The Maze: A shopper asks an AI assistant to find a fragrance-free detergent. The assistant can pull together a shortlist while the shopper still opens a retailer's site to decide and pay. In a December 2025 US survey reported by Boston Consulting Group (BCG), people who already shopped with AI were four times as likely to report using it to find products as to use it for a purchase. For brands, the assistant can influence which products get considered before the retailer sees a visit.
The details
🔎 Research leads. Among US respondents who already shopped with AI, 96% used it to find products and 80% used it to weigh options. The rates describe separate tasks. One person can report both, and the survey does not track a group of shoppers moving from discovery to comparison to purchase.
🛒 Buying involves a different decision. In the same AI-shopper group, 24% used AI for purchasing. That is a share of people reporting a use, not a share of orders. BCG separately says 4% would let an agent complete a purchase alone. That is stated willingness, and the write-up leaves unclear whether that question used the same group.
Our take
Brands and retailers can work on the product recommendation before committing to a checkout integration. The immediate question is whether the assistant can match a shopper's requirements to a real offer.
🧾 Make product fit easy to verify. BCG reports that about 70% of shoppers who use AI compare product attributes with it. A detergent brand can test a prompt for a fragrance-free formula that works in cold water. Check whether the answer identifies the right variant and can point to facts on the product page. A brand can lose the comparison when those facts are only on the packaging, even if the item meets the brief.
🏷️ Follow the price to the listing. BCG says about half use AI to compare prices before buying and describes uneven access to current product information. When two packs look alike, a low price in the answer can belong to a different pack size or an expired promotion. Test whether the recommendation opens the correct in-stock variant at the quoted price. That tells the retailer whether to improve offer data before paying for a wider integration.
🔐 Give checkout its own test. Purchase assistance and willingness to delegate the order are separate measures. A retailer can start by sending shoppers to a product page where they confirm the item, the seller and the total. Track whether that handoff works separately from whether anyone wants the assistant to authorize payment. That keeps a broken link or wrong offer from being mistaken for reluctance to let AI pay.
Sources: Boston Consulting Group
