AI-Generated Damage Photos in Refund Claims: What Online Sellers Can Do

Last reviewed September 29, 2026 · 7 min read · By the PEA team (Lumina Spark)

Short answer: Sellers in China, Europe and the US have reported refund claims backed by AI-generated or AI-edited "damage" photos. The fair response is a consistent process: ask for the original file or a short video, check provenance and metadata, compare the claim with the order, look for damage that couldn't physically happen, and decide on evidence rather than suspicion.

What's been happening

Faking damage used to take editing skill. Now anyone can ask an AI tool to add a crack, a stain or mold to a photo of an item that arrived in perfect condition.

  • China, around Double 11 (November 2025). The South China Morning Post reported that, around the festival, several online shop owners said they had received AI-generated photos of damaged goods: fruit made to look moldy, an electric toothbrush shown with rust, and a dress supposedly fraying at the collar, which the shop caught through mismatched lighting and telltale edges. China Daily later described a fresh-fruit store hit by several suspected AI-assisted "refund-only" claims; the platform said it could not determine whether the images were AI-generated, and the total was below the threshold for a fraud case. In another case, a seafood seller paid out on a claim that six of eight crabs had died, then noticed that the number of male and female crabs changed between the buyer's photos and videos. Police found the video had been faked with AI; the buyer was detained for eight days and the refund was recovered.
  • Europe (March 2026). French and Belgian media reported that some Vinted buyers were using AI-generated images to claim items had arrived damaged. In one case reported in the French press, a seller said a buyer was refunded, without returning the item, after sending a picture of a badly damaged book that the seller said was fake. The OECD's AI Incidents and Hazards Monitor logged the reports as an AI incident affecting Vinted and other platforms.
  • United States (2026). Modern Retail reported brands receiving suspect damage photos, including a bedding company whose "torn" sheets didn't tear the way cotton does, with one image carrying an AI watermark, and a bag maker that received the customer's own earlier photo with damage added.

These cases sit inside a bigger returns problem. In the 2025 Retail Returns Landscape report from the National Retail Federation (NRF) and Happy Returns, US retailers estimated returns at $849.9 billion, or 15.8% of sales, and the report found that 9% of returns were fraudulent. Those figures cover return fraud in general, not AI images specifically; 85% of the retailers surveyed said they use AI to detect or prevent return fraud.

Practical, fair steps

1. Ask for the original file or a short video

A screenshot or a compressed image forwarded through chat carries little information. For claims above a threshold you set, ask for the original photo file as taken on the phone, or a short continuous video showing the item, the shipping label or order number, and the damage, handled in a way you specify. A law professor quoted by China Daily suggested asking for several specific actions in a longer video, saying current AI still struggles to keep video smooth and consistent. One US brand asked for a live video call.

2. Check Content Credentials and metadata

Run the file through a Content Credentials viewer, such as the Content Authenticity Initiative's Verify tool, and through the free official checks from AI providers, such as Google's SynthID check in the Gemini app and OpenAI's Verify page. A positive result is meaningful. A negative one isn't: each official check recognizes only its own company's tools, and credentials and metadata are often lost when files are uploaded, downloaded or screenshotted. Record what you checked and what came back, and read each tool's privacy terms before uploading a customer's photo.

3. Compare the claim with the order

This is often the strongest check. Does the item match the SKU, color, size and packaging you shipped? Does the count match? Is the label yours? The crab seller's case turned on numbers that didn't match. Keep your own records, too: photograph or film high-value orders before dispatch. That won't settle every case (the fruit store filmed every package and its complaint was still rejected), but it gives you facts to put forward.

4. Look for damage that couldn't happen

AI damage often looks plausible at a glance and wrong on inspection. Reported giveaways include a tear that doesn't match how the material actually rips, lighting on the damaged area that doesn't match the rest of the photo, crab legs held stiff and upright in a way that doesn't fit natural death, details that change between photos, and an earlier photo reused with damage added. Compare with how your product really fails; your returns history helps here.

5. Keep a consistent written policy

Publish before purchase what you may ask for (original photos, a video, return of the item for inspection) and apply it to everyone. Consistency is fairer to customers and easier to defend. For card disputes, Stripe suggests including your refund policy and how it was shown to the customer before purchase; the card issuer may or may not weigh it, but it can't hurt. Some brands now require items to come back for inspection before refunding, and track purchase and return history so repeat patterns stand out.

6. Don't accuse without evidence

Most damage claims are genuine; parcels do get crushed. A false accusation can cost you a customer, a review or a dispute, and NIST warns that wrongly labeling authentic content as AI-generated can be extremely damaging. Ask for more information in neutral words ("we ask for this on all claims over a set amount"), base any refusal on specific facts, and don't expose a customer publicly: the crab seller faced a privacy complaint after posting a video about the suspected scam online. Consumer law may also give the customer rights whatever your policy says. In the UK, if goods a trader sells to a consumer turn out to be faulty within six months of delivery, they are generally presumed to have been faulty on delivery unless the trader shows otherwise. In Australia, the ACCC notes that consumer guarantees can't be taken away by anything a business says or does.

7. Follow the platform's and payment processor's rules

Marketplace and card disputes run on their own rules and deadlines. For card disputes where the customer says a product was damaged or not as described, Stripe suggests evidence such as how the product was described before purchase, your messages with the customer, whether the item was returned, and whether you have already refunded or replaced it. Don't assume seller protection covers you: PayPal's Seller Protection, for example, excludes "Significantly Not as Described" claims. Submit your facts through the official process, and where you have strong evidence of fraud, report it through the platform and, for serious losses, to the police.

What PEA can and can't do

PEA (Professional Evidence Authenticator), from Lumina Spark in Japan, runs several independent analyses on an uploaded image and shows its findings, separating confirmed facts from concerns, with a grade from 1 to 10. It checks Content Credentials (C2PA) and, where available, some official watermark checks. When there isn't enough to go on (a frame from a video, a photo of a screen or printout, or no usable material), it says "Not assessed (Grade –)" instead of guessing, and suggests a human review. It can also fix a file with a SHA-256 hash and seal it with an Ed25519 digital signature and an RFC 3161 timestamp from a third-party time-stamping authority, verifiable by certificate number on PEA's public verification page; it does not prove when or by whom a photo was taken. Staff see an image only if you request a human review and consent, and free results are deleted after 7 days.

PEA is preparing a tool for online stores, and the English version is in early access. If that would help your shop, you're welcome to join the early-access list at our early-access page.

This article is general information, not legal advice.

Sources