One in Three Shoppers Regrets a Purchase an AI Recommended. The Return Still Comes to You.
· CX Pulse
One in three shoppers regrets a purchase an AI recommended. The expectation was set somewhere you can't see, and the return lands where you can.
One in Three Shoppers Regrets a Purchase an AI Recommended. The Return Still Comes to You.
SmartCustomer, a business reviews site, surveyed nearly 1,200 U.S. consumers and released the results on Monday, reported by Retail Dive on Tuesday. 76% have used AI to make a purchase in the past year. 89% say they don't fully trust the recommendations. And one in three has bought something on an AI's recommendation and regretted it. Half of those purchases were electronics. 45% were apparel and footwear.
SmartCustomer hasn't published the survey itself, so treat those figures as reported, from a company with an interest in the answer. Gartner's own release from May, which anyone can read, says the same thing from a different sample. Among consumers who used generative AI for a recent purchase, 54% said they had to double-check everything it told them, and 62% said the information turned out to be a waste of their time. Only 11% would let an AI make the purchase decision, and that was in the lowest-stakes categories. Kate Muhl, a VP analyst at Gartner, put it in five words: "Accuracy is now a brand issue."
The recommendation happens somewhere you aren't
When a shopper asks an assistant which earbuds to buy and it names yours, the expectation for that product gets set in a conversation you never see, from sources you didn't choose. Your product page might be one of them. So might a two-year-old review, a competitor's comparison chart, and a forum thread with the wrong model number. 53% of the people SmartCustomer surveyed suspect companies of seeding false or misleading information specifically to steer those recommendations. Whether or not that's happening to your products, the assistant can't tell the difference either.
Then the order ships, the regret arrives ten days later, and it lands in your numbers the same way any other bad purchase does. A return. A "not as described" contact. A one-star review saying the product didn't do what they were told it would, by something you have never spoken to.
If you run returns and support at scale
Your return reason codes were written for a world where the customer formed their expectation on your product page or in your store. Wrong size. Not as described. Changed mind. Arrived damaged. None of them can say "an assistant told me this would work with my phone and it doesn't." So the cost of a bad AI recommendation gets filed under "not as described," and merchandising takes the blame for a description it never wrote.
Two cuts fix that. First, add one question to the returns flow or the post-purchase survey: where did you decide this was the right one? Put an AI assistant on the list of answers. You will get a number nobody in the building has. Second, take your twenty highest-return SKUs and ask the assistants your customers actually use what each product is for and what it works with. Compare the answers to your own product data. Every disagreement is a return you have already paid for and will pay for again next week.
Electronics were half the regretted purchases, and electronics is where compatibility lives. That's a fixable category. Compatibility is a fact, and a fact can be published where the assistants read it.
If you're the whole company
The same assistant is describing your product to someone right now, and its brief is whatever it can find. For most small operations that's one product page, a handful of reviews, and a marketplace listing. If the page doesn't say what the product works with, what it doesn't do, and who it isn't for, the assistant fills that in from somewhere else. You find out which somewhere else when the return comes in with a note.
The move is unglamorous. Take the question behind your last three returns and answer it on the product page, in plain words, near the top. That page is now the source document for a recommendation you will never see. Write it like one.
The recommendation happened somewhere you weren't. The return happens somewhere you are.