Do delivery apps reject AI-enhanced food photos? A photographer's straight answer

I get this question almost every week, usually asked a little nervously: 'Will Wolt ban us for using AI?' I understand the worry. There's a lot of noise online about fake AI food, and nobody wants their menu pulled a week before the weekend. But the fear is pointed at the wrong target.

Platforms don't scan for 'was AI involved here'. They can't do that reliably, and they don't try. What a reviewer looks at is the photo in front of them: is it one real dish, shown clearly, at the portion you actually sell, with nothing slapped on top of it. That's the whole test. A photo either passes it or it doesn't, and how the photo was lit has nothing to do with the verdict.

Do Wolt, Uber Eats and the other apps actually ban AI photos?

Here's what the rulebooks actually say, because they don't all say it the same way. Wolt's published checklist does exclude 'AI-generated images' — but read the next line and it means pictures of food that never existed. Wolt's own merchant photo tool ships an 'Improve with AI' button. So Wolt isn't against AI touching a photo; it's against AI inventing the dish.

Uber Eats never mentions AI at all. Its rule is that the photo must accurately represent the real item you sell, and it refuses stock images. DoorDash publishes a rejection category it literally calls 'mismatch' — the photo has to match the item it's attached to. Glovo, Bolt Food and Deliveroo all land in the same place in their own words: show the food as it's genuinely served, honest portion, no overlays.

Read across all six and the pattern is impossible to miss. None of them polices the tool. Every one of them polices the truth of the picture. If you want the exact specs and content rules per platform, I've written them up — start with /guides/wolt-photo-requirements and /guides/uber-eats-photo-requirements, the two most restaurants ask about first.

What actually gets a menu photo rejected

In practice, almost every rejection I've seen comes down to a short, boring list — and it's the same list whether the photo came out of a camera, a phone, or a relighting tool:

• ◆The portion or ingredients don't match — an invented garnish, a plumped-up portion, a side that doesn't come with the dish. This is the big one, and it's a content failure, not a tool failure.
• ◆Text, logos, watermarks, or a phone's timestamp burned onto the image. Every platform declines overlays; your branding belongs in the listing, not on the tile.
• ◆More than one dish in the frame, or a whole-table collage. One item, one photo — that's universal.
• ◆Blurry, dark, or too small. A screenshotted or WhatsApp-recompressed copy often falls under the resolution floor before you even crop it.
• ◆A stock photo, or another restaurant's dish. Uber Eats and DoorDash both explicitly refuse borrowed and search-engine images.
• ◆Hands, faces, or a cluttered background. Glovo bans body parts outright; DoorDash and Deliveroo don't want faces in the shot.

When AI enhancement crosses the line

This is the part I won't dodge, because it's the real risk. If you use a tool that generates food instead of enhancing your photo, you absolutely can get rejected — and honestly, you'd deserve to. The moment an AI adds a herb that isn't on the plate, fattens a portion, or paints in a sauce the customer won't get, the photo now lies.

That's a 'mismatch' on DoorDash, a portion problem on Bolt Food, an accuracy failure on Uber Eats. And a rejection is the mild version. The worse outcome is the photo passing review, then a customer opening a box that looks nothing like what they tapped 'order' on. Rejections you can resubmit in a day. Refund disputes and one-star 'not as pictured' reviews stick to your listing for months.

So the line isn't 'AI or no AI'. The line is 'is this still the dish you cooked'.

How to enhance a photo and still pass review

The safe lane is narrow and simple: change the light, never the food. That's the whole design principle behind PlateRefresh. It takes the phone photo you actually shot and relights it to studio quality — same ingredients, same garnishes, same plating, same camera angle. It improves the lighting, the texture and the surface, and it adds nothing new to the plate. Nothing gets invented, nothing gets removed, portions don't move.

Which means a properly enhanced photo has to clear the exact same content rules a good phone photo would — because underneath the better light, it is your phone photo. The categories a relight fixes are the ones platforms complain about most: dim kitchen lighting, dull colour, a distracting background. The category it refuses to touch is the food itself, which is precisely the category that gets photos rejected.

My honest advice, whichever tool you use: shoot the real plate, keep the original file as your proof that the enhanced version matches reality, and never ask software to imagine a dish. Do that and 'is it AI' stops being a question anyone can catch you on. If you want to see what an honest relight costs per photo versus booking a shoot, /pricing lays it out plainly.