Guides

AI Chatbot for Restaurants in 2026: Bookings, Allergens and the QR Code on the Table

Around 70% of restaurant messages are hours, menu and practicalities questions a bot can answer from documents you already have. What to feed it, three booking patterns, the QR-code placement most venues miss, and three rules that keep it out of trouble.

Meda
Founder, PopABot
9 min read Last updated:

Key takeaways

  • Roughly 70% of restaurant messages are hours, menu, allergens and practicalities, all answerable from documents you already have.
  • Add your menu as plain text, never as a JPEG or designed PDF poster, because a graphical menu has no reliably extractable text.
  • Allergen answers must pair the fact with a handover: state the ingredients, then ask the guest to confirm with a server on arrival, because kitchens are shared.
  • The QR code on the table and in the window often outperforms the website widget, and multilingual replies convert the tourists who would otherwise walk on.

A restaurant website gets asked the same eleven questions forever. Are you open on Monday? Do you have a gluten-free option? Can I book a table for eight? Is there parking? Do you do the set menu at lunch? Every one of those messages arrives on a phone that is already ringing, at the exact moment the kitchen is at full tilt.

An AI chatbot answers all eleven from your own menu and opening hours, in whatever language the visitor's browser is set to, and only interrupts you when someone actually wants to book a table or asks something you haven't covered. This guide covers what a restaurant chatbot should do in 2026, what it should never do, and how to build one in an afternoon.

What guests actually ask (and what a bot should answer)

Before you build anything, look at your last 50 messages across Instagram DMs, Google Business chat, and the phone. In restaurant after restaurant the split looks like this:

Question type Share of messages Should the bot handle it?
Opening hours, holidays, kitchen closing time ~25% Yes, fully
Menu items, prices, daily specials ~20% Yes, if the menu is in the bot's data
Allergens, vegan, halal, gluten-free ~15% Yes, with a "confirm with staff on arrival" caveat
Table bookings and party sizes ~20% Yes, capture the booking; escalate large groups
Parking, accessibility, dogs, kids, wifi ~10% Yes, fully
Private hire, events, catering quotes ~10% No. Qualify, then hand to a human

Roughly 70% of the volume is answerable from documents you already have. That is the part worth automating first. The remaining 30% is where the money is, and that is the part worth routing to a person fast.

What to feed the bot

A restaurant chatbot is only as good as the four or five sources behind it. In PopABot these are called datasets, and you want these:

  1. Your menu as text, not a picture. This is the single biggest mistake restaurants make. A menu that only exists as a JPEG or a designed PDF poster cannot be read reliably. Paste the dish names, descriptions and prices as plain text. It takes 20 minutes and it is the difference between a bot that sells and a bot that shrugs.
  2. Opening hours, including exceptions. Bank holidays, the Monday you close, the kitchen closing an hour before the bar. Write the exceptions down explicitly; the bot will not infer them.
  3. An allergen and dietary sheet. One line per dish: contains nuts / gluten-free on request / vegan as standard. See the safety section below for how to word the bot's answers.
  4. A practicalities Q&A. Parking, nearest station, step-free access, high chairs, dogs, large screens for the football, corkage. Twenty Q&A pairs covers almost everything.
  5. Your website URL. The crawler picks up the pages you forgot. The free Starter plan crawls up to 100 pages per website dataset, which is far more than any restaurant site needs.

If your menu changes weekly, put the changing part in a single "this week" text dataset and rewrite that one field. Everything else stays put. On Pro, automatic dataset refresh re-crawls your site on a schedule so a menu page update flows through without you touching the dashboard.

Turning a question into a booking

Answering questions is table stakes. The reason to run a chatbot is the booking. There are three patterns that work in restaurants, and which one you pick depends on how you take reservations today.

Pattern 1: book straight into your calendar

If you keep your covers in Google Calendar, the bot can offer real time slots and write the booking in directly, with your own availability rules (service windows, last seating, blocked days). This is the appointment booking flow, and it works well for restaurants with set sittings and for anything consultative like a tasting or a private room viewing. Calendar booking is on Essential and above.

Pattern 2: capture the request, confirm by hand

Most independents still want a human eye on the book. In that case the bot runs a short flow (date, time, party size, name, phone) and submits it as a lead. You get an email, and on Essential and above an SMS to your phone, so a booking request that lands at 22:40 is on your screen before you've locked up. You confirm in one message. The guest never waited.

Pattern 3: hand off to WhatsApp for the awkward ones

Groups of twelve, a birthday with a cake, a wedding lunch, a dietary situation that needs the chef. Don't try to automate these. Have the flow ask two qualifying questions (date, headcount), then push the guest to WhatsApp with the details pre-typed. You pick up a conversation that already contains the facts instead of starting from "hi".

The QR code on the table is the underrated part

Everyone thinks of a chatbot as a website widget. In hospitality, the higher-value placement is often physical. A QR code printed on the table talker, the menu, the window, or the flyer opens a chat that can answer the wifi password, explain the specials, take a drinks question, or push the loyalty signup, all without anyone flagging down a server.

The same mechanism covers the venue that has no website at all. A hosted chat link is a URL you can put in your Instagram bio, your Google Business profile, or a printed code by the door. For a lot of small food businesses that link is a more honest "website" than a page nobody updates.

Tourists, and why multilingual matters more here than anywhere

If you're in a city centre, a coastal town, or anywhere near a station, a meaningful slice of your walk-in trade does not read your language. PopABot detects the visitor's browser language and replies in it, across roughly 100 languages, with no setup and no separate bot per language. A German family scanning the QR code on your table gets the allergen answer in German from the same English source menu you typed once.

This is the single feature most restaurant owners underestimate. It costs nothing extra and it converts the customers who would otherwise have walked to the place with the picture menu.

Three rules that keep a restaurant bot out of trouble

  1. Never let the bot give an unqualified allergen guarantee. The right answer format is factual plus a handover: "The mushroom risotto is made without gluten-containing ingredients. Because we prepare dishes in a shared kitchen, please tell your server about your allergy when you arrive so the kitchen can confirm." Write that caveat into your allergen dataset so it rides along with every answer.
  2. Never let the bot invent availability. If it is not reading a real calendar, it must not say "yes, 8pm is free". It should say "I've sent your request to the team, you'll get a confirmation shortly". Overpromising a table is worse than not answering.
  3. Never let the bot quote for private hire. Events pricing depends on the day, the season, and how much you want the job. Qualify and hand off.

Building it, step by step

  1. Create the chatbot and set the tone. Restaurants do well with warm and short. Two sentences maximum per reply.
  2. Add the menu as a text dataset. Plain text, dish name, description, price, one per line.
  3. Add a Q&A dataset with hours, parking, access, dogs, kids, wifi, corkage, and the allergen caveat.
  4. Add your website URL as a crawl dataset to catch anything you missed.
  5. Build the booking flow in the flow builder: date, time, party size, then a lead form for name and phone.
  6. Add a branch for groups over eight that routes to WhatsApp instead of the form.
  7. Turn on SMS notifications so booking requests reach your phone, not just an inbox you check at 11am.
  8. Test it on your own phone before publishing. Ask it the five things you get asked most. If any answer is vague, the fix is in the dataset, not the prompt.
  9. Publish it twice: the widget on your site, and a QR code for the tables and the window.

What to measure after two weeks

Not "messages sent". Three numbers matter: how many booking requests the bot produced, how many conversations ended with no useful answer, and which questions came up that you had not anticipated. The conversations view shows you the raw transcripts. Every unanswered question is one line to add to your Q&A dataset, and the bot is measurably better the next day. Two rounds of that and the answer rate stops being a problem.

Frequently asked questions

What can an AI chatbot do for a restaurant?

Answer opening hours and holiday exceptions, menu items and prices, allergen and dietary questions, parking, accessibility and wifi, and take table booking requests, all from documents you already have. That is roughly 70% of a typical restaurant's inbound messages. It should escalate private hire, events and catering quotes to a human, because those depend on the day and the season.

How do I get a chatbot to answer questions about my menu?

Add the menu as plain text, not as a picture. This is the single biggest mistake restaurants make: a menu that only exists as a JPEG or a designed PDF poster has no reliably extractable text. Paste dish names, descriptions and prices as text, one per line. If the menu changes weekly, keep the changing part in its own short "this week" document so updating it is a single edit.

Can a chatbot safely answer allergen questions?

It can answer factually, but never with an unqualified guarantee. The correct format is the fact plus a handover: "The mushroom risotto is made without gluten-containing ingredients. Because we prepare dishes in a shared kitchen, please tell your server about your allergy when you arrive so the kitchen can confirm." Write that caveat into your allergen document so it travels with every answer.

Can a restaurant chatbot take table bookings?

Three ways. If you keep covers in Google Calendar, it can offer real slots with your own rules (service windows, last seating, blocked days) and write the booking straight in, on Essential and above. If a human needs to see the book, it captures date, time, party size, name and phone as a lead and sends you an SMS. And for awkward bookings such as large groups, it should qualify briefly then hand off to WhatsApp with the details pre-typed.

Where should a restaurant put a chatbot QR code?

The table talker, the menu, the window and the flyer. In hospitality the physical placement often beats the website widget, because the guest with a question is sitting in your room. A window code also captures the person who arrives after closing, which is otherwise a completely lost enquiry. Print at least 2 cm across and test it on your own phone before a print run.

Do I need a website to run a restaurant chatbot?

No. A hosted chat link works from your Instagram bio, your Google Business profile or a printed QR code, with no website at all. For a lot of small food businesses that link is a more honest front door than a page nobody updates.

Will the chatbot answer tourists in their own language?

Yes. It detects the visitor's browser language and replies in it across roughly 100 languages, generated from the English menu and FAQ you typed once. A German family scanning the code on your table gets the allergen answer in German with no extra setup and no separate bot per language. For city-centre, coastal and station-adjacent venues this is usually the highest-value feature.