It can make sense, as long as you’re clear about which calls it handles. An AI receptionist is good at the questions that interrupt your busiest hour: are you open, where do I park, do you take walk-ins, can I book for six on Saturday. It should not be the final word on anything where a wrong answer could hurt someone or cost you a night’s reputation, above all food allergies, and it shouldn’t confirm a table your team can’t actually hold. The real decision is less “AI or not” and more “what do we want a caller to experience at 7.30pm, when nobody can pick up?”

The call nobody answers at 7.30pm

Picture it from the caller’s side. They’re planning dinner for their parents tomorrow. They ring, it rings out, they try again, and then they book somewhere else that answered. They don’t leave a message. The restaurant may never know that call happened.

Now picture your side. The pass is full, two tables want their bill, someone at the counter is asking about takeaway, and the phone goes again. Whoever answers has to drop something. Either way, someone has a worse experience.

That’s the gap an AI receptionist is meant to fill: not to replace your front-of-house, but to make sure the phone is answered when they physically can’t.

What callers usually ask a restaurant

Before deciding anything, keep a tally for one ordinary week. For many restaurants, calls fall into a handful of groups:

Type of call Examples Who should handle it
Quick facts Opening hours, address, parking, nearest MRT, halal status, dress code AI can answer from what you’ve told it
Booking requests “Table for six on Saturday at 7?” AI can take the details; your team confirms
Changes Running late, changing numbers, cancelling AI can take the message; your team acts on it
Menu questions Set menus, vegetarian options, corkage AI can answer general questions you’ve approved
Allergies and dietary needs “Is the sauce nut-free?” “Can you do gluten-free?” A person who can check with the kitchen
Events and large groups Birthdays, company dinners, buyouts A person, after the AI takes the details
Complaints “Last night’s order was wrong” A person

If most of your calls sit in the top half of that table, an AI receptionist could take a lot off your team. If most sit in the bottom half, it probably can’t.

Why allergy questions need a person

This is the line we’d hold most firmly. A caller asking whether a dish contains nuts, shellfish or gluten is asking a safety question, and the true answer depends on today’s ingredients, today’s supplier and what else was cooked on the same surfaces. An AI receptionist only knows what it was told when it was set up. A chef knows what’s in the kitchen tonight.

The Singapore Food Agency describes food safety as “a joint responsibility shared between the Government, industry and consumers”, and points out that cross-contamination can happen at many stages, including handling (SFA, protecting consumers from food allergens, checked 12 November 2026). That’s exactly why an AI shouldn’t give a reassuring yes.

A safer pattern is for the AI to say something like: “I don’t want to guess about allergies. Let me take your name and number, and someone from the kitchen will call you back to check.” Then make sure someone actually does, before the booking.

If a caller describes a reaction happening right now, they need emergency help, not a restaurant. Whoever answers your phone should tell them to hang up and call 995.

Bookings: take the request, don’t over-promise

The biggest risk with bookings isn’t the AI mishearing a date. It’s the AI saying “you’re confirmed” for a table that doesn’t exist.

Unless your booking system is genuinely connected and kept up to date, the honest approach is:

  • The AI takes the name, number, date, time and party size.
  • It reads the details back so a misheard “sixteen” versus “sixty” gets caught on the call.
  • It says clearly what happens next: “I’ve passed this to the team. They’ll message you to confirm.”
  • Your team confirms quickly. A request that sits until tomorrow afternoon can feel worse than no answer at all.

Decide your own rules too. Do bookings over a certain party size always go to the manager? Do you hold tables for walk-ins on Friday nights? Whatever you decide, the AI should say it the same way every time.

When an AI receptionist is the wrong choice

  • You rarely miss calls. If someone almost always picks up, the problem may not be worth solving.
  • Most calls are about allergies or special diets. That’s a sign your callers need a person who can check with the kitchen.
  • Your bookings are complicated. Lots of deposits, set-menu negotiations or private events need judgement.
  • Your regulars expect to talk to you. Some restaurants are loved partly because the owner answers.
  • Nobody will read the call summaries during service. A fast answer followed by no follow-up disappoints people twice.
  • Callers keep asking for a person. If that happens often, listen to it.

Be open with callers

Two things callers should know from the start: that they’re talking to an AI, and, if calls are recorded, that they are. Callers are giving you their name, phone number and sometimes details about the people they’re dining with. Under Singapore’s PDPA, that’s personal data, so collect only what you need for the booking and say what it’s for. The PDPC publishes guidance on the PDPA’s key concepts if you want to read the detail. This isn’t legal advice.

A hypothetical call

This is an invented example. It’s 7.40pm on a Thursday. A caller rings [Business name], a neighbourhood restaurant. The AI says it’s an AI assistant for the restaurant, and the caller asks for a table for five on Saturday at 7pm, adding that one guest has a peanut allergy.

The AI takes the caller’s name and number, reads the booking back, and says: “I’ve passed your request to the team, and they’ll message you to confirm the table. I can’t check ingredients myself, so I’ve flagged the peanut allergy for the kitchen to call you about.” The manager sees the summary after the rush, confirms the table, and the chef rings the next morning. The AI didn’t decide anything important. It kept a caller who might have rung somewhere else.

A five-minute self-check

  1. How many calls did you miss during service last week? (A rough count is fine.)
  2. What were they mostly about? Quick facts, bookings, or allergies and events?
  3. Who would read a call summary during a busy night, and how fast?
  4. Which questions should never get an automatic answer at your restaurant?
  5. Would your regulars mind hearing an AI first, if it was honest about being one?

If your answers point to “lots of quick facts and booking requests, and someone can follow up within the hour”, an AI receptionist is worth exploring. If they don’t, that’s a useful answer too.

Hear it for yourself

The quickest way to judge is to be your own most awkward caller. We built an AI receptionist for our own enquiries. Her name is Penny. She tells callers she’s an AI in her very first line, and when she can’t answer something she takes the caller’s name, number and question, so it lands in a summary of who called, what they need and what to do next. You can talk to her at penny.pencilglow.com, by voice or chat, with no sign-up and no contact details needed.

For more on calls that can’t wait, see how an AI receptionist should handle urgent call-outs.