Why AI-Generated Restaurant Menus All Look the Same
AI-generated restaurant menus converge on glossy, uncanny food as repeated edits and recycled training data deepen sameness, disgust and public distrust.
Summary
AI-generated restaurant menus increasingly feature unnaturally symmetrical, smooth food, from perfectly round ice cream to malformed shrimp and excessively melted cheese. Reality Defender CTO Alex Lisle says large language and diffusion models behind ChatGPT and Midjourney reproduce patterns in their training data. Fast-food prompts often draw on similar Wendy’s, Burger King and McDonald’s menus, creating what Lisle calls a 2015 Chili’s aesthetic. Maggie highlighted off-putting New York City AI food advertising on X on August 28, 2026.
Generated images returning to training datasets reinforce that sameness. Lisle distinguishes this quality-degrading convergence from model collapse, the more severe failure caused by repeatedly training models on their own output. Demand for fresh material is intense, with Amazon sourcing rare books, scanning them and destroying them afterward. Repeated revisions also compound distortion: on August 19, 2026, Labtec edited a ChatGPT menu 100 times and watched the food become rounder, smoother and less recognizable, an experiment independently replicated with similar results. Lee Rainie, director of Elon University’s Imagining the Digital Future Center, says optimization for pleasing, inoffensive output removes distinctive edges. University of Duisburg-Essen researchers found nearly realistic AI food provokes more disgust and unease than obviously fake imagery. Restaurants therefore risk customer backlash, while Reality Defender and other verification startups address a wider erosion of trust in images, video and courtroom evidence.
Positives
- University of Duisburg-Essen research gives restaurants evidence that nearly realistic AI food can repel customers more than obvious fakes.
- Reality Defender and other verification startups are building tools to identify synthetic media and authenticate content.
- Labtec’s 100-edit ChatGPT experiment exposed how routine menu revisions can progressively distort generated food.
- Obviously artificial food images produced less disgust than nearly realistic versions, offering designers a clearer warning about uncanny imagery.
Risks & concerns
- One hundred ChatGPT edits made menu food progressively rounder, smoother and less recognizable.
- Wendy’s, Burger King, McDonald’s and 2015-era Chili’s styling can steer generated menus toward the same visual formula.
- AI-generated material returning to training datasets can intensify convergence and, at greater levels, risk model collapse.
- Nearly realistic AI food triggered disgust and unease in University of Duisburg-Essen research.
- Synthetic images and video weaken assumptions underlying customer trust and courtroom evidence.
- Amazon’s scanning and subsequent destruction of rare books illustrates the intense demand for fresh training data.