The Maze: An AI shopping agent is software allowed to find products, carry out tasks or buy within a customer's instructions. Those permissions are very different businesses. Across 16 countries, most surveyed consumers welcome shopping help, but few want software to complete purchases independently. Merchants face two jobs: make their offers easy for machines to understand, then give customers control over how far those machines can go.
Help has a much larger audience than autonomous buying. Accenture's 2026 research finds 85% open to working with an agent on purchases and 74% open to having it execute specific tasks. Finding a suitable product and handling an instructed return both remove effort without surrendering the whole decision. These are shares of surveyed consumers willing to allow each activity. They are independent answers, so adding them together would misrepresent the evidence. For a retailer, the larger immediate opportunity may be assisting a shopper who still wants to participate. A readable product offer can matter before an automated payment ever happens.
Choosing and paying remain separate permissions. Openness falls to 32% when an agent decides what to buy but the customer pays. Only 9% would allow independent purchasing within set limits. That leaves a 76-percentage-point difference between collaboration and autonomous buying. It does not mean 76% abandoned a checkout: these are different hypothetical permissions, not stages in a tracked sales funnel. A useful design implication is to distinguish finding an option, selecting it and authorizing payment. A customer can accept one without accepting all three. Human approval is part of the service people may want, rather than a temporary obstacle that every shopper inevitably outgrows.
Fast purchase growth does not establish broad merchant readiness. BrightEdge, a company that helps brands appear in search and AI answers, describes agent purchases growing 43% month on month. It also says only 3% of ecommerce companies it works with allow agents to transact. That is a client measure, not the percentage of all online stores. Neither the absolute purchase volume nor the merchant sample size is disclosed. Rapid growth and limited implementation can coexist when a channel starts small. The commercial question is therefore whether an agent can reliably read an offer and follow the shopper's permission through to a completed order, not whether every store should immediately remove human approval.
Treat willingness as a design signal, not a revenue forecast. Accenture surveyed 25,590 consumers online in January 2026. Multicountry coverage makes the permission boundary relevant beyond one national market, but the answers do not establish actual usage, spend or demand in a particular retailer's category. The same consumer can delegate a routine task while keeping a meaningful choice personal. Merchants can test a narrow process, such as finding an in-stock item and requesting confirmation, before granting wider purchasing authority. That is an operational interpretation of the permission gap, not a measured conversion benefit or a claim that the survey tested that particular design.
Why it matters: Stores can lose a place in the customer's shortlist before they lose control of checkout. If an agent cannot understand price, availability or the conditions of purchase, it may struggle to present a useful offer. Making those facts clear serves assisted shoppers as well as automated ones. Test discoverability, permission and payment as separate steps. The evidence supports preparing for delegation while keeping the customer's final say visible.
Sources: Accenture Consumer Pulse Research 2026 · BrightEdge's October 16 measurement summary. Source discovery: Jim Yu's LinkedIn post.
