You already know the shape of it. An order goes out, everything looked fine at the pass, and forty minutes later there is a refund on the account with a photo attached: a burger that looks grey, a pizza with a hair on it, a curry that is "cold and half missing". You never see the plate, you rarely get to argue, and the money comes out of your next payout.

Some of those complaints are real. Kitchens have bad nights and couriers drop bags. But a growing share of them are not, and the reason is simple: on most delivery platforms the photo is the case. Nobody inspects the food. A person or an algorithm looks at the picture, checks the order value against a threshold, and pays.

This guide is about the ones that are not real: how they are produced, why they work, and what an owner can do without turning every customer into a suspect.

How a fake complaint is actually made

There are three flavours, in rough order of how often we see them.

The darkened original. The customer photographs the real food, then drops brightness and warmth in the phone's built-in editor or a free filter app. A perfectly good chicken dish photographed under a yellow kitchen light becomes something that reads as greasy and grey in thirty seconds. Nothing is added — the photo is just made to look worse than the plate.

The added defect. Something foreign is put into the frame or into the image: a hair, a piece of plastic, a dark spot that reads as mould. Sometimes it is physically placed on the food and photographed; increasingly it is painted in with a retouching tool, or "inpainted" by an AI feature that most phones now ship with. Consumer tools have made this trivial. Ten seconds, no skill required.

The borrowed or generated photo. The complaint photo is not of your food at all. It is pulled from an earlier complaint, from a review of another restaurant, or generated outright by an image model prompted with "burnt lasagne in a takeaway box, phone photo". Generated food photos have got good enough that the giveaway is often not the food but the container, the receipt, or the table around it.

The common thread is that the person doing it does not think of it as fraud. Refund abuse is discussed openly on social media as a hack — "how to get free food on [app]" — and the perceived victim is a billion-dollar platform, not the person who cooked the meal.

Why the platforms pay out

Delivery apps compete on customer retention, and the cheapest way to keep a customer is to refund them quickly. The dispute process is designed for speed and scale, which means:

  • Low-value claims are often auto-approved. Below a certain order value or refund amount, no human looks at the ticket at all.
  • The photo is treated as evidence, not as a claim. An attached image raises the chance of approval, and there is no forensic step between "image attached" and "refund issued".
  • The cost is passed to you where the platform can argue the kitchen was at fault: wrong item, missing item, quality. "Order error adjustments", "quality refunds" — the names differ per app, the direction of the money does not. Our guide on how Uber Eats, Deliveroo and Glovo handle complaints goes through each platform's flow.
  • Disputing costs you time, and the platforms know that a restaurant on a Friday night will not fight a €9 refund.

None of this is malicious on the platform's part. It is simply that the incentives point one way, and the photo is the weak link nobody was checking.

What it costs, honestly

We are not going to give you an industry statistic, because the ones that circulate are mostly guesses. Do your own arithmetic instead. Pull last month's refunds and adjustments from each platform's back office and sort them by reason. Count the ones where the only evidence was a photo and the order value was above what you would refund without a second thought. Multiply by twelve.

For most independent restaurants doing meaningful delivery volume, that number is somewhere between "annoying" and "a part-time salary". It is rarely nothing.

What you can do this week

You cannot make the platforms inspect the food. You can make your side of the dispute stronger, faster and less emotional.

1. Photograph what leaves the pass. A single photo of the packed order, taken at the pass before the bag is sealed, changes every later conversation. It is timestamped, it shows the food, and it is yours. Many kitchens already do this for order accuracy; do it for evidence.

2. Save the customer's photo as a file, not a screenshot. When you can, download the original image from the dispute or the chat. A screenshot throws away the camera metadata that often gives an edit away. Our guide to spotting an edited food photo explains what to look for.

3. Dispute the ones that matter, with a technical answer. "The food was fine" loses. "The submitted image shows editing artefacts in the region of the alleged defect and carries no camera metadata; our pass photo at 19:42 shows the order as sent" is a different conversation. The evidence guide has a template.

4. Track repeat claimants. Platforms anonymise customers, but names, addresses and patterns recur. A second "hair in the food" from the same address in a month is worth a note; a third is worth escalating to the platform's account manager.

5. Check the photo before you decide. This is what FraudBite does: you upload the complaint photo and get back a risk score, the suspect regions marked on the image, and a note on the metadata — in seconds, in language you can paste into a dispute. It is a signal, not a verdict; a high score is a reason to look closer and to answer with evidence rather than an apology.

The line you should not cross

It is worth saying plainly: a risk score is not proof that a specific customer committed fraud, and you should not accuse anyone in writing on the strength of one. Use it to decide whether to dispute, and let the evidence — the pass photo, the metadata, the marked-up image — do the talking. The goal is to stop paying for photos that lie, not to start a fight with a customer who had a genuinely bad meal.

If you want to see what the check looks like on a real complaint photo, try it on one of our sample photos — no account needed.