adoption evidence reviewed on October 31, 2026
Source and estimate
Who, when, and what counts
How to interpret it
McKinsey, 2026 surveyNearly nine in ten report regular AI use in at least one business function.
1,719 respondents in 97 countries. Fieldwork: May 4 to June 8, 2026. Respondents report their organizations' use. Results are weighted by countries' contributions to global GDP.
A respondent-based measure of organizational use. Thirty-six percent work at organizations with revenue above $1 billion. It is not a count of all firms or proof of enterprise-wide deployment.
US businesses in Ramp's transaction sample. The index describes coverage of more than 85,000 businesses. Adoption means an observed positive payment for an AI product or service during the month.
Observed buying, with a commercial customer-sample boundary. Free tools and purchases outside Ramp can be missed. A payment does not establish active usage, production deployment, or a return.
A nationally representative business survey in the US. This estimate asks about AI use in a business function during the previous two weeks. It is a dated Census observation, not an October estimate.
A broader business denominator and a short use window. The question was broadened in November 2025, creating a break from earlier estimates focused on producing goods or services.
Why the estimates differ
Start with the denominator. A survey of managers around the world, a sample of US spending-platform customers, and a representative US business survey describe different populations. Their percentages cannot be averaged into a global adoption rate. A difference between them is not evidence that one source is wrong.
Then check what the measure can see. A company could use a free AI tool without an observed payment. It could pay for a subscription that staff rarely use. A team could run AI in production while most of the company has made no change. These situations belong at different stages of adoption.
Finally, align the period and the question. A monthly payment measure and a two-week use measure are not interchangeable. Changes in the question, supplier coverage, or customer sample can change a series even when behavior has not changed by the same amount.
The 2026 report's business analysis examines spending intensity, inference economics, and how companies change their workflows. This reference complements that annual snapshot by comparing measurement methods across sources.
Does adoption mean AI is paying off?
No. McKinsey's 2026 survey reports that 37% of respondents attribute some positive impact on earnings before interest and taxes to AI. That is a reported financial outcome, not an adoption rate or a causal estimate of what earnings would have been without AI. Read the survey and methodology.
To judge a deployment, connect use to a defined task and outcome. Track completion quality, elapsed time, human review, error rates, and the full operating cost. Compare those results with the previous workflow under similar conditions. Paying for a tool or generating more tokens is evidence of activity, not sufficient evidence of value.
Which measure should you use?
Business reach: use a survey with a population and weighting appropriate to the businesses you want to describe.
Paid demand: use transaction evidence, retaining the provider coverage and customer-sample limits.
Production deployment: ask whether a workflow runs reliably with real users, at what scale, and with how much supervision.
Financial returns: compare measured outcomes and all relevant costs against a credible baseline.
State the population, period, and definition alongside every percentage. If those details are missing, the number is not yet ready to support a market-wide claim.
This is State of AI's comparison of attributed public findings, not a new survey. Figures retain their source definitions and dates. Updates will follow source verification, with changes to definitions recorded rather than spliced into an apparently continuous trend. Review is monthly.
Cite this page
Benaich, Nathan. “AI adoption in 2026: use, spending, and returns.” State of AI Report. October 31, 2026.
