Using LLMs to Optimize Restaurant Menus

A restaurant's menu is its most important piece of marketing, yet most menus are written once and barely touched. We experimented with using LLMs to analyze and rewrite menu descriptions — the results were surprising.

The setup: we fed Claude the restaurant's existing menu, its brand voice guidelines, the target audience profile, and a dataset of top-performing dish descriptions scraped from high-rated restaurants in the same cuisine category.

The model suggested three types of changes:

• Sensory language. Replacing "grilled salmon with vegetables" with "wood-fired Atlantic salmon, charred broccolini, lemon-herb butter." Descriptions that engage multiple senses (sight, smell, texture) correlate with higher order rates.
• Price anchoring. Reordering dishes so the most expensive item appears second in each section, not first. The LLM identified this pattern from the training data and recommended the restructure.
• Allergen clarity. Rewriting vague labels like "contains nuts" into specific, confident statements: "prepared with pine nuts; nut-free preparation available on request."

We A/B tested the rewritten menu at a 40-seat bistro over two weeks. Average check size increased by 8%, and the kitchen reported fewer allergy-related questions from servers.

The takeaway: LLMs are not replacing chefs or menu designers, but they are excellent at pattern-matching what works across thousands of menus and applying those patterns to yours.