How to use AI in restaurants: from bookings to reviews

AI is used in restaurants for six jobs: replying to booking requests from WhatsApp, Instagram and the website, picking up the phone during service, sending a confirmation on the day of the booking, offering cancelled tables to the waiting list, answering menu questions and drafting replies to Google reviews. The food, the table plan and the relationship with guests stay with the team.

A restaurant's phone tends to ring at the busiest moment, in the middle of service. A call nobody can answer while carrying plates is a lost booking. An Instagram message waits until after service, sometimes until the next day. AI takes over the repetitive part of this load and leaves the kitchen and the floor alone.

Our own research shows that booking and menus often exist on a restaurant's website but cannot be reached from its Google listing. On 22 September 2026 we checked the Google listings of 93 fish restaurants along the Bosphorus. Only 17 of them (18 per cent) had a booking link on their listing, and only 13 (14 per cent) had a menu link. Yet 35 of the 54 websites we checked offered booking and 45 had a menu. The full results are in our Bosphorus fish restaurant study.

Which jobs can AI help a restaurant with?

1. Booking requests: free tables are shown by party size and time, and the booking goes into the book.
2. Phone calls: calls during service are picked up by a voice assistant and the request is logged.
3. Confirmations and reminders: on the day of the booking, the guest receives a WhatsApp confirmation.
4. Filling empty tables: a cancelled table is offered to guests on the waiting list.
5. Menu questions: questions about vegan options, the children's menu, parking and opening hours are answered from the restaurant's own information.
6. Review replies: Google reviews gather on one screen with a reply drafted for each.

How are booking requests handled during service?

Requests come in by phone, WhatsApp, Instagram and the website form. The setup gathers these channels into one queue. The table plan, sitting length, capacity and closing days are written as rules once: tables for four, two-hour sittings, closed on Mondays. Each request is answered with the times that fit those rules, the guest picks one and the booking goes into the book.

Guests are told clearly that they are writing to an AI assistant. Large groups, birthdays, set menus or terrace requests pass to a member of staff, and when the answer is not known nothing is made up. On the phone, the call is picked up by a voice assistant and the request is logged; the restaurant gives the final confirmation. We describe this on our restaurant reservation system and online booking and scheduling pages.

Here is how the setup works in a restaurant. On a Saturday evening, in the middle of service, a WhatsApp request comes in for a table for six. Free times are shown, the guest picks one and the booking goes into the book. At noon on the day, the guest receives a confirmation. If they cancel, the table opens up again and the first guest on the waiting list is told.

How can no-shows and empty tables be reduced?

A no-show is a table standing empty at the busiest hour of the evening, while the guest at the door may already have been turned away. The simplest answer is confirmation and reminders. In the booking flows we build, a reminder goes out a day before and two hours before, with a cancellation link inside. The guest cancels without calling, the slot returns to the calendar and the waiting list is told.

If you want, a deposit can be taken for certain times or for large groups. The restaurant sets the amount and the refund rule, and the system never issues refunds on its own. At the end of each month, a report shows occupancy, the no-show rate and the busiest hours, so the evening for a promotion is chosen from that table rather than by guesswork.

What goes into a confirmation matters too. Turkey's Regulation on Commercial Communication and Commercial Electronic Messages does not require prior consent for notices about a purchase and similar matters, but it bans promoting goods or services inside them. A confirmation therefore carries only the booking details, and promotions go only to guests who have agreed to receive marketing messages.

WhatsApp has its own rule. Under Meta's policy, when the business starts the conversation, or more than 24 hours have passed since the guest's last message, it must use a template Meta has approved in advance. Confirmations and reminders are therefore set up as approved templates. We cover this on our WhatsApp Business integration page.

How are the menu and prices kept up to date from one place?

When a price changes, several things in a restaurant have to change with it: the price list at the entrance, the QR code on the table, the menu on the website and the menu link on the Google listing. If one is missed, a guest sees one price online and another at the table.

In Turkey this is more than a matter of guest satisfaction. Under the price labelling regulation, restaurants, cafes and patisseries display the price list at the entrance and on the tables. At the table the list may be shown by QR code, but it must be given to any guest who asks for it. Every item on offer must also appear on the price list.

In the setup, the menu and prices are updated from a single admin screen. The QR menu and the website menu read from the same source, so a price changes once and shows in both places at the same time. A sold-out dish can be hidden with one tap. Printing the list at the entrance from the same record makes it easier to keep every price in line. We describe this on our QR menu page.

How are guests' menu questions answered?

Before booking, guests tend to ask about the same things: vegan or gluten-free options, a children's menu, parking and closing time. These are answered from the restaurant's own information, read from the menu record, so when the menu changes the answer changes too. We describe this side of the setup on our chat and voice assistant page.

Allergen questions are kept apart. Which dish contains which allergen is known in the kitchen. Because the question concerns the guest's health, it is not answered by the assistant; it passes to the kitchen or the floor manager.

An allergy written in a booking note also needs care. Article 6 of Turkey's data protection law (KVKK, Law No. 6698) treats data about a person's health as a special category of personal data. The note is kept where only the staff handling that booking can see it, and is not stored longer than needed. Guests are told in a privacy notice why the data is processed.

How are replies to Google reviews drafted?

A guest searching for a restaurant on Google Maps also sees its rating and reviews. New reviews gather on one screen, and a reply is drafted for each in the restaurant's own tone; a reply to a negative review includes an apology and a way to get in touch. The restaurant decides whether to post it, and no reply goes live without its approval.

Asking for reviews has limits. Google bans asking for reviews in return for discounts, free items or money. Asking only happy guests for reviews, or trying to discourage negative ones, breaks the same rule. So the review request goes to every guest in the same way.

A booking link and a menu link can be added to the Google Business Profile. If the booking link points to the restaurant's own booking page, guests book straight into the restaurant's calendar. We describe this on our SEO and digital marketing page and the restaurant website on our restaurant website design page.

What changes for foreign guests?

A report by the Istanbul Provincial Directorate of Culture and Tourism, based on the Ministry of Culture and Tourism's border statistics, puts foreign arrivals to Turkey in 2025 at 52.8 million, of whom 18.97 million entered through Istanbul. For a restaurant in a busy tourist area, offering bookings and the menu in a second language is no small detail. In our Bosphorus study, 24 of the 54 websites we checked had an English version.

In the setup, replies go out in the language the guest writes in. The menu is managed in Turkish and English from the same screen; a price is entered once and shows the same in both languages. The booking flow is built in both languages too.

Where should AI not be used in a restaurant?

• Allergen and health questions: the answer comes from the kitchen, and the question passes to staff.
• Complaints and bill disputes: a manager speaks to the unhappy guest.
• Prices and promotions: the restaurant sets prices, discounts and deposits.
• Messages and photos without consent: promotions go only to guests who agreed to them, and no guest's face is posted without permission.

Where should a restaurant start?

1. For one week, count which channel each booking comes from, how many calls go unanswered and how many tables stay empty.
2. Write down the table plan, sitting length, closing days and deposit rule.
3. Set up bookings and confirmation messages.
4. After a month, look at the no-show rate and occupancy, then decide whether menus or reviews come next.

ALTAI Digital starts with a discovery of about a week, covering the restaurant's booking flow, channels, table plan and menu, and then prepares a demo. If you like it, your requirements go into a detailed brief form and we build. First live use usually opens within four to eight weeks. After launch we keep measuring, sending a monthly report and maintaining the setup.

Every part joins one panel, so bookings, messages, the menu and reviews can be followed on one screen. We describe our approach for the sector on our restaurants and cafes page and the method on our digital transformation page.