A Jamaican barbecue pop-up recently drew attention after its AI-generated menu images struck some diners as more unsettling than appetizing. Denver nurse Jill Sennett shared the pictures with her 26,000 X followers, and they were reshared more than 500 times. She still ordered lunch, however, showing that an unappealing image does not automatically translate into a lost sale.
Generative AI gives restaurants a faster, cheaper way to produce menu and marketing visuals, part of a broader push toward AI in restaurants. But new research suggests fully synthetic food images can fall short at the exact job restaurant photography is meant to do: make someone want to eat.
In a recent Scientific Reports experiment, 87 participants rated AI-generated food images as less realistic and reported lower willingness to eat the depicted food than with matched real photographs. The results also linked greater perceived realism in AI images with higher willingness to eat.
How diners respond to AI food images
Researchers showed participants 60 pairs of AI-generated and real food images matched for calorie content and flavor. Each image was rated for realism, willingness to eat, healthiness, and calorie content.
Real photographs scored significantly higher for realism and willingness to eat. Healthiness and calorie estimates did not significantly differ after the researchers applied their statistical correction.
The experiment did not measure orders, add-to-cart rates, or restaurant sales, so it cannot show that AI menu photos cost operators business. It does point to a practical risk: a picture can suggest roughly the same nutritional profile while doing a weaker job of making the food look desirable.
Earlier research found that AI-generated food images were often preferred when participants did not know their origin, while identifying an image as genuine increased its appeal. The newer study likewise found that more realistic AI images tended to score better on willingness to eat.
Where AI fits in menu photography
Restaurants do not have to choose between a professional shoot for every menu change and an entirely synthetic plate. Good restaurant food photography can start with a smartphone, consistent lighting, and accurate plating.
DoorDash’s AI Photo Enhance tools offer one middle ground. AI Retouch keeps the original dish while adjusting lighting, sharpness, and backgrounds. AI Replate changes its presentation on different dinnerware, while Match Style applies the lighting, color, and background of a reference image. Enhanced photos still go through the platform’s standard review process.
Start with real photos for the images carrying the most sales weight: signature dishes, premium items, and prominent delivery-menu photos. A strong menu design already uses photography strategically to draw attention to key dishes. Before publishing, compare each image with the current recipe, portion size, ingredients, garnishes, and plating. If an AI edit invents toppings, inflates portions, or changes the food enough to create a different expectation, pull it back.
When AI food images risk misleading diners
AI-generated food imagery is also subject to ordinary truth-in-advertising rules. The Federal Trade Commission considers both words and images when deciding whether an advertisement is likely to mislead a reasonable consumer about something material to a purchase.
Using AI does not automatically make an image deceptive. Problems are more likely when the picture communicates something the actual dish cannot deliver, such as a different portion size, ingredients, preparation, or presentation.
Keep AI edits tied to the real dish
Treat AI imagery as part of the restaurant’s photo workflow, not as a substitute for checking what actually leaves the kitchen. Use real photos for dishes customers are most likely to judge visually, and use AI editing where it improves presentation without changing the product being sold.
Then track the result. Watch conversion where available, item mix, complaints, refunds, and customer comments after meaningful image changes. AI can clean up a restaurant photo, but the finished image should still set an expectation the kitchen can meet.
