In the National Restaurant Association's State of the Restaurant Industry 2026 report, 26% of operators use AI-related tools, with marketing the leading use (19% of full-service, 15% of limited-service operators)Source ↗
In Deloitte's 2025 survey of 375 restaurant executives, 82% plan to increase AI investment in the next fiscal year and only 2% plan to decrease itSource ↗
Sodexo reports cutting recipe development from three days to half a day using AI, and AI-assisted menu engineering commonly raises profit 10-15% by flagging which items to feature, reprice, or cutSource ↗
A 2024 audit found more than 70% of AI-powered restaurant bots made definitive dietary-safety claims without access to real-time kitchen data or clear disclaimersSource ↗
creativeChatGPTClaudeGemini

Specials and new-dish ideas from what is already in your walk-in

Weekly specials and surplus product force fast creativity, and few chefs have hours in the test kitchen to chase flavor pairings. This is where AI earns its keep as a sounding board — Velvet Taco's Chat GPTaco came from feeding ChatGPT the chain's own ingredient list, and Sodexo uses AI to shorten seasonal development. The model has never tasted anything, so it proposes directions; you cook, taste, and decide.

Prompt
You are a culinary creative partner brainstorming with a chef. Propose {{count}} special or new-dish concepts for a {{cuisine}} kitchen.

What I have to work with: {{ingredients_on_hand}}. My station and equipment: {{equipment}}. I need these dietary slots covered: {{dietary_targets}}. Target food cost is {{food_cost_target}}.

For each concept give me: a working title, the core flavor idea and why the pairing should work, the main components, a one-line plating note, and the likely allergens so I can plan (I will verify them).

Rules:
- Build mainly from the ingredients I listed; note any small addition a concept needs.
- Every concept must be executable on the equipment I named — no sous-vide or smoker if I did not list one.
- Do not present any pairing as "classic", "traditional", or "proven" unless it genuinely is; otherwise mark it [UNTESTED SUGGESTION].
- Do not give me final recipes with cook times or temperatures — I will develop and test those in the kitchen.

Fill in your details and the prompt updates live — then copy.

analysisChatGPTClaude

Recipe costing and menu engineering you double-check by hand

Ingredient prices keep moving, and knowing your true plate cost, food-cost percentage, and contribution margin is the difference between a profitable menu and a slow bleed. AI can structure the classic menu-engineering grid — stars, plowhorses, puzzles, dogs — and the payoff is real, but chatbots miscalculate arithmetic with total confidence, so the math is something you verify, never trust on sight.

Prompt
You are a menu-engineering analyst. Work only from the data I paste below — do not invent any number.

For each dish I give you: the ingredients with quantities and unit costs, the yield (portions), the menu price, and units sold last month.

Produce a table with, per dish: total plate cost, food-cost percentage, contribution margin (price minus plate cost), and monthly contribution (margin times units sold). Then classify each dish by popularity versus margin (star, plowhorse, puzzle, or dog) and recommend one specific move — reprice, re-portion, re-plate, or remove.

Rules:
- Use ONLY the figures I provide. Do not estimate or fill in prices, yields, or quantities. If something is missing, write [MISSING: what you need] and skip that calculation.
- Show every calculation step so I can check it against my spreadsheet.
- State your food-cost-percentage formula explicitly.
- Recommendations only — I make the final call on pricing.

My data: {{dish_data}}. My target food cost is {{target_food_cost}}.

Fill in your details and the prompt updates live — then copy.

planningChatGPTClaude

Prep lists and par sheets scaled to the covers you expect

Turning a forecast — expected covers, a catering headcount, a private event — into station prep lists and par levels is daily math where errors cost money both ways: over-prep becomes waste, under-prep becomes 86'd dishes and unhappy guests. AI is good at translating covers and per-portion yields into quantities to prep and pars to hit, so long as you keep it inside your own numbers and your food-safety hold times.

Prompt
You are a sous chef building prep lists. Using only the numbers I give you, produce a station-by-station prep list and a pull/order list.

Menu items with per-portion yields: {{menu_yields}}. Expected covers or event headcount: {{forecast}}. Current on-hand pars: {{current_pars}}. Stations: {{stations}}.

For each item output: quantity to prep to hit par, the pull/order quantity given what is on hand, and the station it belongs to. Show the scaling math (covers x per-portion yield) so I can check it.

Rules:
- Base every quantity only on the covers, yields, and pars I provided. Do not assume a par I did not state.
- Do not invent yields or portion sizes. If one is missing, write [MISSING: yield needed].
- Flag items with long lead time or that cannot be safely prepped far ahead as [HOLD-TIME CHECK].
- Where a judgment call is needed (waste factor, safety buffer, weather), write [CHEF TO SET] rather than guessing a number.

Fill in your details and the prompt updates live — then copy.

communicationClaudeChatGPT

Guest-facing allergen and dietary language that routes to a human

Guests ask "is this gluten-free?" or "can the kitchen do this nut-free?" every service, and servers need consistent, cautious language. This is the single most dangerous place to let AI decide anything: it cannot see a shared fryer, a supplier's substitution, or cross-contact on your line. Use it only to format allergen facts you have already verified, and to make sure every uncertain case gets routed to a person.

Prompt
You are helping write front-of-house allergen communication. You do NOT determine whether any dish is safe — I have verified the facts below and you only format them.

Dish: {{dish}}. Allergen facts I have VERIFIED from my recipe and supplier specs: {{verified_allergens}}. Kitchen realities that affect cross-contact: {{cross_contact_notes}}.

Write two things:
1. Server-facing talking points: what a server can accurately say, in plain language.
2. A short guest-facing note stating what has been verified and directing anything beyond it to the chef or manager.

Absolute rules:
- Use ONLY the allergen facts I provided. Do not infer, assume, or fill any gap.
- Never state that a dish is "safe", "free of", or "okay" for an allergy. Describe what is in it and what was verified.
- For any allergen I did not explicitly address, insert: "Please speak with the manager so we can check with the kitchen."
- Always include the shared-equipment / cross-contact caveat I gave you.

Fill in your details and the prompt updates live — then copy.

automationChatGPTClaude

Station SOPs and training guides built from your own recipes

Onboarding line cooks and standardizing stations eats hours you do not have, and the same prep standards get rewritten again and again. AI is genuinely useful for turning your recipes and standards into clean, repeatable SOPs and training sheets — a system you build once and reuse. The judgment stays yours, especially on any food-safety critical limit, which is never a number you let the model invent.

Prompt
You write kitchen SOPs and training documents. Using only the recipe and standards I provide, create a station training sheet.

Station: {{station}}. My recipe and standards: {{recipe_standards}}. Trainee level: {{trainee_level}}.

Produce: mise en place list, step-by-step sequence, plating standard, quality checkpoints (what "right" looks like), and close-down/cleaning steps. Add a line for chef sign-off.

Rules:
- Use only my recipes and standards. Do not invent steps, ingredients, portions, temperatures, times, or quality specs.
- Wherever a food-safety critical limit applies — cook temperature, cooling, hot/cold holding, reheat — do NOT state a number. Write [VERIFY AGAINST HACCP / FDA FOOD CODE] instead.
- Keep language simple and direct for a multilingual kitchen; short sentences, no jargon.
- Note any step that needs hands-on demonstration as [CHEF TO DEMONSTRATE].

Fill in your details and the prompt updates live — then copy.

Common questions from chefs

Can I use AI to tell guests whether a dish is safe for their allergy?

No. AI cannot see your kitchen — it does not know about a shared fryer, a supplier swapping an ingredient, or cross-contact on the line, and audits have found most AI restaurant bots make confident dietary-safety claims with no real kitchen data. Use AI only to format allergen facts you verified yourself against your recipes, supplier specs, FALCPA's Big 9, and local rules, and route every uncertain case to a manager. A wrong claim is both a health emergency and a liability.

Can AI calculate my food costs and menu prices accurately?

It can structure the analysis and lay out the menu-engineering grid, but chatbots frequently make arithmetic errors. Always re-check plate costs, food-cost percentages, and totals in a spreadsheet or on a calculator, and confirm that every ingredient price is current from a recent invoice before you reprice a single dish.

Is it safe to put my proprietary recipes into ChatGPT?

Be careful. Consumer chatbots may retain and train on what you paste, so signature formulas, unique techniques, and supplier pricing are trade secrets you may not want to expose. Use generic descriptions for public-facing tasks, keep crown-jewel specs out, or use an enterprise or no-train tier that contractually excludes your data from training.

Will AI replace chefs?

The evidence points the other way. Adoption is still mostly in marketing and admin, and even the flashy examples — like Velvet Taco's Chat GPTaco — had a chef curating the output. AI drafts menus, costs plates, and writes SOPs, but taste, food-safety judgment, and the creative call stay human. The Association of Professional Chefs frames it as support that keeps the chef the final decision-maker.

Related professions