McKinsey's AI survey samples opinion, not a random slice of firms

The setup

McKinsey publishes 'The State of AI' as a recurring global survey of executives on AI adoption, one of the most frequently cited sources in coverage of business AI use, accessed here through an archived copy because the live page blocks automated retrieval. The August 2026 edition's own 'About the research' note, as retrieved, states the online survey was fielded from 4 May to 8 June 2026 and drew 1,719 respondents across 97 nations, spanning a range of regions, industries, company sizes, functions and tenures, with 36% from organizations reporting more than $1 billion in annual revenue.

What the documents show

The same note explains that responses are weighted by each responding nation's share of global GDP, a disclosed adjustment made to offset uneven response rates across countries rather than a raw, unweighted tally. The survey reports adoption in terms like the share of respondents whose organizations use AI regularly in at least one business function, and the share scaling AI across three or more functions, figures the note frames as self-reported organizational behavior, not an independently audited deployment count. Stanford's AI Index lists McKinsey among the named external partners it draws on for exactly this kind of survey data.

The friction

What the methodology note does not claim is that its panel is a random or representative sample of businesses worldwide: respondents opted into an online questionnaire, and McKinsey's own description does not describe a probability-based sampling frame comparable to a government business register. The GDP weighting adjusts for response-rate differences among the countries and firms that did respond; it does not manufacture a representative sample out of a self-selected one.

What changed in the work

What a large, consistently repeated executive survey like this adds is a comparable time series of self-reported sentiment and stated behavior, useful for tracking how executives describe their own AI programs from wave to wave, even though executives describing their programs is a different kind of evidence than logged usage data or government statistics measuring the same phenomenon.

• Is a McKinsey adoption figure being treated as an economy-wide measurement, or reported as what surveyed executives said?
• Does the citing article mention the survey's sample size and country spread, or only the headline percentage?
• Would the same adoption definition used here match a government survey's stricter production-use question?

Repeated year over year, this kind of survey is most useful for tracking direction of change in executive sentiment rather than as a precise level of true global adoption, a distinction the survey's own weighting note implicitly acknowledges by correcting only for response rates, not for who chose to respond in the first place.