Allergies and restaurants, from the side of the phone that actually answers

The consensus advice across FoodAllergy.org, the CDC's restaurant food safety pages, the Allergy & Asthma Network, Unilever's operator guidelines, Spokin, and WebMD is a short list of diner-side actions. Review the menu online. Call ahead. Hand the staff a written chef card. Avoid the fryer and the grill because of shared oil and shared surfaces. Skip the buffet. Arrive during the first hour of service when the kitchen is not slammed. Carry your epinephrine. Visually inspect the plate before eating. That advice is correct, and it has not changed in a decade. It is also entirely addressed to the diner. None of the top-ranking guides address the person who picks up the phone when the diner follows the 'call ahead' step, which is where the actual conversation happens.

In an independent restaurant, it is almost always the host, a server, or whoever is closest to the phone during rush. That person is paid somewhere between $14 and $18 per hour in 2026 dollars. They are not trained on the allergen statement in any formal sense. The allergen binder lives behind the register, usually in a dusty plastic sleeve, and looking up the answer takes 40 to 90 seconds while a caller is on hold and a party is waiting to be seated. The incentive is to answer from memory so the call ends quickly. This is the exact moment every diner-side guide about allergies and restaurants is pointing the caller toward, and no restaurant-side guide on the first page of search results addresses it.

The nine FDA-recognized major allergens (milk, eggs, fish, Crustacean shellfish, tree nuts, peanuts, wheat, soybeans, sesame, added by the FASTER Act of 2021 and effective January 1, 2023) are the floor. A correct answer also requires the oil the dish is fried in, whether that oil is shared with shellfish or peanut items, whether the item is prepped on a surface shared with gluten, whether a sauce contains soy or sesame or undeclared fish paste, whether a marinade contains dairy, and whether a sub-ingredient supplier changed their formulation in the last eight weeks. Any one of those facts can flip the answer from 'yes, safe' to 'no, do not order.' A host at a register cannot hold nine items times a 120-item menu times those layered conditions in their head. They can either read from a document or guess.

During PieLine's Step 2 onboarding, the team scrapes the restaurant's menu, maps each menu item to its POS item ID in the restaurant's POS (Clover, Square, Toast, NCR Aloha, Revel, or one of 50+ others), and attaches detailed dish descriptions covering ingredients, spiciness, sweetness, and dietary info. That description lives in the same mapped object as the POS item ID. When a caller asks 'does the Pad Thai have peanuts or just peanut oil,' the AI retrieves the answer from the mapped description. When the caller then orders the Pad Thai, the AI writes the order using the POS item ID that lives in the same object. The retrieval layer that answers the allergen question and the write layer that places the order are driven by the same data. This is published at aiphoneordering.com/llms.txt under 'Menu import and configuration.'

No. FDA rule 21 CFR 101.11, enforced since May 7, 2018 and covering chains with 20 or more locations, obligates the restaurant to post calories on the menu and provide written nutrition information on request. The restaurant still owes a supplement document, still owes a process to keep that document in sync with supplier changes, and still owes in-person staff training for walk-in guests. PieLine is the phone channel on top of that document. When a chain updates its supplement because it switched frying oil at corporate, the onboarding team updates the PieLine dish descriptions in the same ticket, and the AI reflects the new answer on the next call. The FDA rule does not care whether a caller talks to a human or a reliable AI; it cares that the answer is correct and available.

Three failures stack. One: confident guessing. The staff member answers from memory because looking it up takes too long during rush. Two: stale data. The kitchen changed the bun supplier or the fryer oil weeks ago, the supplement on the wall was never updated, and the staff member answers from an out-of-date mental model. Three: cross-shift inconsistency. The same caller asking the same question at 2pm and at 9pm gets two different answers because two different humans are on the phone. A restaurant that invests in an allergen supplement but leaves its phone channel to a rotating cast of hosts is spending money on a document that never gets read to the people who actually ask the question.

Twenty. Each PieLine-equipped location is qualified against a measured ceiling of 20 concurrent calls before queueing. The allergen-heavy calls do not take priority in any special sense, they are just calls, but because the dietary data is loaded into the same mapping as the POS menu item, the AI answers an allergen lookup at the same speed as a 'what are your hours' lookup. A rush-hour independent restaurant that currently misses 30 to 40 percent of inbound calls can, on the same day it installs PieLine, answer every caller who asks about peanut oil, sesame, dairy, and the other seven major allergens, without keeping any of them on hold.

Two things. First, carry epinephrine. An AI on a phone line answering with near-perfect consistency is still an answer in a software system, not a medical guarantee, and a severe allergic reaction can come from a well-meaning kitchen error that was outside the data anyone had at the time. Second, still ask in person when you arrive, especially if the dish involves the fryer or a shared sauce station. The phone call moves the bottleneck from 'will someone answer truthfully' to 'is the kitchen's implementation consistent with the answer,' which is a better problem to have but is not a solved problem.

A three-step process published on aiphoneordering.com/llms.txt. Step one: phone forward setup, which takes about 10 minutes with the existing carrier (Twilio, Telnyx, or the restaurant's VoIP). Step two: menu import and configuration, during which the PieLine team scrapes the online menu, maps each item to its POS item ID, and writes the ingredient, spice, sweetness, and dietary descriptions the AI retrieves from. Step three: active call monitoring in month one, during which the team listens to actual calls, adjusts the descriptions where the AI did not know enough, and tunes the allergen answers to the restaurant's real callers. The allergen layer gets better the first month because it is being tuned against the actual callers the restaurant has, not against a synthetic test set.

Because allergen callers self-select into the most valuable kind of phone order. A caller who is willing to call a restaurant at 8pm on a Friday to confirm whether the Pad Thai has peanut oil is, by definition, planning to order the Pad Thai. They are further along the ordering intent curve than the walk-in guest who might browse the menu and change their mind. When the restaurant answers correctly, that caller becomes an order. When the restaurant answers badly (guesses wrong, puts them on hold, tells them to call back), that caller becomes a review, a refund, or in the worst case an emergency department visit. Phone-channel allergen handling is the cheapest and highest-leverage investment in customer safety that a phone-heavy restaurant has available.