A restaurant's phone rings with the same questions every evening, usually at the exact moment the floor is busiest. Are you open on Monday? Do you have Jain food? Is there parking? Do you deliver to Baner? Can twenty of us book the upstairs room on Saturday? Almost every one of those answers is already written somewhere on your website, in the menu, the timings, the contact page or the FAQ.
A chatbot that reads those pages can answer them while your staff look after the people already seated. It cannot take a booking or an order, and it should never guess about allergens. The product steps below are TailGPT's, and the plan figures are as of 4 October 2026.
Which diner questions can a chatbot answer from your website?
Any question whose answer is written on a page or document the chatbot has read. TailGPT answers only from the pages you tick and the files you upload, with no general knowledge mixed in, so the useful question is not "what can the bot do" but "what does my website actually say".
For most restaurants and cafes, the questions fall into a handful of groups:
| What diners ask | Where the answer has to be written |
|---|---|
| Are you open today? What time does the kitchen close? | Timings page or footer: days, hours, last order time, weekly off, holiday closures |
| Where are you? Is there parking? | Contact or visit page: address, landmark, floor, parking or valet, nearest metro |
| Do you have veg, Jain or vegan options? | The menu, with each dish marked in words |
| How much is the thali? Is there a lunch buffet? | The menu with prices, or a separate offers page |
| Can we book for a birthday? Do you have private dining? | An events page: capacity, set menus, minimum spend, advance notice |
| Do you deliver to my area? Is takeaway available? | A delivery page listing areas, minimum order and hours |
| Can we bring a cake? Are children and pets allowed? | An FAQ or house rules page |
If the answer is in that right-hand column, the chatbot can give it in a sentence, at eleven at night or during the Sunday lunch rush. If it is not, the chatbot says plainly that the website does not cover it, instead of inventing a closing time.
How should it handle allergens and dietary questions?
Only from what your menu says, and never by guessing. A diner asking whether the paneer tikka is nut-free is asking a safety question, and the honest answer depends entirely on what you have written down.
Because TailGPT answers only from your pages and documents, it will not reach for a general recipe and assume the makhani gravy has cashews in it. But that cuts both ways. If your menu describes a dish as "paneer in a rich tomato gravy" and does not mention the cashew paste, the chatbot has nothing to warn anyone about. It can only be as complete as your menu is.
So treat allergens as something you either list properly or do not list at all:
- If you list allergens, list them per dish, in words. "Contains: nuts, dairy" beside each item is something a chatbot can quote. A row of small icons with a key at the bottom of a designed menu often is not.
- Write a kitchen statement in your own words. Something like "Our kitchen handles nuts, gluten, dairy and sesame, and we cannot guarantee any dish is free of them. Please tell your server about any allergy before you order." The chatbot can then give that sentence to every allergy question, which is the safest answer you have.
- Mark Jain and vegan dishes explicitly. "Jain" means no onion, garlic or root vegetables, and many kitchens can make a dish Jain on request. Say which dishes, or say "most curries can be made Jain on request; ask your server". Vegan is harder, because ghee, cream and butter hide in dishes people assume are plant-based. If you have not checked, do not mark it.
Then test it with "Is the paneer tikka nut-free?" and "What can a Jain guest order?". If either answer says more than your menu does, the menu needs a line. The chatbot should never be the allergy check; your staff are.
What about menus that live in a PDF or a photo?
Upload the PDF, and replace any photographed or scanned menu with a version that has real text. Many restaurant websites show the menu as images, or link to a PDF that is really a photograph of the printed card. Neither has any words for a chatbot to read.
TailGPT takes documents alongside your pages: PDF, Word (.docx) and text files, with no limit on how many. An uploaded menu is read the same way a page is, and answers start using it within seconds. A scanned PDF with no text layer fails with a clear reason, so you will know.
A quick test before you upload: open the PDF and try to highlight a dish name. If you can select the words, there is text in it. If the whole page highlights as one block, it is an image. In that case:
- Ask whoever designed the menu for the original file, exported as a PDF with selectable text.
- Or type it out. A plain Word document with dish names, descriptions, prices and dietary marks is perfectly good, and often reads better than a designed menu.
- Better still, put the text menu on your website as a page. Search engines and diners on slow phones benefit too.
Keep separate menus as separate files with plain names: menu-lunch.pdf, menu-dinner.pdf, menu-bar.pdf. And keep the names stable. Uploading a corrected file with the same name replaces the old one, which is exactly what you want when prices change; a new name such as menu-dinner-october.pdf would sit beside the old menu and answer with both. The full routine for documents is in how to train a chatbot on your PDFs and documents.
What can't a restaurant chatbot do?
It cannot take a table booking or an order, and it cannot see what is happening in your restaurant tonight. TailGPT is a chat window on your website that answers from your pages. It is not connected to a reservations system, a point-of-sale or a delivery app, it does not collect a diner's name and phone number, and it does not pass a conversation to a member of staff.
| A diner asks | What the chatbot can do |
|---|---|
| "Book a table for four at 8 tonight" | Tell them how to book, using what your page says: the reservations page by name, or the phone number written on it |
| "Is there a table free right now?" | Nothing live. If your page says "for same-day tables, please call", it can say that |
| "I want to order the biryani for delivery" | Say how ordering works, as your delivery page describes it |
| "Where is my order?" | Nothing. Your website does not hold order status |
| "Is the mutton sold out today?" | Nothing. Daily stock is not on a page |
| "Can I get 20% off for my group?" | Only repeat an offer that is published. It cannot agree a discount |
One detail matters here. Answers appear as plain text, without clickable links, so when the chatbot points a diner to your booking page it names the page in words. It can give a phone or WhatsApp number only if that number is written on a page it has read. So put the booking route in a sentence on your contact or reservations page: "To book a table, call the number below or use the Book a Table button on this page. We hold tables for fifteen minutes." Make sure that page is ticked, or the commonest question of all gets an honest but useless answer.
How do you keep menus and prices current?
Update the website or the menu file first, then let TailGPT read it again. A chatbot quotes what it last read, and TailGPT has no automatic refresh: the website is read again when you ask, from a monthly allowance. As of 4 October 2026, Starter includes 2 re-reads a month and Growth includes 8.
Restaurants change often, so plan for it:
- Keep fast-moving prices in a document. Replacing a menu PDF with a file of the same name does not use a re-read at all. A restaurant that revises prices every few months can do it without touching the allowance.
- Publish seasonal timings a day early. Diwali hours, a Christmas Eve set menu, Ramzan evenings, a monsoon closure: change the page, read the site, and the chatbot is right from the first diner who asks.
- Put dates on time-bound offers. "Weekday lunch buffet, Monday to Friday, until 31 October" is better than a bare offer, because a stale answer with a date on it is at least visibly stale. Even better, remove the offer when it ends.
- Untick last season's pages. An old festive menu page keeps answering until it is unticked and the site is read again.
The same habits for every kind of site are in how to keep your chatbot's answers up to date.
How many questions will a restaurant chatbot use?
Fewer than you might think for a single outlet, and more on weekends. Questions are one pool for the whole account. Every question counts, including one your pages do not cover. Only three kinds of refusal are free: attempts to give it orders, requests to write code, and too many questions too fast.
| As of 4 October 2026 | Starter | Growth |
|---|---|---|
| Price | ₹4,000 a month | ₹8,000 a month |
| Websites live at once | 1 | 3 |
| Chatbots | 1 | 3 |
| Questions a month | 4,000 | 10,000 |
| Website re-reads a month | 2 | 8 |
A single restaurant usually sits comfortably on Starter; a group with three brands on three websites is what Growth is for. TailGPT emails you at 80% and 100% of the month's questions, and at the limit the chat bubble disappears until the month resets, you upgrade, or you buy a top-up of questions. Enterprise is custom. The full table is on the pricing page.
What should you test before it goes on your website?
Ask it the questions your staff answer every day, and check each answer against your own pages. The free trial needs no card: one website, up to 500 pages read, one chatbot and 20 test questions in the dashboard. It cannot upload documents or go on the website, so a PDF menu has to wait for a plan, but the pages can all be tested.
- Are you open on Monday, and what time does the kitchen close?
- Is there parking?
- Do you have Jain options?
- Is the paneer tikka nut-free?
- Can I book a table for eight tonight?
- Do you deliver to (an area you do not serve)?
- Can we host a birthday for 25 people?
Each answer should come from your page, say plainly that the website does not cover it, or name your booking route. After launch, read the conversations in Show chats once a week and search for words like "Jain", "parking" and "book". The dashboard also lists the questions your pages could not answer, with how often each was asked. Each one is a sentence your website is missing.
