What is a chatbot knowledge base, and what should go in yours?

When a chatbot gives a crisp, confident answer about your cancellation policy or your service area, it is not being clever. It is reading. Behind every good website chatbot is a knowledge base: the approved information it draws on to answer questions. Get that right and the chatbot sounds like your most experienced team member. Get it wrong and it sounds like someone who started yesterday.

What a knowledge base actually is

It is simply the collection of content your chatbot is allowed to use. Your services, prices, policies, opening hours, service areas, the questions your team answers every week and the answers they give. A well-built chatbot finds the relevant piece first and writes its reply from it, rather than guessing from general knowledge. That is the difference between an answer and an opinion.

The eight things to include

Services and products, described the way you would explain them to a customer, not the way they appear on an invoice. Prices or price ranges, including what affects them. Policies: cancellations, deposits, guarantees, returns. Practical details: opening hours, locations, service areas, parking, access. Your process, step by step, so the chatbot can set expectations. The questions you are asked most often, with your best answers. The questions you would rather it did not answer, and who should handle them instead. And your tone: a few examples of how your team actually writes to customers.

The mistakes that make chatbots vague

The most common one is pasting in the website and stopping. Websites are written to persuade, not to answer, and they leave out the practical detail people actually ask about. The second is including too much: old brochures, internal documents, three versions of the same price list. The chatbot cannot tell which is current, so it hedges. The third is forgetting to say what the chatbot should do when it does not know, which is how confident guesses happen.

Keep it alive

A knowledge base is not a one-off upload. Prices change, a new service launches, a question starts coming up that nobody expected. Part of running a chatbot well is reviewing what people asked regularly and updating the content behind it. In our experience the first month after launch surfaces a handful of gaps; after that, a short monthly check keeps it accurate.

How we build yours

We start with a conversation, not a form. We ask how enquiries actually arrive, listen to how your team explains things, read your site and your inbox, and then build the knowledge base in your voice. You approve every line before it goes live, and the monthly report tells you what to add next.