Key takeaways
- Check the population, question wording and survey period before comparing adoption rates.
- Use relevant size and industry groups, with subgroup counts and uncertainty where available.
- Price the actual proposed task and distinguish spending already incurred from planned investment.
In May 2026, the Census Bureau reported that 19.8 percent of U.S. businesses were using artificial intelligence. McKinsey’s 2025 global survey put reported adoption at 88 percent. A business owner consulting both reports needs to examine who answered, what they were asked and how the answers were weighted.
The Census findings came from its Business Trends and Outlook Survey. For the period ending May 3, the survey asked about AI use during the preceding two weeks. McKinsey’s survey asked respondents about regular AI use in at least one business function at their organizations.
McKinsey received answers from 1,993 participants in 105 countries between June and July 2025. Thirty-eight percent said they worked at organizations with annual revenue above $1 billion. The firm weighted responses by each country’s contribution to global economic output. Its 88 percent figure describes that survey, while Census estimates business use across the United States.
Read the question alongside the percentage
Even a continuous survey can change its definition. In November 2025, Census broadened its question from AI used to produce goods or services to AI used in any business function. Its May 2026 article explains the change. A comparison across that boundary must account for both the wider question and changes in actual use.
Save the questionnaire wording with any figure included in a budget proposal. Note whether it covers all AI or only generative tools, such as chatbots; whether it asks about use today or plans for later; and whether trying a tool once qualifies as adoption.
Then check the unit being counted. A percentage of employees, a percentage of companies and a percentage of companies’ tasks answer different questions. A large company using AI in one small department counts as an adopter in a company-level survey, even if most of its staff never use the tool.
Choose comparisons that resemble the business
Census reported use among 37 percent of firms with at least 250 employees and 32 percent of firms with 100 to 249 employees in the May 3 period. Its national rate combined businesses of different sizes and industries. Use the relevant breakdown when it is available, and check uncertainty estimates before treating a small difference as meaningful.
For a survey focused on midsize businesses, read the publisher’s definition of midsize. An employee range and a revenue range can include different firms. Industry, country, operating costs and whether respondents already buy AI services also affect how closely a sample resembles your company.
Ask how many responses support a particular comparison. A report can have a large total sample but only a few respondents in your sector. If the publisher gives no subgroup count, leave that limitation beside the figure rather than presenting it as a precise peer benchmark.
Build a budget from proposed work
For a project proposal, ask what the company would purchase and which expense or constraint it would change. Include software, integration, employee preparation, ongoing review and support. A survey of AI spending can provide context, but suppliers still need to price those items for the actual scope.
Use a common accounting basis when comparing budgets. One company may count engineering salaries and cloud costs; another may report only licenses. Record whether each figure describes money already spent, an approved budget or an intention to spend.
Also separate a company’s performance from claims about what caused it. Businesses reporting high AI use may differ from others in staffing, management or prior technology investment. To attribute a profit increase to a particular tool, look for a study design that addresses those differences or for a controlled trial inside the company.
A useful request to a research publisher is specific: provide the questionnaire, recruitment method, field dates, weighting, relevant subgroup counts and definition of spending. If those materials are unavailable, show which comparison remains uncertain and use a limited trial to collect the operating data needed for the purchase decision.
