How to Build a Chatbot Knowledge Base: What to Include
You've set up a bot, but the answers feel vague? The problem is usually not the bot but the information you gave it. From FAQs to return policies, from site pages to what must never go to the bot, here is how to build a knowledge base step by step.
How well an AI bot answers depends largely on the information you give it. A chatbot knowledge base is exactly that: the texts about your business that the bot looks at when it answers customer questions. Below we go through what to put in it, what to leave out and how to write the texts, in that order.
What is a chatbot knowledge base, and why does it matter so much?
An AI bot can use language very well. But it can't know your delivery time, your return terms or which districts you serve. You tell it.
The knowledge base works like the bot's memory of your business. When a customer asks "Are you open on Sundays?", the bot looks at these texts and builds its answer from what it finds there. If the text is incomplete or contradicts itself, the answer will too.
We see one situation often. A business sets up the bot, isn't happy with the answers and looks for the problem in the bot. Most of the time, what's missing is the knowledge base.
The knowledge base doesn't update itself When a price changes, you open a new branch or your working hours change, you have to add that information to the knowledge base.
What content should go into the knowledge base?
We recommend four main sources. You don't need to prepare all of them in full on day one; start with what's asked most and add the rest over time.
1. Frequently asked questions
This is the most productive place to start. Look through the last few weeks of WhatsApp, Instagram and phone conversations and you'll see the same questions coming up again and again. Write them as question-and-answer pairs.
- Working hours, address and directions
- Delivery time and the regions you ship to
- Payment options: cash on delivery, installments, bank transfer
- How to book an appointment or a reservation
- How invoices and warranty documents are sent
2. Product and service information
What does the product do, who is it for, what are its size and material, how is it used? If you run a service business, write down what the packages include, how long they last and what the prerequisites are. When a customer asks "Is this cream suitable for sensitive skin?", the bot will rely on this text.
Information that changes often, like price and stock, is a separate matter. If you keep it in a text you write by hand, it goes stale quickly. In Chatopya, if your store is connected, stock and price are updated from the connected store at regular intervals. In texts you prepare by hand, it's safer to use a sentence that points the customer to the current source than to write the price as a number.
3. Policy texts
Return, exchange, warranty, cancellation, delivery and privacy terms belong in this group. The bot has to pass these on to the customer briefly and correctly, so use the current version of the official text on your site.
One method that works: add a plain-language summary next to the legal text. For example, "We accept returns within 14 days of delivery for products with the tags still attached." The bot can build its answer from both the official text and this clear summary.
4. Site pages
About us, FAQ, shipping and returns pages, service pages. In Chatopya's Bot Center, the knowledge base covers both your business information and your site pages, so you don't have to rewrite what's already written on your site.
The only condition is that the pages are up to date. If the page for a campaign that ended three years ago is still live, the bot can read it too.
What information should you not give the bot?
The purpose of a knowledge base is to answer customers. Anything you put there could be said to a customer one day, so choose the content with that in mind.
- Customers' personal data: name lists, phone numbers, addresses, order records. Personal information belongs in its own systems, not the knowledge base.
- Special categories of data: sensitive data such as health information. Businesses like clinics, pharmacies and beauty centers need to be especially careful here.
- Internal documents: cost tables, supplier prices, profit margins, staff information, internal correspondence.
- Access credentials: passwords, API keys, panel login details.
- Old or unconfirmed information: campaigns that have ended, new things you've only talked about as "maybe next month".
- Contradicting texts: two documents that give two different return periods will make the bot answer wrongly.
Don't skip the KVKK side KVKK is Turkey's personal data protection law. The Personal Data Protection Authority has published a question-and-answer guide on generative AI and the protection of personal data. Keeping personal data out of the knowledge base entirely is the easiest way to reduce the risk from the start. If you're unsure, ask your legal adviser.
How should you write the texts for a bot?
A bot works much better with organized, clear text. Keep these rules in mind as you write:
- Write each topic in one place. Don't let the delivery time appear in five separate documents; keep it in one.
- Write short paragraphs that make sense on their own. References like "as we noted above" mean nothing to a bot.
- State figures and conditions clearly. Write "1-3 business days" instead of "shortly".
- Mention the exceptions. Which orders get free shipping, and which don't?
- Use the customer's words. A customer asks "when will my parcel arrive", not "shipment process".
- Write down what the bot can't do as well. For example, "For a custom price quote, our agent will get back to you."
| Topic | Wording that is hard for a bot | Wording that suits a bot |
|---|---|---|
| Delivery | Orders are shipped as soon as possible. | Orders placed by 2:00 pm on weekdays are shipped the same day; delivery takes 1-3 business days. |
| Returns | See the contract for return terms. | You can return a product with its tags still attached within 14 days of delivery. Underwear cannot be returned. |
| Working hours | You can reach us during our working hours. | We are open 9:00 am-6:00 pm on weekdays and 10:00 am-3:00 pm on Saturdays. We are closed on Sundays. |
The figures in the table are examples; write your own terms. The idea here is that whatever the customer asks, the bot should be able to find the answer in a single paragraph.
Let's build your knowledge base together. Setup and training are free, and our Turkish-speaking support team is with you on the first texts as well.
Book a free demoWhat should the bot do when the knowledge base has no answer?
A good bot isn't one that knows everything. It's one that knows what it doesn't know.
AI models have a known weakness: where they aren't sure, they can produce an answer that sounds reasonable but is wrong. A bot that tells a customer the wrong return period costs more than a bot that gives no answer at all.
In Chatopya, when a question comes in that has no match in the knowledge base, the bot doesn't invent an answer; it hands the conversation to an agent. Because the handover includes the chat history, the agent continues the conversation without asking the customer the same things again. You set the working hours and the handover rule in the Bot Center. We describe this structure in detail on our WhatsApp customer service and AI chatbot page.
The questions that land with an agent are also a to-do list. They show you what is missing from the knowledge base. If you look at them once a week and add the ones with clear answers, the topics the bot can answer grow step by step.
How do you test before going live?
In the Bot Center you can test flows before putting them live. We suggest this checklist:
- Ask the 15-20 most frequent questions one by one and compare the answers with the official text.
- Write like customers do: try questions with typos and abbreviations ("when delivery arrive").
- Ask a question with no match in the knowledge base and check whether the conversation passes to an agent.
- On policy questions such as returns and warranty, check that the bot passes on the terms in full.
- If you have foreign customers, try the same questions in another language as well.
The bot doesn't work only on WhatsApp. The Web Chat window you add to your site answers from the same knowledge base, so testing on both channels makes your job easier. If you're setting up the bot for the first time, our WhatsApp chatbot setup guide shows the order step by step.
How often should you update the knowledge base?
In short: the day the information changes. Holiday and New Year working hours, new products, ended campaigns, a changed courier company... All of these should go into the knowledge base that day.
Beyond that, read all the texts from start to finish once a month. That's where we catch two sentences that contradict each other or a condition that no longer applies. A chatbot knowledge base is not a document you write once and leave; it's living content that needs regular care.
Factor in the cost too In Chatopya, AI Bot usage runs on credits; Pro includes 500 AI credits a month and Business 1,000, and extra credit packs are available. On the WhatsApp side, under Meta's pricing in force since 1 October 2026, the first 1,000 service messages per number per month in the 24-hour window after the customer writes are free, then charged. Details are in our WhatsApp Business API pricing guide.
Frequently asked questions
What content should I add to a chatbot knowledge base?
Frequently asked questions, product and service information, policy texts such as returns, delivery and warranty, and the current pages of your site. We suggest starting with the questions asked most.
How long should the knowledge base be?
What matters is not length but coverage and consistency. A set of texts that answers the questions your customers ask most and doesn't contradict itself gives better results than long but scattered documents.
What does the bot answer to a question that isn't in the knowledge base?
In Chatopya, when a question has no match in the knowledge base, the bot doesn't invent an answer; it hands the conversation to an agent with the chat history. You set the hours and the situations for handover with rules.
Can I put customer information into the knowledge base?
We don't recommend it. Content in the knowledge base is used in answers given to customers. Personal data, order records and internal documents should therefore stay out.
Do I pay separately for the bot's answers?
In Chatopya, AI Bot usage runs on credits. On WhatsApp, Meta keeps the first 1,000 service messages per number per month free in the 24-hour window and charges after that. For packages, see our pricing page.
Related articles
Help your bot describe your business correctly
Let's set up your knowledge base and handover rules together. Setup and training are free, and our Turkish-speaking support team is with you at every step.