E-commerce return fraud. Global trends 2026.
The photo of the damaged product has become the key that unlocks a refund. Today it can be made with AI in seconds. In this report we review the global return fraud figures and point out what works to detect it.
Vouch teamOctober 21, 202610-minute read
The numbers
of online merchants suffered refund and policy abuse in the past year: the most widespread type of fraud.1
Survey of 1,278 eCommerce professionals in 37 countries, Nov–Dec 2025.of merchants received returns of used or damaged products, the most common abuse after “never arrived”.1
Survey of 1,278 eCommerce professionals in 37 countries, Nov–Dec 2025.in merchandise will be returned by U.S. consumers in 2025: 15.8% of sales. Online, 19.3%.2
National Retail Federation estimate.of fraud attacks on merchants already use AI, and claims with fabricated damage are the fastest-growing return abuse.3
Forter report on UK merchants, Jun 2026.For scale: across retail, returns fraud and abuse caused about USD 100 billion in preventable losses in 2025, 14.2% of all returns.4
1. What changed: proof of damage is now manufactured
Free returns and instant refunds have become a commercial promise. To keep the experience smooth, many merchants approve the refund with a photo of the damaged product, before the product comes back.
Today that photo can be made with AI in seconds. And abuse is no longer individual: there are groups selling fraudulent refunds as a service.
- Dec 2023United StatesAmazon sues REKK, a network selling fraudulent refunds on Telegram: its main channel had more than 35,000 subscribers and charged a 30% fee.5
- Mar 2024United StatesGroups offering fraudulent refunds for a fee openly advertise on TikTok, Telegram and Reddit.6
- Apr 2025ChinaTaobao and Tmall drop the “refund only” option, which returned money without requiring the buyer to send the product back.7
- Nov 2025ChinaDuring the Double 11 season, sellers report AI-altered photos used to claim damaged goods. In Suzhou, a buyer submitted a fake video of dead crabs and received eight days of administrative detention.87
- Dec 2025Latin AmericaUsers create fake images of “defective” products with AI to get refunds, and the trend reaches the region.9
- Jan 2026ChinaTaobao and Tmall launch their own model to detect fake images in after-sales claims. The largest platforms now treat the claim photo as a risk.10
- Mar 2026United StatesA Boll & Branch customer claims “torn” sheets with a photo carrying an AI watermark. The brand asked for a video call to see the damage and the customer never replied.11
- May 2026MexicoThe owners of a small business find that the photos of “damaged” products a customer sent them to request a refund were generated with AI.12
- Jun 2026United KingdomForter reports that claims with AI-fabricated damage are the fastest-growing form of return abuse and that some groups offer “returns as a service”.3
- Aug 2026European UnionThe AI Act’s transparency obligations take effect. Watermarking of AI-generated content becomes enforceable in December 2026 for tools already on the market.1314
2. How the fraud works
A fake photo shows up in almost every tactic, from the product that “arrived broken” to the box that “arrived empty”.
AI-fabricated damage
The product arrives fine and the photo shows it broken, stained or defective. The Boll & Branch case was caught because the tear didn’t look like the way cotton actually rips.11
Recycled photo
The image is taken from another listing, from other customers’ reviews or from the internet, or reused across several claims and accounts.
“Never arrived” or “arrived empty”
Still the most common form of refund abuse, and one in three merchants already faces manipulated shipment tracking.1
Used and returned
The customer uses the product and returns it as new. More than half of UK merchants report this problem, known as wardrobing.3
Swapped product
A different, counterfeit or cheaper product comes back. 29% of merchants received returns of counterfeit items.1
Refunds as a service
Organized groups request the refund on the customer’s behalf for a percentage, with scripts, accounts and staged evidence.5
3. Who pays
On a marketplace, the cost usually falls on the seller: the platform refunds the buyer and the seller gets back a damaged, incomplete product or nothing at all.15 On a brand’s own store, the merchant absorbs it directly and pays for the return shipping too.
The law protects buyers, and that is right: in Argentina, buyers have 10 days to cancel a distance purchase16 and in Brazil, 7.17 The problem is that abuse hides behind that right. Without a fast way to verify the photo, rejecting a false claim costs more than paying it.
A quick estimate
Fraud loss = returns with photos × fraudulent share × average order value. Replace the values with your store’s.
Estimated lossUSD per month · USD per year
Example values, not market data. Does not include return shipping, reverse logistics or support time.
4. Why current controls fall short
Instant refunds
To offer a good experience, money is returned before the product is received. Forter stopped a USD 800,000 ring that abused exactly those policies.3
The photo as sufficient proof
Many return policies were designed when a photo was hard to fake. Accepting it without analysis today means accepting the claim.
The support team’s eye
The cases that get caught depend on someone who knows the product very well, like the way a fabric tears.11 With hundreds of tickets a day, that doesn’t scale.
Controls that look at the payment, not the photo
Traditional fraud tools analyze the card, the device and the account. None of them analyzes whether the image of the damaged product was edited.
The product comes back late or never
By the time the product reaches the warehouse, the refund has been paid. And with “never arrived” or “arrived empty”, nothing comes back to check.
5. What works: a layered approach
No single signal is enough. What works is analyzing every photo with several layers at once, before the refund is released, and saving people for the doubtful cases.
Image analysis
Look for signs of generation or editing in the damage: edges, textures, reflections and lighting that don’t match a real photo. The file’s metadata helps, but it can’t be the only check.
Image history
Know whether that photo has appeared before: in the product listing, in reviews, online or in another claim, even if cropped or rotated.
Claim context
Cross-check the photo with the customer’s history, return frequency, order value and what is being claimed.
Human review
A specialist reviews only the gray-area cases, with the evidence already organized, and records the reasoning for disputes and chargebacks.
6. Latin America: returns as a right and as a loophole
The region’s e-commerce is growing with ever simpler returns, and photo fraud is arriving at the same time as in the rest of the world.
Argentina: the seller pays
Marketplace sellers describe refunds approved for the buyer and products returned damaged or incomplete, with no one to claim against.15
Brazil: 7 days to change your mind
The Consumer Protection Code lets buyers cancel an online purchase within 7 days of receiving it.17 It is a legitimate right, which is why abuse is hard to tell apart.
Mexico: AI-generated photos
Small businesses are already receiving AI-generated photos of “damaged” products in refund requests.12
The photo arrives by WhatsApp
Many claims are handled by chat, which compresses images and strips much of their metadata. The analysis has to look at the image itself.
7. What’s next
Faking gets cheaper
Each new generation of image tools makes editing faster and more realistic. An analysis of the AI Incident Database concluded that deepfake fraud now operates on an industrial scale.18
Abuse gets organized
Refunds as a service turn fraud into a business: scripts, accounts and photos ready for each store.6
More investment in detection
The fake image detection market is estimated at USD 1.87 billion in 2026 and could reach USD 7.43 billion by 2031.19
More labeling, but not everywhere
The EU AI Act requires AI-generated content to be disclosed, and watermarking becomes enforceable for tools already on the market in December 2026.1413 It doesn’t reach anyone using tools that don’t follow the rule.
Photos signed at capture
Some phones already sign photos with the C2PA standard.20 For now few devices do, and a photo without a signature proves nothing. It complements analysis; it doesn’t replace it.
8. What to ask a vendor
If you are evaluating a tool to review return photos, these questions separate a complete solution from a partial one.
Does it analyze the image, its history and the claim’s context?
Each signal alone leaves gaps: a new edited photo won’t show up in any search, and a recycled photo may carry no trace of editing.
Does it answer before the refund is released?
If the analysis takes hours, the instant refund gets paid anyway. It has to answer in seconds.
Does it explain the reason behind each result?
To reject a return or fight a chargeback, you need to know why it is suspicious.
Can you use it without integrating anything?
If the only option is a technical integration, testing takes months. With a web tool you start in days.
Does it catch photos from your own listing?
One of the most common tricks is claiming with a photo taken from the product listing or a review.
Can I test it with my own cases?
A pilot with your real returns shows the results before you sign.
9. Checklist for operations and support teams
- Measure what share of your refunds is approved with a photo alone.
- Analyze the photo before releasing an instant refund.
- Ask for the photo through the app or browser camera, in the moment, not from the gallery.
- Compare every photo with those in your listing and reviews.
- Review the customer’s return history before approving.
- Photograph and log the product when you ship it, especially high-value items.
- Define what evidence you need to reject a return and keep it on record.
- Separate clear cases from doubtful ones, and send only the doubtful ones to a person.
How Vouch solves it
Vouch analyzes every return photo in under 30 seconds with three layers: detection of images generated or edited with AI, a search for the image online and in previous claims, and the customer’s context. It returns a 0–100 risk score with the reason, and gray-area cases go to a specialist. It works from Vouch Web, with no integration, or via API.
Frequently asked questions
What is e-commerce return fraud?
It’s requesting a refund for an online purchase with a false claim: that the product arrived damaged, never arrived, arrived empty or was a different item. Increasingly, that claim comes with an AI-edited or recycled photo.
How do you detect an AI-made photo of a damaged product?
It’s getting harder by eye. What works is combining technical image analysis, a search for that same photo online, in your listing and in past claims, and the customer’s context, with human review for doubtful cases.
What tools detect AI-edited return photos?
There are general detectors for AI-made images, but for returns you want a tool that combines image analysis, repeated-photo search and the customer’s context, and answers before the refund is released. Vouch, for example, analyzes every photo in under 30 seconds, returns a 0–100 risk score with the reason and routes doubtful cases to a specialist.
What is refund-as-a-service?
Groups that request fraudulent refunds on behalf of others for a percentage of the value. They advertise on social networks and messaging channels and use scripts, accounts and staged evidence.
Sources
This report is based on public sources only; every figure can be verified.
- Merchant Risk Council. 2026 Global eCommerce Payments & Fraud Report (Mar 2026)
- National Retail Federation. Consumers Expected to Return Nearly $850 Billion in Merchandise in 2025 (Oct 2025)
- Retail Rewired. Forter: Retailers report rising returns fraud, with 80% of fake claims fuelled by AI evidence (Jun 2026)
- Appriss Retail. 2026 Total Retail Loss Benchmark Report (Feb 2026)
- Sourcing Journal (WWD). Amazon Lawsuit Names 27 Defendants Running Global Refund Fraud Scheme (Dec 2023)
- CNBC. Refund fraud schemes promoted on TikTok, Telegram are costing Amazon and other retailers billions of dollars (Mar 2024)
- China Daily. AI-driven refund scams spur calls for stronger e-platform protection (Dec 2025)
- South China Morning Post. China consumers use AI to alter product photos, claim refunds for goods that appear damaged (2025)
- Infobae. Alerta por nuevo fraude con IA: usuarios crean imágenes falsas de productos “defectuosos” para obtener devoluciones (Dec 2025)
- OECD.AI (AI Incidents Monitor). Taobao and Tmall Deploy AI to Combat Fraudulent Refunds with Fake Images (Jan 2026)
- Modern Retail. From Boll & Branch to Bogg, brands are battling a surge of AI-driven return fraud (Mar 2026)
- Infobae México. Emprendedora se viraliza al detectar imágenes falsas creadas con IA y evita presunta estafa (May 2026)
- William Fry. EU AI Act Omnibus Deal Reached: Postponed Deadlines, Watermarking Compromise, and the Nudification Prohibition (May 2026)
- Jones Walker LLP. The EU approved a high-risk AI delay, but most transparency obligations remain (2026)
- iProUP. Fraude en primera persona: así te estafan en Mercado Libre cuando vendés un producto y te lo devuelven
- Infoleg (Argentina). Ley 24.240 de Defensa del Consumidor, texto actualizado
- Planalto (Brasil). Lei nº 8.078/1990, Código de Defesa do Consumidor
- The Guardian. Deepfake fraud taking place on an industrial scale, study finds (Feb 2026)
- Mordor Intelligence. Fake Image Detection Market Size and Share Research Report, 2031 (2026)
- Google Security Blog. Pixel and Android: trusted images with C2PA Content Credentials (Sep 2025)