How to Take Restaurant Reservations on WhatsApp (2026 Guide)

A restaurant phone has one line and one pair of hands. Every call it takes during service is a person on the floor not being served, and every call it misses is a table that goes to whoever picked up. Guests know this, which is why they increasingly message instead of ringing, and why a "we'll answer tomorrow" reply loses the booking to the place that answered in ten seconds.

WhatsApp is where that message lands in most of the world. It is also the one channel where the conversation persists: a guest who booked in March still has the thread in May, so the next booking is one message away instead of a search for your number.

The catch has always been staffing it. Somebody has to read the message, check the book, and reply, at exactly the moment nobody is free. An AI agent takes that shift: it answers from your menu and your policies, it reads your calendar for open slots, and it hands anything unusual to a human. The rest of this guide sets one up, from the number to the first booking, without writing code.

Four things, none of them technical:

• A number for the restaurant. A new SIM, a virtual number, or the landline guests already call (Meta verifies landlines by voice call). One rule: a number cannot be in the WhatsApp app and on the Business API at once, so if your host stand answers WhatsApp on a phone today, start the bot on a second number.
• Your menu, in any form. A menu page on your website, a PDF, or a plain list. Allergens and dietary labels matter more here than photography.
• A calendar the kitchen already trusts. Google Calendar or Cal.com. The agent reads free slots from it and writes confirmed bookings back into it, so the reservation appears where your team already looks.
• A bot platform. This guide uses Chatloom's WhatsApp AI chatbot; the general setup is covered step by step in how to make a WhatsApp AI bot, so here we stay on the restaurant-specific parts.

Most restaurant bots fail here, not at the technology. Owners upload a beautiful PDF menu and nothing else, then wonder why the bot cannot answer "are you open on Monday?"

Write down the questions your staff answer twenty times a week. They are almost always the same list: opening hours including the days you are closed, whether you take walk-ins, parking and the nearest metro, whether there is a vegetarian and a gluten-free option, whether children are welcome and whether high chairs exist, whether the terrace is heated, the corkage policy, the largest group you seat, and what happens when a booking is late.

Then add the menu with prices and allergens, and your reservation policy in plain words: how far ahead you take bookings, how long you hold a table, deposit rules for large groups. Feed all of it to the agent, whether by pointing it at your website or uploading documents. Answers are then generated from your material rather than invented, which is the whole difference between a useful bot and an embarrassing one; the mechanics are explained in what a RAG chatbot is.

One tip that costs nothing: write the allergen information as text, not only inside a menu image. A photo of a menu is a photo to a machine, and allergen questions are exactly where a wrong answer stops being an inconvenience and becomes a health risk.

A bot that says "noted, see you Friday" and writes it nowhere is worse than no bot. Connect Google Calendar or Cal.com and the agent gets two abilities: it can read your open slots before promising anything, and it can write the confirmed booking back as an event with the guest's name, party size, and phone number.

That means the reservation shows up in the same calendar your front of house already has open, on every device, without anyone copying anything across. When a guest asks for 8pm and 8pm is taken, the agent offers the nearest slot it can actually see rather than guessing.

Two honest limits, because they decide whether this fits your restaurant. First, this is calendar booking, not table management: it does not model floor plans, covers per slot, or table turning. If you run a 200-cover room on a dedicated system, keep that system as the source of truth and use the bot for the enquiries around it. Second, if a guest never gives an email, no automatic reminder goes out the day before, because the reminder rides on the calendar invite. Restaurants that care about no-shows send a WhatsApp template the morning of the booking instead, which is a two-minute job from the dashboard.

Write this list before you go live, not after an incident.

Allergies beyond what is written. The agent should quote your allergen sheet and, for anything not on it, put a human in the conversation. Chatloom's agent says when something is not in its material rather than improvising, and a request for a human hands the thread to your team with the full history attached.

Large groups and private hire. A twelve-top on a Saturday is a commercial decision with a deposit attached. Let the bot collect the details and pass them on.

Complaints. Someone typing about a bad meal wants a person, quickly. Route it to a human every time; the bot's job is to notice and hand over, not to apologise on your behalf.

Anything involving money. The agent talks; it does not take payment. Deposits and prepaid events stay on whatever you use today.

Everything left over after those four is the bulk of your inbox, and that is what the bot is for.

Owners test with the questions they wish guests asked. Guests write "u open tmrw?" at 1am, in the wrong language, with a photo of a dish they liked on Instagram.

So test that way. Ask in fragments. Ask in a second language if your city has tourists, and check the answer comes back in that language, not in English with a foreign menu pasted in. Send a photo of a plate and see whether the reply engages with it. Ask something your documents deliberately do not cover and confirm the agent says so instead of inventing a Monday opening you do not have.

Then book a table as a stranger would, and go look at your calendar. The event should be there, at the right time, with the party size and a way to reach the guest. If that loop works end to end, you are ready. If it does not, fix it now rather than discovering it on a Friday.

Set the tone while you are here. A neighbourhood trattoria and a tasting-menu restaurant do not sound the same, and the agent's personality is a settings field, not a development project.

The measurable part is quiet: the phone stops ringing for things that were never worth interrupting service for. Hours, address, parking, vegetarian options, whether you take walk-ins, whether the terrace is open. Those are the calls that pull a host away from a queue of guests, and they are the ones a bot absorbs completely.

The second change is the bookings you were never getting. Messages that arrive at midnight, during service, and on your closed day currently wait for someone to notice. Answered instantly, a share of them become tables. That is the number worth watching in your first month: not "conversations handled", but covers that came from a message nobody would have answered in time.

The third is language. In a tourist city, the guest who writes in Italian and gets a fluent Italian reply books; the one who gets an English wall of text often does not. The agent detects and answers in the guest's language across 95+ of them, which is one of those features that is invisible until you see a booking come in through a language nobody on your staff speaks.

What does not change: the food, the room, and the judgement calls. A bot that knows its limits protects all three.

Two meters, and for a restaurant one of them is usually zero.

Meta's meter. Replying to a guest inside the 24-hour window that opens when they message you is free. That covers essentially every reservation conversation, because guests start them. You only pay Meta when you start the conversation with a template, such as a booking reminder or a "we have a table tonight" note, priced per delivered message by category and country.

The platform's meter. On Chatloom, WhatsApp is included on every plan including the free one, and plans differ by how many AI-answered messages you get per month, where one exchange means one guest question plus the bot's reply. A single restaurant's reservation traffic is small in these terms; the honest advice is to start free, watch a real month, and only then pick a plan. The full picture, including where competitors add per-seat and per-contact fees, is in our AI chatbot pricing comparison.

If you would rather see the whole thing laid out for a restaurant specifically, including menus and tourist languages, the WhatsApp chatbot for restaurants page covers it.