Local-context mishaps force Just Eat to pull AI-generated food images

Just Eat, a food ordering and delivery platform, apologized after a wave of widespread ridicule on social media, following its use of AI-generated images for the menus of some partner restaurants. The experiment, which aimed to cover dishes that lacked marketing photos, quickly turned into a showcase of surreal dishes that surprised consumers. The digital platform, which relies on accelerating menu digitisation, found itself facing a storm of criticism after its image-generation algorithms failed to grasp the specific names and everyday vocabulary of local dishes.

The results revealed a stark gap between generative models and a understanding of the cultural and local food context; for example, a soda-bread dish with egg and bacon at a café in Belfast was rendered as cooked egg and slices of meat floating inside a glass of carbonated cola after the algorithm conflated “soda bread” with soft drinks. In Manchester, describing a popular crispy-potato sandwich known locally as “chip muffin” generated an image of a muffin dessert stuffed with baked potato sticks within the batter itself. Visual mishaps continued, turning cheese-and-bread rolls into sushi-like pieces made from pure cheese slabs, while a pepper-and-salt potato dish appeared in a pale form lacking onion, garlic and chili. Confusion peaked when a typographical error in a menu printed the word “plain” using the letters of the English word “plane,” prompting the model to draw a garlic-bread loaf shaped like a full-size airplane with wings.

Facing a flood of sarcastic comments on the platform’s Reddit forum and Instagram accounts, Just Eat announced that it would begin removing inaccurate images and halt the generation of new ones until the technology is further developed and fine-tuned. A company spokesperson told the British Broadcasting Corporation (BBC) that the platform works with tens of thousands of small restaurants to support photographing their dishes by funding professional local photographers, but it decided to trial AI-enhanced images for menus that lack photos, noting that emerging technologies do not always get it right in their first attempts. Despite this official retreat, some of the odd designs remained displayed on live menus within the app, bearing a label indicating they were automatically generated, which did not excuse them in the eyes of shoppers.

The crisis goes beyond fleeting ridicule to reveal a clash between excessive automation efficiency and the core of commercial trust.Major platforms aim to reduce the cost of onboarding restaurants and accelerate the inclusion of meals without waiting for field photography sessions, but shifting the visual-representation burden to models that do not understand food backgrounds undermines the foundation of e-commerce; consumers commenting on the experience explained that a customer decides to purchase based on the image matching what will be delivered to their table, and that artificial, machine-generated copies only lead to disappointment at the moment the actual meal arrives, even if the image is labelled as AI-generated.

This retreat carries operational and regulatory implications of great significance for those managing commerce platforms and delivery services in the Gulf, Egypt and the Arab Levant; the heated regional race to onboard thousands of cloud kitchens and popular stores pushes many tech teams toward easy automated generation to cut visual-production costs. Yet the risks in our markets appear sharper because of the wide span of local dialects and the multitude of popular names for traditional dishes such as “maṭbaq”, “musakhan” and “hawawshi”; any visual model that relies on generic translations or ignores the fine details of dishes will fall into the same trap, turning sales catalogs into a marketing-misinformation arena that raises order-cancellation rates, consumer disputes and drains the corporate reputation of the brand.

Savings in photography budgets through generative agents quickly turn into a severe loss in brand value and customer loyalty.What the Just Eat experiment demonstrates is that automation pathways require strict human-review gates and models that comprehend local dialects before being deployed to consumer interfaces; the machine can sketch imaginary dishes in seconds, yet it is unable to grasp the fundamental difference between bread culture and soda drinks.