Key takeaway: Merchants often rely on photo evidence to approve damage and wrong-item refunds. That evidence will face growing pressure from generative AI. Merchants should add verification steps, such as video confirmation or return-history checks, before refund losses scale.
Retailers including Boll & Branch and Bogg are catching shoppers who submit AI-generated photos of “damaged” or “wrong” items. The shoppers are trying to win refunds on goods that arrived undamaged, according to reporting by Digiday and PYMNTS.
At Boll & Branch, a customer claimed a set of sheets arrived torn. The rip did not match how cotton actually frays, and one submitted image carried a visible AI watermark. When the company asked for a FaceTime verification, the customer never responded.
At Bogg, a customer first complained that a bag’s straps looked twisted. Bogg’s team explained that the straps could twist normally and sent fix-it instructions. The customer then sent a second photo claiming a rip in the material. It was the same photo as before, but with damage added. The tear did not match how the bag’s material actually rips. Bogg says it sees cases like this a couple of times a month.
Fraud platform Yofi has flagged similar cases. In one, a shopper claimed a liquid makeup product spilled in transit, but the shipping box itself looked dry. In another, a shopper claimed a wrong item arrived. The submitted photo showed the same flaw across multiple boxes, suggesting AI had duplicated the image.
US consumers returned nearly $1 trillion in merchandise in 2024, according to McKinsey estimates cited by PYMNTS. That figure is more than double the total from four years earlier. Retailers now spend an estimated $200 billion a year recovering value from those returns, McKinsey estimates.
Why it matters: Generative AI has made convincing fake evidence cheap to produce at scale. For merchants, that means photo-evidence-only refund workflows are becoming easier to game. More retailers are turning to video verification, purchase-and-return history checks, and their own AI-assisted image forensics.