The Allergen Question Your Website Should Never Answer With ‘It’s Safe’

A guest planning dinner messages your restaurant’s website: "My son has a severe peanut allergy. Is the pad thai safe for him?" The assistant, built to be welcoming, answers "Yes, it’s safe, you’ll be fine." No chef saw the question. No one checked whether the kitchen uses peanut oil, shares fryers, or preps that dish near nuts. The family books on that answer. If the kitchen gets it wrong, the consequence is not a refund. It is an emergency room, and a transcript where the restaurant’s own website promised safety.

Most businesses can run a chatty, eager assistant. A restaurant operates where a confident wrong answer can hurt someone. Allergens and accessibility and alcohol service are not soft hospitality topics. They sit on top of food-safety and consumer-protection rules. A general-purpose chatbot answers every one of them with the same cheerful certainty, whether or not it has any basis to.

An Allergen Clearance Is a Safety Representation

The allergen answer is the one that can do real harm. Telling a guest a dish is "safe" for a serious allergy is a representation about cross-contact in a kitchen the assistant has never seen. Food-safety regulation treats allergen handling as a serious matter, and most state food codes, built on the FDA model, expect staff to manage allergen information with care rather than guess. A bot that clears a dish to keep the conversation friendly has made a safety claim no one in the kitchen verified, in a situation where being wrong is dangerous and the company’s name is on the message.

Beyond the physical risk, a flat "it’s safe" that turns out false is also a misrepresentation, and the FTC’s authority over deceptive practices reaches misleading claims to consumers. The honest answer to an allergy question is never a breezy yes. It is a conversation with the kitchen.

Service Animals and Accessibility Are Legal Questions

The next trap looks harmless. A guest asks whether they can bring their service animal, or whether the patio is wheelchair accessible. Restaurants are public accommodations under Title III of the Americans with Disabilities Act, and the rules about service animals are specific. A chatbot that says "sorry, no pets allowed" to someone with a service animal has stated a position that conflicts with the law, in writing, under the restaurant’s name. The same goes for a confident but wrong answer about accessible seating or entrances. These are not menu trivia. They are the kind of statements that turn into complaints.

The Easy Answers the Bot Gets Wrong

Then there is the everyday accuracy problem. A general assistant will confirm a reservation it cannot actually place. It will promise a table at 8 on a night you are fully committed. It quotes a private-event price no manager approved. Ask it on behalf of a twenty-year-old and it may say the bar will serve them, even though alcohol service and age rules are set by state law and carry real penalties for the license. Each answer is a small promise the restaurant has to honor or walk back, made by a tool that knows nothing about your book, your kitchen, or your liquor license.

"Will Not" Is a Suggestion. "Cannot" Is an Architecture.

The reflex is to tell the assistant the rules. Never clear an allergy. Never quote private-event pricing. Treat that as the boundary. It is not, because of how the model handles a question phrased a little differently than expected.

You tell it never to say a dish is allergy-safe. A parent asks, "there’s no nuts in this, right?" The model hears a simple yes-or-no and reassures them. The instruction was loaded the whole time. The wording just did not match what it was told to refuse. A system built to be agreeable will keep being agreeable, and a disclaimer in the corner of the chat does not pull back the "you’ll be fine" that already went to a family making a medical decision.

Who Answers for It

Take the software away and the picture is clear. Picture a new host who cleared allergic guests without asking the kitchen, waved off a service animal as a pet, and booked tables the restaurant did not have. That host would be a problem the first night. When the website assistant does the same things, the responsibility still belongs to the restaurant. Now there is a written record, no chef or manager was in the loop, and food-safety and accessibility rules govern the exact answers it gave.

The restaurants that get burned are not the ones that put an assistant on the site. They are the ones that dropped in a generic chatbot, assumed friendly meant safe, and found out when a guest had a reaction or filed a complaint. The fix is not to take the assistant down. Run one that sends allergy questions to the kitchen, gets accessibility and service-animal questions right, and never invents a reservation, a price, or a clearance it cannot back.

Frequently asked questions