Consumers expect AI agents to handle 15% of their purchases within five years, according to research Global Payments released on September 23, 2026. The comparable expectation was 9% in its earlier survey. That is a rise in what respondents anticipate, with actual purchase behavior still to be established.
The Global Payments release also says 45% of Americans have used or would consider using an AI agent for purchases. Combining those two groups makes the figure a measure of experience or openness. It cannot be read as a 45% adoption rate.
The research used two survey waves: 8,000 respondents in August and September 2025 and 8,027 in May 2026. Both covered seven markets: Australia, Brazil, China, France, Singapore, the U.K. and the U.S. The release also reports that 33% of Americans want to approve every payment.
For a retailer, a repeat purchase provides a concrete way to examine that preference. Imagine a customer asking an assistant to reorder the same laundry detergent. “The same” could refer to brand, fragrance, pack size and quantity. A substitute that preserves only the brand may change the product the buyer intended to receive.
A useful approval screen would show the exact item, the number of packs and the final charge, including delivery. It should also identify whether the order creates any recurring commitment. If the requested size is unavailable, present the replacement and revised total before asking for confirmation.
That example gives a pilot team a manageable set of decisions. Which changes can the assistant propose? Which require the buyer to review the order again? What happens when the buyer declines a substitution or cannot complete payment? Write those expectations in plain language before the first customer uses the service.
Keep the purchase record understandable after checkout. The confirmation should identify the merchant and ordered variant, then give the buyer a direct path to order support. A smooth recommendation is only useful if the customer can recognize and manage the resulting order.
The survey can help a team choose questions for an AI shopping pilot. Budget decisions should then follow the pilot’s completed orders, customer feedback and support issues, with expectations from the survey kept in context.
Common questions
Are AI agents already making 15% of purchases?
That is not what this research establishes. The figure describes consumers’ expectations for purchases within five years.
Does the 45% U.S. figure measure active users?
It combines people who have used an agent with people who would consider doing so. The combined figure should not be presented as current use.
What should a retailer test before allowing repeat purchases?
Check exact-product matching, substitutions, final totals and recurring-order disclosures. Make it clear when the buyer must approve a changed order.