Allergen Management & Dietary Substitutions

This is the highest-stakes use of AI in your kitchen, and it deserves a lesson of its own. Done well, AI lets you offer thoughtful gluten-free, dairy-free, vegan, nut-free, and other restricted versions of your menu without the usual three-cooks-in-the-weeds scramble. Done badly, AI sends a guest to the emergency room.

The whole game is structure: never trust AI's allergen output, always verify it, and build a workflow that makes verification fast.

What You'll Learn

• A safety-first prompt framework for allergen substitution
• How to use AI to map hidden allergens across your entire menu
• The "verify-against-labels" workflow you must put in place
• How to handle the most common requests (GF, DF, vegan, nut-free, shellfish-free)

The Safety-First Framework

Every allergen prompt should follow this structure:

1. State the original dish in full, including all bases, stocks, and oils
2. State the specific allergen restriction explicitly
3. Ask for a substitution plan with reasons
4. Explicitly tell AI you will verify against real labels
5. Ask AI to flag what you should double-check in your specific kitchen

Here's the template:

Act as my allergen-aware sous chef.

Original dish (FULL detail, including bases, stocks, and oils used):
- Pan-seared scallops in clarified butter
- Cauliflower puree (cauliflower, whole butter, heavy cream, salt)
- Beurre blanc reduction (shallot, white wine, white wine vinegar, butter)
- Hazelnut crumble (toasted hazelnut, panko, parsley, lemon zest)
- Garnish: micro celery, lemon oil

Guest restriction: DAIRY-FREE and TREE-NUT-FREE (no almonds, hazelnuts,
cashews, pistachios, walnuts, pecans, macadamia, brazil, pine nuts).
Coconut is OK for this guest.

Give me:
1. A revised component-by-component substitution plan that preserves the
   texture and flavor architecture
2. For each swap, one sentence explaining the choice
3. A list of "hidden allergen risk points" in MY kitchen I should verify
   (e.g., shared fryer oil, cross-contamination from prep, ingredient
   labels I should re-check)
4. Three follow-up questions I should ask the guest before serving

I will verify every swap against my actual product labels and our
cross-contamination protocol. Do not approve allergen safety -- only suggest
substitutions for me to verify.

Notice the last paragraph. You're explicitly framing AI as a suggestion engine, not as a safety authority. This matters both for your team's mindset and for liability if something goes wrong.

Mapping Hidden Allergens Across Your Menu

This is one of the most valuable AI uses chefs underuse. Take your full menu, paste it into Claude with ingredient detail, and ask:

Below is our full menu with ingredient lists for every dish.

For each of the Top 9 US allergens (milk, eggs, peanuts, tree nuts, fish,
shellfish, wheat, soy, sesame), tell me:

1. Every dish where this allergen is a component
2. Every dish where this allergen is HIDDEN (in a stock, oil, base, sauce,
   or sub-component most guests wouldn't expect)
3. Every dish at risk from cross-contamination based on shared cooking
   surfaces, fryer oil, or prep stations

Output as a markdown table I can print and put on the line.

The hidden-allergen column is the one that has prevented real incidents. Things like soy in your Worcestershire-based steak sauce, gluten in your roux-thickened velouté, sesame oil in your finishing drizzle, milk solids in clarified butter (technically reduced but not eliminated). AI will flag dozens of these in a single pass.

Print the result. Walk the menu with your team. Verify each flag against your real labels. This is one of the highest ROI two hours you'll spend with AI.

The Verify-Against-Labels Workflow

After you generate a substitution plan, the workflow is:

1. Pull the actual product labels for every ingredient in the new plan.
2. Read the "contains" and "may contain" statements line by line.
3. Check shared equipment risk. Same fryer? Same cutting board? Same wiping cloth?
4. Brief the line cook(s) on the modification before service.
5. Tag the ticket clearly ("DF/TNF" + table number + initial of cook who built the dish).
6. Have the modified dish reviewed by a second set of eyes before it leaves the pass.

This is non-negotiable. AI does not replace any step. AI accelerates step 0 — getting you to a smart substitution plan in 90 seconds instead of 20 minutes.

Handling the Most Common Requests

Gluten-free

Common hidden gluten: roux-thickened sauces, soy sauce (use tamari), wheat starch in spice blends, malt vinegar, beer in marinades, panko or breadcrumbs in meatballs/forcemeats, dusting flour on the line, flour-thickened jus, shared fryer with breaded items.

Dairy-free

Common hidden dairy: ghee (technically dairy-derived, often a no for severe allergies), Parmesan in a "vegan" pesto, milk solids in browned butter, butter in finishing oil, milk wash on house bread, cream in a "vegetable" puree base.

Vegan

Beyond dairy and egg: gelatin in mousses/panna cotta/marshmallows, fish sauce in "vegetarian" curries, anchovy in Caesar/Worcestershire, honey, beeswax glazes, refined sugar with bone-char filtration (case-by-case depending on the guest's strictness).

Tree-nut-free / peanut-free

Hidden risk: pesto (pine nut), Romesco (almond), some commercial breads, dukkah, mole, frangipane, certain ice creams, marzipan glazes, almond extract in pastry, peanut oil in some Asian-style sauces.

Shellfish-free

Hidden risk: oyster sauce, fish sauce (some are made with crustaceans), shrimp paste in Thai curries, surimi in California rolls, dashi made with bonito only (usually safe) vs with shellfish.

For each, the right prompt is the safety-first template above with the specific allergen named. The dietary-specific knowledge above is the kind of context you should add to the prompt so AI knows where to look first.

Key Takeaways

• The safety-first framework: state the dish, state the restriction, ask for a substitution plan, declare you will verify, ask for hidden risk points
• Use AI to map hidden allergens across your full menu — print and post the result
• AI accelerates getting to a smart substitution plan; it does not replace label verification or cross-contamination protocol
• Common hidden traps: roux in stocks (gluten), milk solids in butter (dairy), pine nut in pesto (tree nut), fish sauce in vegan curries (animal)
• Never let an AI-generated substitution be served without a trained human's label-verification and a second set of eyes at the pass