The refund photo had an AI watermark on it
Customers have started generating pictures of damage that never happened and sending them in for refunds. Big retailers have fraud teams for this. A small online shop has a returns page and a gut feeling — so the returns page had better do more of the work.
The head of a bedding company noticed something odd about a run of refund claims. Customers were sending photos of torn sheets. The tears looked wrong — not the way cotton actually frays.
Then he noticed the watermark. Some of the photos still had the mark of the AI tool that made them.
The sheets had arrived perfectly fine. The damage was generated.
This went from rare to routine very quickly
A fraud-prevention platform told a retail trade publication this year that the practice had “exploded overnight” — from isolated incidents to daily occurrences across thousands of merchant accounts. Another fraud firm now ranks AI-generated damage claims as the fastest-growing form of return abuse. UK retail trade press reported the same surge in June.
The mechanics take minutes. Buy something. Photograph it. Ask a free tool to add a crack, a stain, a tear. Submit the picture through the returns form. Get the refund. Keep the item.
It sits inside a bigger problem that was already expensive: fraud researchers link nearly one in four refund dollars to some form of abuse, and one deepfake-detection firm estimates that three in ten retail fraud attempts now involve AI in some way.
Why this lands hardest on small shops
A large retailer has a fraud team, detection software that reads image metadata, and enough volume to spot patterns.
You have a returns email address and a feeling about whether a customer seems genuine.
For twenty years that was fine, because a photo was decent evidence. Faking one convincingly took skill and time, and almost nobody was going to bother over a £40 lamp. So small shops quietly built their entire returns process on a single assumption: if they can show me a picture of it broken, it is broken.
That assumption stopped being true this year, and most small shops have not noticed yet.
The part nobody wants to say out loud
The tempting response is suspicion — treat every damage claim as a probable fake, make everyone jump through hoops.
Don’t. Most claims are genuine. Parcels really do get crushed. A shop that greets every honest complaint with an interrogation will lose more in goodwill than it ever saves in fraud, and the honest customer will say so in public.
The aim is not to catch people. It is to stop relying on the one piece of evidence that just became free to forge — and to say so calmly, in advance, on a page, so nobody feels accused.
It cuts both ways
There is an uncomfortable mirror image here, and it is worth facing.
Plenty of shops have started using generated images for their own product photos. Cleaner backgrounds, perfect lighting, a lifestyle scene that never happened. Then the real item arrives looking different — a slightly different colour, a different texture — and the customer sends it back as “not as pictured.”
That is not fraud. That is a fair complaint, and the shop caused it.
It also quietly destroys your standing in every other dispute. You cannot ask a customer to prove their photo is real when yours isn’t.
Real photographs of the actual item are now doing two jobs at once: they set honest expectations, and they give you something solid to check a complaint against.
The fair question
devkoi builds the parts of this a customer actually reads: the returns page, the product pages, and the photographs on them.
It does not run your checkout, your returns portal or any fraud detection, and it cannot tell you whether a picture someone emailed you is real. That judgement stays with you.
What it can do is make the rules clear before anyone needs them, which is genuinely most of the battle. A policy written calmly on a page in advance is a policy. The same words typed in an email after a suspicious claim is an accusation.
What to do this week
1. Write your evidence requirements on your returns page, now. Before there is a dispute, not during one. Say what you will need for a damage claim, and it applies to everyone equally.
“Add to the returns page: for damage claims, please send a photo with your order number handwritten on paper next to the item, or a short video.”
The handwritten order number is the cheap trick here. It is trivial for a genuine customer and much more awkward to generate convincingly.
2. Above a value you choose, refund on return, not on a photo. Set a threshold that suits your margins. Under it, a photo is fine. Over it, the item comes back first and the refund follows on arrival.
“For items over £75, say we refund once the item reaches us.”
3. Photograph high-value orders as you pack them. Thirty seconds with your phone. If a claim ever arrives for that order, you have your own dated picture of exactly what left the building.
4. Look at claim photos properly before you pay. Zoom in. Check for leftover watermarks, damage that does not match the material, packaging that is not yours, and lighting that does not match the rest of the picture. None of this is conclusive. All of it is worth a second look.
5. Use real photos of the real item on your own site. Your own phone, good daylight, the actual thing — not a generated version of it. It cuts genuine “not as pictured” returns and keeps your own evidence honest.
“Replace the product images with these photos I took of the actual stock.”
6. Fast-track the customers you know. Repeat buyers with a clean history get the easy path. Extra checks are for first-time, high-value claims. That is roughly how the large retailers are responding too, and it keeps the friction away from the people you most want to keep.
7. Say it kindly. Explain that you check claims carefully because it keeps prices fair for everyone. People respond well to reasons, badly to suspicion.
(General information, not legal advice. Whatever your policy says, it cannot remove a customer’s legal rights over genuinely faulty goods — check the rules where you trade.)
For a long time, a photograph was the cheapest proof anyone had. That is the whole of what changed: it is still cheap, and it is no longer proof.
The shops that adjust will not be the ones with fraud software. They will be the ones that wrote down, in plain words and in advance, what they need — and showed their own customers real pictures in the first place.