TL;DR

Just Eat has stopped rolling out a trial of AI-generated dish photos and is removing the pictures already shown, The Grocer reported on 20 October. The tool used independent takeaways’ own menu wording as prompts, and in several cases read the names too literally. Just Eat apologised and said it will keep developing the tool before using it again.

What went wrong

The idea was sound enough: give small independent restaurants a quick, cheap way to show pictures of their food, since images are known to lift orders. The execution was not. The AI took menu names at face value or misread them altogether.

A bread bap sold as a “Breakfast Soda” became sausages and a fried egg bobbing in cola, which one customer called “the actual worst thing I’ve ever seen”. A “Garlic Bread Plane” came out as an aeroplane made of bread. A listing for burger sauce produced a tiny burger sitting in a sauce pot, a “Cheese Roll” was drawn as sushi with cheese inside, and a sauvignon blanc whose tasting notes mentioned watermelon was pictured with chunks of the fruit floating in the wine.

The results, which The Grocer described as “disgusting”, have been shared widely on social media and online forums.

Just Eat’s response

A Just Eat spokesperson said the company wants restaurants to show photos that match what customers will receive, and that its imagery work already spans tens of thousands of smaller restaurants, including paying for local professional photographers. “We’ve been trialling AI-enhanced images and as with any new technology, it isn’t always quite right the first time,” the spokesperson said. “We’re sorry for any inaccurate images that have been posted.”

Looking forward

In our view, this is a clear example of where generative AI fails without context. A menu line is written for people who already know the local shorthand; an image model sees only the words. For UK retailers and platforms tempted to auto-generate product imagery at scale, the lesson is cheap to apply: put a human check, ideally the business owner’s, between the generated image and the live listing, especially where the picture stands in for something a customer is about to pay for.