How to Write a Chatbot Knowledge Base That Stops Wrong Answers

How to Write a Chatbot Knowledge Base That Stops Wrong Answers

Most chatbot mistakes are content problems. If the information the chatbot answers from is missing, vague or out of date, it will either refuse to help or improvise. A short, well-structured knowledge base fixes most of that.

Last reviewed: November 2026.

Start from real questions

Collect the questions customers actually ask: from your inbox, phone notes, contact-form messages and reviews. Group them. For most small businesses, twenty to forty questions cover the large majority of conversations.

Write one clear answer per question

• Lead with the answer. "Yes, we serve Surrey. Travel is included within 30 km." Then add detail.
• Be specific. Exact hours, exact prices or price ranges you publish, exact areas.
• Say what you do not do. "We do not offer emergency service after 9 pm" prevents the bot from guessing.
• Use your customers' words. If people say "quote" rather than "estimate", use both.

Add the facts the bot needs everywhere

Business name, address or service area, hours, phone, booking link, payment methods, cancellation policy and how to reach a person. Keep these in one place so updates happen once.

Decide what it must never answer

List topics that always go to a person: medical, legal or financial advice, complaints, refunds, pricing exceptions and anything about a specific customer's account. The chatbot should say it will connect them, then do it.

Keep it current

• Name one person who owns the content.
• Update it whenever prices, hours or services change.
• Review chatbot transcripts weekly for the first month, then monthly. Every unanswered or wrong answer becomes a new or corrected entry.

Test before launch

Ask the chatbot every question on your list, then the awkward ones: rude messages, questions outside your business, attempts to get it to promise discounts. Fix the content, not just the prompt.

How we build this

Our chatbots answer from content you approve, with clear handoff rules to a person. See AI chatbots and lead capture and our guide to human-in-the-loop AI.