Restaurant FAQ Checklist for AI Phone Answering: What to Prepare

Before an AI phone host answers a guest's first question, the restaurant has to write down the answers. Here are the categories worth preparing, what each one needs from you, and where a person has to take over.

A restaurant phone line carries two different kinds of call, and only one of them is comfortable for a business to leave unattended. A guest who wants to order dinner has usually decided what they want by the time they dial. A guest with a question has not: they are looking for the one piece of information that stops them from turning around. Preparing for the question side is the work. It is a list of facts about your restaurant, written down in the words your guests use, and it is the difference between an AI phone host that answers confidently and one that hands every caller to a busy host stand. This checklist is that list: the categories of question worth preparing, what each one needs from you before go-live, and the two categories that fall outside what this checklist can cover.

Start with the categories, not the script

It helps to know what the shape of the question traffic looks like, because vendors in this space organise it differently. Palona, a voice-AI vendor, published a guide to the FAQs an AI phone agent should be ready to answer, and grouped them into seven categories: hours and location information, menu options and modifications, ordering and pickup rules, coupons and promotions, reservations and large parties, catering questions, and how to reach staff. That is a useful map of what guests ask. It is not a description of what any particular restaurant is asked, and every restaurant's mix differs. Two of those categories — coupons and promotions, and catering questions — fall outside the fact base this checklist is drawn from, so they are not covered here and no claim is made about how they are handled.

1. Hours, address, parking and directions

This is the category to get right first, because the answers are the same every day and they already exist somewhere: on your website, on the door, on Google, in the window. The documented scope of DineCall's FAQ handling covers business hours, address, parking, directions, allergen warnings (for example MSG, nuts and gluten), and license details — the operational facts of your restaurant rather than anything interpretive. Write them as sentences a caller would accept, not as a list of fields. A line like "we close at 10 PM on weekdays and midnight at the weekend" is a fact; the same schedule written as "weekdays 22:00 / weekend 24:00" is data, and data has to be translated before it can be spoken. Dimsum Caddy in Richmond, BC is a good illustration of the category: a venue whose kitchen runs until 12:30 AM and which closes on Tuesdays, where the calls that make up most of a new venue's volume are about hours and the Tuesday closure, how the indoor golf simulator bays work, and whether a group can come — and where the phone host answers from the venue's own published facts rather than a script.

Two details decide whether the answer survives contact with a real caller. First, the exception. Every restaurant has some version of it: the Tuesday closure, the private event, the holiday menu, the room that is out of service. If the exception is not written into the FAQ, the host will hit it mid-call and either guess or escalate. Second, the parking answer usually needs more words than the hours answer, because it covers directions and the difference between the restaurant's lot and the street. Write the full sentence once, in the words you would use on the phone.

2. Menu options and modifiers

The menu category is different from the others because it is the part the AI already knows. DineCall's AI host is grounded in the restaurant's own menu knowledge graph: the entire menu, including complex modifiers such as "extra spicy" and "sauce on the side", sizes, substitutions, combos, and current pricing. So for menu questions your preparation is not transcription — the menu is already the source. Your preparation is the modifiers and the exceptions: which substitutions you allow, which dishes cannot be modified, which combos exist and what comes in them, and which items are seasonal.

Getting this right also depends on where the menu lives, and that is a publishing decision rather than an AI decision. Changes to items, prices or availability in the bilingual menu CMS reflect instantly across the AI voice agent, the restaurant website and the QR codes, so a price change made once in the editor reaches a caller on the phone and a guest scanning a QR code at the table at the same time. A second copy of the menu, maintained separately for the phone, is where stale prices come from. Keep one source and the question category solves itself.

3. Allergens and dietary questions

Allergen questions are the category where being wrong costs the most, and they sit oddly with the rest of the list. DineCall's documented FAQ scope includes allergen warnings such as MSG, nuts and gluten, so the host can state the warnings your kitchen has documented. What it should not do is settle a medical question on the caller's behalf. The pattern is that a request the AI should not decide on its own is recognised as such and either handed to a person live or captured and flagged for the team, whichever way the restaurant sets it up. Yuan's Garden in Richmond, BC uses exactly that shape: a large party, a private room, or an allergy goes to a person, while the ordinary booking is written straight to the reservation board.

So prepare two layers. The first is the factual layer: the allergens that appear in your kitchen, the cross-contact risks, the ingredients a regular customer asks about every week. The second is the decision layer: which of your questions are answered by a machine and which are answered by your host stand. Write the second layer down as an explicit list, because the default under pressure is to answer, and the default you want on an allergy is not to.

4. Reservations and large parties

Reservation calls are where an unprepared restaurant gets into trouble fastest, because the question and the booking are the same call. The same DineCall AI host that takes food orders can also complete a table reservation during the call. It checks live availability before offering a table — service hours, available time slots, party-size limits, per-slot capacity and dining duration — and the restaurant can either confirm eligible bookings instantly or hold new requests for manual approval. Bookings taken on the phone appear immediately in the owner dashboard, where staff can create bookings, seat guests, reassign tables, complete a visit or cancel a reservation.

What you must prepare is the policy set, and it is longer than most owners expect: the booking horizon, minimum notice, maximum party size, active-booking limits per phone number, table-lock duration, last seating, slot size, and dining durations by party size. Those are configuration values, and they belong to you rather than to a vendor default, because they encode what your room can actually seat. Around them sit two more questions worth answering in advance. What happens to a no-show? DineCall supports reservation reminders, late-arrival grace periods and no-show protection, and it can automatically tag and warn staff about callers with a history of repeat no-shows or fraudulent ordering, with blacklist restrictions permitted. And what does the host do with a table that a caller negotiated down to a time you would not have offered? Decide the floor before a caller asks you to move it.

5. Ordering and pickup rules

Ordering questions are the ones you already know best, and they are worth preparing anyway because the phone version of an ordering question is not the same as the website version. The questions with rules attached are the ones to write down first: how long a pickup order takes, whether an order can be changed after it is placed, and what to say when an item is unavailable that morning. The ordering surface itself is documented: the kitchen order board displays incoming orders from phone, web and tables instantly, so a phone order and a web order land in the same place with the same item modifiers and totals. Hao's Restaurant Group in Richmond, BC runs three restaurants where DineCall answers all three phone lines, taking pickup orders and booking tables on the same call.

6. The same question in the caller's language

For a bilingual restaurant this is not a category so much as a multiplier on every category above, and it is the one most often left unprepared. DineCall's AI host speaks English, Cantonese and Mandarin, detects the language the caller is using on the fly within one or two words, and switches vocabulary and accent without requiring key presses. It also handles bilingual code-switching: ordering "beef chow fun" inside an English sentence, or mixing Cantonese dish descriptions with English modifiers, works the way a guest would actually speak. Speech recognition is trained on diverse acoustic models including regional accents and common bilingual phrasing, and where confidence falls below a threshold the host asks for clarification or redirects to a human operator rather than guessing.

The preparation this requires is smaller than it sounds and more valuable than it looks: a bilingual menu that holds both languages for the same dish, and a FAQ list whose answers exist in both languages rather than one language plus an improvised translation. Lao Cai on Marine Drive is the clearest example of why that matters — the host detects the caller's language within a word or two and answers in it, switching mid-call if the caller switches, and dish names work in either language, so 油潑麵 and "spicy hot-oil noodle" come out as the same order.

Where the AI stops and a person starts

Every checklist in this category eventually reaches the same boundary, so it is worth stating plainly rather than leaving it to a demo. DineCall's human handoff fail-safes are explicit: if speech confidence falls below 70%, or if the caller requests a human manager, the AI immediately alerts front-of-house staff by phone ring or dashboard notification, and logs a call-back ticket if the lines are busy. Reservations can be held for manual approval instead of confirmed automatically, which puts a person in the loop on exactly the bookings a host would want to see. And the seventh category on the operator map — how to reach staff — is where those handoff rules do the work rather than the script.

That boundary is the reason the checklist matters more than the model. The questions a host stand should never receive are the ones you did not write down: the allergy, the private event, the large party, the caller who wants a manager, the caller the voice did not understand. Write those down, mark them as hand-offs, and the AI can take everything else.

How to build the list before go-live

The preparation is a working session, not a writing project, and it is shorter than it looks. Start with the facts already published: hours, address, parking, directions, license details and the menu itself. Then add the exceptions — the closure day, the sold-out dish, the private dining room, the allergen that appears in the kitchen. Then add the decisions: the party size that goes to a person, the allergy that goes to a person, the booking you want approved manually, and the callback promise you want made. DineCall's own deployment sequence fits around this work, with configuration and AI training typically completed in 2 to 3 business days, including menu digitization, bilingual translation verification and routing tests. The restaurant keeps its existing public phone number, DineCall provisions a localized routing number, and the restaurant sets up conditional call forwarding for busy, unanswered or offline calls, or forwards all calls directly. No hardware installation is required in the restaurant; the order board runs on any web-enabled device.

Three operational points close the loop. Your existing point-of-sale system stays in place: DineCall operates alongside it and provides an open, documented ordering API that lets developers synchronise tickets into select POS terminals. The menu stays in one editor, with side-by-side English and Chinese translation management and bulk CSV import, so the phone and the QR code cannot disagree. And the line stays fast: response latency is under 800ms, with speech-to-text and text-to-speech tuned for rapid response to mimic natural human phone interaction.

FAQ

Which FAQ content should a restaurant prepare first?

Start with the operational facts that repeat every day — business hours, address, parking, directions, license details and allergen warnings, which is the documented scope of DineCall's FAQ handling — then the exceptions to each of those, then the menu modifiers and substitutions. The categories worth preparing in full are the ones this guide walks through: hours, address, parking and directions; menu options and modifiers; allergens and dietary questions; reservations and large parties; ordering and pickup rules; and the same question in the caller's language.

Does the AI phone host need to know the whole menu?

Yes, and it does. DineCall's AI host is grounded in the restaurant's own menu knowledge graph — the entire menu including complex modifiers such as "extra spicy" and "sauce on the side", sizes, substitutions, combos and current pricing — and it handles bilingual code-switching, so a Cantonese dish name inside an English sentence works the way a guest would say it. Because the menu is the source, what you prepare is the modifier rules and the exceptions, not a transcription.

What happens when a caller asks about an allergy?

The host answers on the facts it has — allergen warnings such as MSG, nuts and gluten are inside the documented FAQ scope — but a request it should not decide on its own, such as a large party, a private room or an allergy, is handed to a person live or captured and flagged for your team, depending on how you configure it. Write down which of your questions are answers and which are hand-offs before go-live, because the default under pressure is to answer.

Can it take a table reservation as well as an order?

Yes, on the same line and the same host. It checks live availability against service hours, available time slots, party-size limits, per-slot capacity and dining duration, then either confirms the booking or holds it for staff approval, and phone bookings appear immediately in the owner dashboard. Booking policy settings cover the booking horizon, minimum notice, maximum party size, active-booking limits per phone number, table-lock duration, last seating, slot size and dining durations by party size.

What if the caller wants a person, or the AI is not sure?

Both routes lead to your staff. If speech confidence falls below 70%, or if the caller requests a human manager, the AI immediately alerts front-of-house staff via phone ring or dashboard notification, and logs a call-back ticket if the lines are busy. A caller with a heavy accent is not lost either: recognition is trained on diverse acoustic models, and below the confidence threshold the host asks for clarification or redirects to a human operator.

Will the FAQ list still be right after the menu changes?

That depends on whether there is one source. Changes to items, prices or availability in the bilingual menu CMS reflect instantly across the AI voice agent, the website and the QR codes, so a change made once reaches the caller on the phone and the guest at the table together. Your FAQ list needs the same treatment: keep the hours, the exceptions and the policy values in one place that you update, rather than in the heads of staff who are on shift.