Customers Sent AI-Generated Photos of "Damaged" Sheets That Never Existed. Some Still Had the Watermark.
E-commerce retailers spent the last decade streamlining damage claims into something almost frictionless: send a photo of the torn or wrong item, get refunded, keep the item, no box to ship back. Late in 2025, Scott Tannen, CEO of the New Jersey bedding company Boll & Branch, went public with what his team had actually been receiving — AI-generated photos of "torn" sheets that had arrived perfectly intact. Some of the fake images still had the AI generator's watermark on them.
A Photo Was Cheaper to Trust Than a Return Was to Process
Returnless refunds exist because verification costs more than the fraud it's meant to catch, for the overwhelming majority of claims — it's cheaper to take a customer's word and their photo than to pay return shipping and restock a damaged item nobody can resell anyway. The whole system assumed the photo was real, because for most of its history, it reliably was.
The Watermark Was Still On the Fake
Tannen described what his team found on LinkedIn, later picked up by PYMNTS: customers submitting AI-generated images of torn sheets to claim refunds on product that had shipped fine and stayed that way. The tell, in some cases, wasn't subtle — the generator's own watermark was still visible in the image being submitted as proof of damage.
Photo Verification Was Built to Catch a Different Kind of Fake
The system was designed to flag obvious mismatches — a photo of the wrong item, a stock photo lifted off the internet — not a synthetic image generated to order and tailored to whatever complaint would get a refund approved. Generative AI didn't break the verification process so much as walk straight past what it was actually checking for.
Every Return Is Now a Risk Score, Not a Policy
By the 2025 holiday season, a UPS subsidiary was deploying AI-based inspection technology specifically to screen incoming returns, while vendors like Riskified and Narvar began selling risk-scoring products that sort customers into different return experiences based on history rather than treating every claim the same. The National Retail Federation reported in October 2025 that nearly two-thirds of retailers planned to tighten their returns process within six months — the industry's own description of the shift is a move from one uniform policy for everyone to a "risk-based" one, meaning even a first-time customer with a genuinely damaged order now faces more scrutiny than a photo alone used to require.
Filed from public reporting:
