AI Chatbot for Restaurants: Hours, Bookings and Hand-Offs to Staff

A restaurant chatbot should answer the fixed facts itself: opening hours, address, parking, the booking link, the current menu and delivery or takeaway rules. It should hand over to a person for allergen questions, complaints, large parties, special occasions, refunds and anything it cannot find in the information you gave it.

Decide those rules on paper before you pick any tool. Vendors will happily sell you a bot that tries to answer everything, and the failures that hurt a restaurant, such as a wrong allergen reply or a promised table that does not exist, come from a bot that was never told where to stop. What follows is the checklist I would work through first.

Follow me on Instagram@sagnikteaches

Sort every guest question into three piles

Start with a list of real questions, not imagined ones. Open your last 50 messages across email, Instagram, Google Business Profile and the website contact form, and copy each question onto its own line. Phone calls count too: ask the front-of-house team to note what they were asked in one busy week. Then sort each question into one of three piles.

Connect on LinkedInSagnik Bhattacharya

• Answer alone. The answer is a fact that sits in one place and changes rarely. Hours, address, nearest car park, whether you take walk-ins, whether dogs are welcome on the terrace, how to reach the booking page.
• Answer, then offer a person. The bot can give a general answer but a human should follow up. Private hire, a table for eight, a birthday cake, a request to bring a pushchair into a tight dining room.
• Hand over immediately. The bot must not attempt an answer. Allergens and intolerances, a complaint about last night, a refund, an injury, a lost item with a name attached, a press or supplier enquiry.

Here is how the sort looks for an illustrative 40-seat neighbourhood restaurant that opens for dinner Tuesday to Sunday and lunch at weekends. The counts are made up to show the shape, not a benchmark.

Subscribe on YouTube@sagnikteaches

In this sample, 23 of the 50 questions can be answered without a person, 13 need an answer followed by a person, and 14 need a person at once. That split is the whole business case: the bot clears the repetitive questions so staff spend their phone time on the ones that need judgement. If your own sort comes out at 90 percent hand-overs, a chatbot is the wrong first project and a shared inbox rule or a phone service may fit better. The phone answering cost tutorial covers that route.

The facts the bot may state, and where each one lives

A chatbot is only as accurate as the page it reads. The safest set-up gives it a single short document, no longer than two printed pages, and tells it to answer only from that document. Everything it may state as fact goes in, each item with an owner and a review date. If you have not yet written the question list, the chatbot FAQ tutorial shows how.

Fixed facts

• Opening hours for each day, plus the kitchen's last order time, which is often 30 to 45 minutes before closing and the thing guests actually need.
• Address, the nearest parking or transport, step-free access yes or no, and whether you have a high chair.
• The booking link, the cut-off for online bookings, and the phone number for same-day requests.
• Walk-in policy, how long you hold a table after the booking time, and the deposit or cancellation rule if you have one.
• Dress code, dogs, children, and terrace rules.

Facts that change

• The menu. Paste the current menu with dish names and the description as you print it. Mark which dishes are seasonal and add a line telling the bot the menu changes and to point to the live menu page for the latest.
• Public-holiday and special-event hours. Add a dated block at the top of the document and delete it afterwards.
• Offers, set menus and gift-card rules, each with an end date.

Write the document in plain sentences, not a data dump. "Last food orders are 9.15pm Tuesday to Thursday and 9.45pm Friday and Saturday" works better than a table of times, because the bot will quote it nearly word for word.

Hours: the answer that goes wrong most

Opening hours look easy and cause the most visible mistakes, because the bot has to combine a date, a day of the week and an exception. Test this specifically. Ask "Are you open on the 26th?" and "Can I get a table at 10pm on Friday?" The second is a last-orders question, not an opening-hours question, and a weak bot says yes because you close at 11.

Give the bot the rule in its instructions:

Opening hours are listed below. A table can only be booked up to the kitchen's last-order time, not the closing time.
If someone asks about a date that falls on a public holiday, a private event or the exceptions list, use the exception, not the usual day.
If the date they mention is more than 14 days ahead and not on the exceptions list, say hours may change and give the booking link.
Never say "we are open" for a date you cannot match to the list.

The last line matters. It gives the bot a legitimate way to say "I cannot confirm that date" rather than guess.

Bookings: link, check or take

A chatbot can do one of three jobs with a booking, and each carries a different risk.

1. Send the link. The bot answers "how do I book?" with the link and one line on what the page does. Lowest risk, works with any chat tool, and covers most small restaurants.
2. Check availability. The bot reads live availability from your booking system. This needs an integration with the booking tool you already use. Ask the vendor to show a real booking appearing in your system, not a demo screen.
3. Take the booking. The bot collects name, party size, date and time, and writes it into your system. Only do this when step 2 has worked for a month with no mismatches.

For a small restaurant I would start at step 1 and move up only if staff time on bookings justifies it. The reason is plain: a booking the bot confirms but your system does not hold is a guest arriving to no table. When you do reach step 3, make the bot repeat back the details and end with a line like "You'll get a confirmation by email; if you don't, please phone us."

Party size is the commonest edge. Set a number, say seven or eight, above which the bot stops handling the booking and hands over with the date, party size and contact details already collected, so the staff member does not start from scratch.

Hand-off triggers to write down

A hand-off trigger is a rule the bot follows without judgement. Write them as plain conditions the bot can match, and keep the list short enough that staff remember it.

• The guest mentions an allergy, intolerance, coeliac disease, "safe to eat", "contains", "traces" or a medical condition.
• The guest says complaint, refund, sick, ill, poisoned, injured, cold food, wrong order, overcharged or a similar word.
• The guest asks for a human, a manager or "someone".
• The party is larger than your limit, or the request is a private hire, a buy-out, a wedding or a function.
• The guest has asked the same question twice and the bot has answered twice.
• The bot cannot find the answer in its document.
• The guest is upset. Tell the bot to look for signs such as capital letters, repeated punctuation, or swearing, and hand over rather than keep chatting.

Hand-over has two parts: what the guest sees, and what the staff member receives. For the guest, one calm sentence: "I'll pass this to the team now. You can also phone us on [phone number] until [time]." For staff, a short summary with the guest's name, contact details, what they asked, and what the bot already said. Without the summary, the guest repeats everything and the hand-over feels like a wall.

Say when a person will actually reply. If the restaurant is closed and nobody reads messages until noon, the bot should say that. "We reply to messages between 11am and 10pm" is honest and manageable; "someone will be with you shortly" at midnight is not.

Allergen questions stay with staff

This is the one rule I would not negotiate. An AI chatbot reads text and produces a likely answer. It does not know that the chef swapped the stock this week, that the fryer oil is shared, or that a supplier changed a sauce. A wrong "yes, that's nut-free" can put someone in hospital, and the duty to give allergen information rests with the business, not the software.

Give the bot one fixed reply and forbid anything else. A version that works:

If anyone mentions an allergy, intolerance or dietary medical need, do not say whether any dish is safe or contains anything. Reply exactly:
"Thank you for telling us. Allergen questions are answered by our kitchen team so you get an accurate answer. Please phone us on [phone number] or tell your server when you arrive. I'm passing your message on now."
Then trigger the hand-off. Do not add anything.

"Exactly" is deliberate. If you let the bot rephrase, it can drift into helpfulness. The allergen tutorial goes deeper on what a restaurant can safely publish. Also check your local food-safety guidance on what you must tell guests, because the rules differ between places, and ask a qualified adviser if you are unsure.

Menu questions that are not allergy questions, such as "Do you have vegetarian options?", can be answered from the menu, but add a sentence: "For allergies or strict dietary needs, please speak to the team." Ask the bot to include it whenever dietary words appear.

A worked example: the Friday-night rewrite

Take the 40-seat restaurant again. The owner builds a bot on a plain website-chat tool, gives it a two-page document, and tests it with five questions. Illustrative results:

1. "What time do you close Saturday?" Correct, but it said "11pm", not the 9.45pm last-order time. Fix: add the rule from the hours section.
2. "Table for 2 at 8pm tonight?" The bot said "Yes, that's available". It cannot see availability. Fix: forbid confirming availability and send the link plus the phone number for same-day.
3. "Is the risotto gluten-free?" The bot said "it looks like it uses rice, so likely yes". Fix: the fixed allergen reply above, with the trigger words.
4. "Do you do a kids' menu?" The bot invented one. Fix: "Only mention dishes that appear in the document."
5. "My food was cold last night, I want a refund." The bot apologised and offered a 10 percent discount it had no authority to give. Fix: the hand-off list, plus "never offer refunds, discounts or compensation".

All five bad answers come from one cause: the bot was allowed to be helpful beyond its document. Each fix is a sentence in the instructions, and the second round of tests passed. Keep a record of what you changed and why in a one-line-per-fix list. It becomes your change log when something breaks later.

A test script you can run in 20 minutes

Before launch, and after every change, run the same script. Use a private browser window so the bot treats you as a new guest, and score each reply pass or fail.

If any of the allergen, complaint or hours tests fail, do not launch. Fix the instruction and run the whole script again, not just the failed line, because a change to fix one answer can break another. The guardrails tutorial goes into the instruction wording that stops a bot promising what you do not offer.

Choosing a tool after the rules are written

Once the rules exist, tool choice is simpler because you can test each one against them. Check these four things in any trial:

• Can you restrict the bot to answering from your document only, so it does not fall back on general knowledge?
• Can it hand a conversation to a person, with a transcript, by email, a messaging app or a shared inbox that your team actually reads?
• Can you see every conversation afterwards? You need this to improve it.
• What happens when your monthly conversation allowance runs out: does it stop, switch off, or bill more?

As of October 2026, free tiers exist but are small: Botpress lists 100 conversations a month on its free plan, Tidio gives 50 lifetime AI conversations, and Chatbase has no free plan, with its Hobby plan at $32 a month billed yearly or $40 month to month. Vendors change these often, so treat them as a guide and read the vendor's pricing page before relying on one. A restaurant with a quiet website might stay inside a free allowance; a busy one will not. The restaurant AI vendor questions tutorial lists what to ask before signing anything.

If your restaurant has customers in the EU, the chatbot transparency rules mean guests should be told they are talking to an AI. Do it anyway everywhere, in the first message. The disclosure tutorial has wording.

What staff need to do for the bot to work

The bot is not a separate system that runs itself. It creates three small jobs for your team, and if nobody owns them, it quietly degrades.

1. Someone reads hand-offs. Name the person or role for each shift. If nobody is on during the afternoon prep, say so in the bot's reply.
2. Someone updates the document. One person, with a calendar reminder at the start of each month and a rule: if you change the menu or hours anywhere, you change the bot's document in the same ten minutes.
3. Someone reviews conversations weekly. Read ten, not all. Look for questions the bot could not answer, wrong replies, and hand-offs that should not have happened. Each finding becomes one new line in the document or the rules.

The review is where the value compounds. After a month you will see that guests ask about things you never thought of: "Is there step-free access to the toilets?", "Can I bring my own cake?", "Do you have a quiet table?" Add the answers to the document or the hand-off list.

Three more restaurants, three different rule sets

The same checklist gives different answers depending on the business. Illustrative cases:

A takeaway-led kitchen

Most questions are "is my order on its way?" and "can I change my order?" The bot can state delivery areas, minimum order and typical wait times, but order changes and anything about a live order should go to a person at once, because the kitchen may already have cooked it. Hours matter less than the "kitchen closes for orders at" time.

A pub with food

The questions split between "is the kitchen open?", "can I book the function room?" and "do you show the match?". The match listings change weekly, so the document needs a weekly update or the bot should say "check our social page for this week's fixtures". Function-room enquiries are a hand-over with a date and headcount collected first.

A fine-dining room with a tasting menu

Guests ask about the menu a great deal, dietary adjustments are frequent, and deposits are common. The bot should explain the deposit and cancellation rule exactly, never agree to a dietary adjustment, and hand over every dietary request with the booking date attached. Here the bot's value is capturing detail and being polite at midnight, not answering many questions.

Signs the rules need tightening

Watch for these in your weekly review, and treat each as a reason to tighten the rules rather than blame the tool:

• Guests arrive for a table they believe is booked and is not. The bot confirmed something it could not see.
• Staff get hand-offs with no context. The summary format needs fixing.
• The bot repeats the same non-answer. Add the missing fact or a hand-over trigger.
• Guests say "it was no help" in reviews or messages. Read those conversations first.
• Hours are wrong on one day only. You changed them in one place and not the other.

A quick sum helps decide whether to keep going. If the bot clears 25 questions a week and each would have taken a minute and a half of staff time to answer, that is about 37 minutes a week, which is modest. The larger gain is the questions that arrive when nobody is in: at night, at weekends, and during service. If those are converting to bookings through the link, that is where the bot earns its place. If you want to map this against your own guest questions and booking tool, that is the kind of thing my AI implementation consultation covers.

Restaurant chatbot questions owners ask next

Can the chatbot take the booking itself?

Only if it connects to your booking system and can see real availability. Most small restaurants do better with a chatbot that sends the guest to the booking link and confirms what the link does. A bot that promises a table it cannot see creates double bookings and angry arrivals.

What should it say when it does not know?

Say it is not sure, give the phone number or opening times of the person who can answer, and offer to pass the message on. Write that sentence yourself in the setup. A confident guess about a menu or a closing time is worse than an honest hand-over.

Does it need to say it is a bot?

Yes, say so in the first message. Guests are happier when they know, and some places have rules that require it. The wording is covered in the disclosure tutorial linked below, and your own chat tool will usually have a setting for the opening line.

How often should I update what it knows?

Update it the same day anything changes: hours for a public holiday, a new menu, a closed kitchen. Put one person in charge, and add a monthly ten-minute check where you ask it the five questions guests ask most and compare against your website.

Further reads

• AI Bot for Restaurant Reservations: Chat Booking Setup and Cost — Set up chat booking once the link-only stage works.
• When Should an AI Chatbot Hand Over to a Human? — General hand-over rules for any business chatbot.
• How Much Does AI Cost a Small Restaurant Each Month? — What a small restaurant spends on AI tools in total.
• How to Answer Restaurant Complaints With AI Without Escalating — Handle the complaints your bot hands over.
• How to Set Up Human Review for AI Work Without Slowing Down — Set up checking that does not slow the team down.
• Can a B&B Use an AI Chatbot and Keep the Personal Touch? — Keep a human tone when a small venue uses a chatbot.
• Maple AI for Restaurants: Voice or Pro Plan, and Who Needs Which — Maple Voice answers and transfers calls; Pro adds orders and reservations. Compare the monthly cost, break-even orders and the test to run first.
• Slang.ai for Restaurants: Is It Worth It for an Independent? — Slang's Core and Premium tiers, the booking systems it connects to, private-events capture and a break-even sum for an independent.
• AI Chatbot Online for Hospitality: Questions a Café Can Safely Automate — A checklist that sorts café guest questions into chatbot-safe answers, answers that need a check, and hand-offs to a person.
• Restaurant No-Shows: Deposit Rules by Party Size, AI Confirmations — Set deposit and card-hold rules from your own no-show log by party size, then use AI for policy wording, reminder texts, reply sorting and waitlist offers.
• AI for Hospitality: Which Jobs to Delegate and Which Stay With Staff — Which guest messages AI can answer, which it can only draft, and which stay with staff, with a one-page rule sheet and a worked week of 120 messages.
• WhatsApp AI Chat for Restaurants: Should You Switch On Meta's Agent? — Meta Business Agent can answer WhatsApp questions about hours and menu. Who it suits, what it costs and how a restaurant should trial it first.
• Restaurant Waitlist Software: Where AI Helps at the Door — AI can quote waits by party size, text when a table is ready and answer waitlist calls. Here is what it still needs a host at the door to do.
• What Is a Ghost Kitchen? How an Existing Restaurant Can Test One — What a ghost kitchen is, and how a restaurant can test one from its current kitchen on one delivery app: menu, per-order economics and a four-week log.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Botpress, Tidio and Chatbase pricing pages (checked October 2026); general restaurant front-of-house practice.