What the survey actually measured
On October 28, 2026, McKinsey published an analysis of its Agentic Commerce Consumer Survey. It was fielded in June among about 21,000 adults in Australia, Brazil, China, France, Germany, India, Japan, Mexico, Saudi Arabia, the United Kingdom and the United States. The authors caution that the base varies by question and that comparison with 2025 is available only for the United States. [1 · McKinsey & Company · agentic-commerce survey · October 28, 2026]
The headline reach figure is 53%: across the 11 markets, that average share of respondents had used AI chat, AI search or a retailer assistant for at least one shopping task in the previous three months. McKinsey reports a weighted, representative sample of 9,988 adults for that exhibit. It is a broad definition—product search and comparison count alongside later stages of the journey. [1 · McKinsey & Company · agentic-commerce survey · October 28, 2026]
Far fewer people reach an AI-executed transaction. In a weighted base of 11,088 AI users, 14% said they had completed a purchase directly through an AI tool; McKinsey maps that to about 9% of the full population sample. In the United States, the share of AI users rose from 10% to 17% versus December 2025, but that year-over-year result cannot simply be extended to the other ten markets. [1 · McKinsey & Company · agentic-commerce survey · October 28, 2026]
Standalone AI chat and search tools reached 56%, versus 27% for retailer assistants. At checkout the gap almost disappeared: 14% said a standalone AI completed a purchase, while 15% said a retailer assistant helped complete a purchase or checkout. The researchers explicitly note the different wording, so the one-point difference does not establish that retailer assistants outperform. [1 · McKinsey & Company · agentic-commerce survey · October 28, 2026]
How an online store should measure the effect
The study points to two distinct funnels: external AI services participate more often in discovery and choice, while a merchant’s own infrastructure matters more near payment, where current inventory, price, delivery and payment are required. This is not a reason to build a universal agent immediately; it is a reason to separate stages in measurement and not confuse reach with revenue. [1 · McKinsey & Company · agentic-commerce survey · October 28, 2026]
At discovery, useful measures include the share of products an AI finds correctly, attribute accuracy, price freshness and inclusion in recommendations for target queries. At consideration, track product-page visits, cart additions and correction of wrong attributes. At checkout, track initiated and successful payments, handoffs back to a person, delivery errors and abandonment after a total changes. [1 · McKinsey & Company · agentic-commerce survey · October 28, 2026]
The business case should rest on incremental orders, not total conversations. A control group without the new assistant or a comparable period can help separate an AI effect from seasonality, discounts and rising brand awareness. Costs include product-data preparation, inventory and payment integration, error monitoring, disputed-order support and returns. [1 · McKinsey & Company · agentic-commerce survey · October 28, 2026]
The survey does not provide a ready-made conversion benchmark. It records consumer recollections in June but does not verify receipts, checkout logs or gross merchandise value. Consequently, 53% is not the share of sales flowing through AI, and 14% is not the share of all shoppers: the figures use different bases and meanings. A credible pilot must define an “AI purchase” in advance and make it auditable against a transaction. [1 · McKinsey & Company · agentic-commerce survey · October 28, 2026]
Sources
- McKinsey & Company · agentic-commerce survey · October 28, 2026 — Primary source: methodology for the June survey of about 21,000 adults in 11 markets, question-level sample sizes, findings on AI-assisted shopping and limits of the year-over-year comparison.
Expert commentary
The survey’s most useful result is not the large reach figure but the gap between advice and action. Shoppers already find it convenient to delegate research and narrow a choice, while handing over an order and money requires exact inventory, delivery terms and clear accountability. The market is therefore not moving through one funnel: a choice interface can become mainstream before transaction infrastructure becomes dependable. [1 · McKinsey & Company · agentic-commerce survey · October 28, 2026]
This creates two points where a merchant can lose a customer. First, incomplete or stale data can keep a product out of the recommendation set. Then a selected product can still fail at checkout because payment is incompatible, delivery is unclear or the customer must repeat the journey manually. The winner is not the most fluent assistant, but the data-and-operations stack that carries a promise through to fulfilment. [1 · McKinsey & Company · agentic-commerce survey · October 28, 2026]
The convergence between standalone and retailer assistants at checkout should not be read as one interface winning. The questions differ and the findings rely on respondents’ recollections. A more cautious interpretation is that systems already connected to assortment, inventory and fulfilment become more valuable close to a transaction. That mechanism is plausible, but a merchant experiment must establish the specific advantage. [1 · McKinsey & Company · agentic-commerce survey · October 28, 2026]
Customer protection matters more than speed. If an assistant shortens search but hides alternatives, misunderstands constraints or changes the total without clear approval, convenience turns into distrust and returns. A store needs a visible order summary, a route to correct the choice, an intelligible consent moment and an easy human handoff. Those controls protect the shopper and reduce the cost of investigating disputed orders. [1 · McKinsey & Company · agentic-commerce survey · October 28, 2026]
The broader social effect is still concentrated at the choice stage: recommendations increasingly determine which products a person sees at all. That raises the stakes for ranking transparency and data quality, especially for smaller merchants with fewer resources for technical optimization. One conditional risk is that a handful of AI intermediaries concentrate access to demand; the survey establishes channel use, not concentration or discrimination. [1 · McKinsey & Company · agentic-commerce survey · October 28, 2026]
Over the next two quarters, maturity will be visible not in more AI mentions but in a narrower gap between a found product and a confirmed order without worse returns or support outcomes. Watch recommendation accuracy, completed payments, cancellations, disputed charges and repeat purchases. Until transactional evidence appears, this survey is best treated as a map of hypotheses rather than a revenue forecast. [1 · McKinsey & Company · agentic-commerce survey · October 28, 2026]