A chatbot for your business is a chat box on your website that answers customers’ questions from material you give it: your price list, your opening hours, your policies, your FAQ. Done well, it answers the same ten questions you hear every day, at any hour, in your words, and passes everything else to a person.
It is good at repeating what you have already written down. It is bad at anything you have not written down, at judgement calls and at anything that needs a person to take responsibility. So the work of building one is mostly not technical: it is choosing what it may answer, writing that content clearly, and deciding when it must hand over. This guide walks through each of those, plus privacy and how to tell whether it is helping.
What is a chatbot that answers from your own content?
Most small businesses need a chatbot that knows their business and nothing else. One built on your own content works like a well-briefed new employee at the counter: it has read your menu, your rules and your notes, it answers from them, and when a question is outside them it says so and fetches someone.
That last part matters. A chatbot that is allowed to fill gaps from general knowledge will sooner or later invent a discount you never offered or a delivery area you do not cover. One that answers only from your content, and hands over otherwise, is less impressive in a demo and far safer on your website.
The best ones also show where an answer came from. If a reply links the FAQ entry or policy page it used, the customer can read more, and you can check the bot against the source in seconds. The FAQ and knowledge base chatbot page shows what that looks like in practice.
What is an AI chatbot for a small business good and bad at?
Write down the last twenty enquiries that came in by phone, email or at the counter, then sort them with the table below.
| Good fit for the chatbot | Keep it with a person |
|---|---|
| Opening hours, holidays, branches and directions | Complaints and anything a customer is upset about |
| What is on the menu or in the range, and what it costs | Discounts, special prices and negotiation |
| Your return, exchange, cancellation and delivery rules as written | Exceptions to those rules |
| How to book, what to bring, how to prepare for a visit | Medical, legal, tax or financial advice of any kind |
| Collecting an enquiry: name, contact, what they need | Refunds and anything that moves money, unless a person approves it |
| Pointing to the right page, form or document | Anything about one customer’s private records without checking who they are |
A restaurant is a good example. Guests ask whether you are open on Monday, whether a dish is spicy and whether you can seat a group. A restaurant chatbot can answer the first two from your menu and house rules, and pass the group booking to the manager. A shop is similar: stock, opening hours and the exchange policy, with the rest handed on, as on the retail chatbot page.
Where chatbots go wrong is usually predictable. They answer confidently from an old price list nobody updated. They try to be helpful about something they were never given. They loop, asking the customer to rephrase again and again instead of handing over. Each of these is fixed by content and rules, not by a cleverer bot.
What content should you give your chatbot?
Give it less than you think, but make every piece clear and current. A chatbot answers only as well as the content behind it. Good starting material:
- Your FAQ, rewritten so each answer stands on its own. “See above” means nothing to a chatbot.
- Prices and what they include, with the date they apply from. Say whether tax is included, so the bot never has to guess.
- Policies: returns, exchanges, cancellations, delivery areas, deposits. The written rule, not the exceptions you sometimes make.
- Practical details: opening hours for each branch, holidays, parking, what to bring, how long things take.
- A short “what we do not do” note: services you do not offer, areas you do not deliver to, questions the bot must never answer.
- Who to pass things to: which person or team handles bookings, complaints, bulk orders or anything urgent.
Leave out internal notes and anything with customers’ personal details. If you would not print it on a poster in your shop, do not give it to a public chatbot.
Then decide who keeps it up to date. Prices and opening hours change; a chatbot that quotes last season’s fee does more harm than having no chatbot. Put one person in charge of the content and give them a simple place to edit it, so a change takes minutes and does not need a developer.
When should a chatbot hand over to a person?
Decide the handover rules before you go live, and write them into the request for the bot. A good chatbot hands over:
- when the question is not covered by your content, instead of guessing;
- when the customer asks for a person, the first time they ask;
- when the customer is unhappy, or the chat goes round in circles;
- when the question touches health, law, money or safety;
- when a decision is needed: an exception, a refund, a special price, a large order.
How it hands over matters as much as when. The worst handover is “please call us”. A better one takes the customer’s name, contact details and question, tells them who will reply and roughly when, and puts the whole chat in front of the person who picks it up, so the customer is not asked the same things twice. On a doctor’s practice chatbot, for instance, opening hours, fees and visit preparation can be answered from what the clinic has written, while anything medical is left to the doctor.
What about privacy when you add a chatbot to your website?
People type all sorts of things into a chat box, including phone numbers, addresses, order details and sometimes things they should not share. Treat every chat as personal data from day one. A few practical rules:
- Ask only for what you need to answer or to follow up. A name and a phone number is often enough.
- Tell people, next to the chat box, what you keep, why and for how long, and link to your privacy policy.
- Do not let the bot read out one customer’s records (orders, bookings, fees) until it has checked who it is talking to, for example with an order number and the phone number on the order.
- Decide who in your team can read chats, and keep that list short.
- Delete old chats you no longer need, on a schedule you can explain.
Most countries have data protection rules, such as the GDPR in the European Union, the UK GDPR in the United Kingdom and the Digital Personal Data Protection Act, 2023 in India. What they require of your business depends on what you collect and how you use it, so check with your lawyer before the bot starts collecting details, especially in health, finance or education.
How do you know if your chatbot is working?
Not by how clever it sounds. In the first weeks, read as many chats as you can yourself. Then watch a few signals:
- Questions it could not answer. A list of these, grouped by topic, is the most useful thing a chatbot produces. Each group is either content to add or a topic to hand over.
- Answers that were wrong or out of date. When you find one, fix the content, not just the reply.
- Handovers. Are they reaching a person, and are they being answered? A handover nobody reads is worse than no chatbot.
- What customers say. A simple thumbs up or down on each answer shows which ones need rewriting.
- The calls and messages you still get. If the same questions still come in by phone, either the bot is not answering them well or people are not finding it.
A chatbot is never finished. The businesses that get value from one treat it like a shop window: someone looks at it every week and changes what is not working.
How to build a chatbot for your website without writing code
You can fill a ready-made chat widget with your content, hire a developer, or use an AI app builder that builds it from a description. Whichever you choose, write down the decisions above first.
With Deployflix, you describe the chatbot in your own words: what it should answer, what content it answers from, when it hands over and to whom, and what your team sees. Deployflix writes a one-page plan of its screens, the data it keeps and what the build can cost at most, and nothing is built until you approve it. See how it works. If the bot needs to reach something else, such as a phone line, an inbox or a system you already use, describe it, and the spec will say how it is done before you approve it.
Answers come from your own content, and anything the bot drafts for a customer’s case, such as a reply to a complaint, should be reviewed by a person before it goes out. Every version is kept, so when you change the content or the rules you can go back. For ideas by industry, see the hotel chatbot, which answers from a property’s room list and house rules, and the customer support chatbot, which raises a ticket when it cannot help.
Questions
Can a chatbot for my business answer questions it was not given?
It should not. Set it to answer only from your content and to say it does not know otherwise, then pass the question to a person. Questions it could not answer are a list of content to add.
Do I need a lot of content to start?
No. A clear FAQ, your prices, your opening hours and your main policies are enough for a first version. Add more as you see what customers actually ask.
Will a chatbot replace my staff?
No. It takes the repeated questions so your team has more time for the conversations that need a person: complaints, exceptions, sales and anything sensitive.
Can the chatbot take bookings or orders?
It can collect a booking or order request with the customer’s details for your team to confirm. Whether it does more depends on where your bookings and orders live, so describe that when you plan it.
Describe your app
Write what you need in your own words. Deployflix writes a one-page plan with its cost, and builds only after you approve it.
Copied. Paste it into Deployflix once you have signed in.
