How to Build a Knowledge Base for Your WhatsApp AI Chatbot
Short answer
To build a knowledge base for an AI chatbot, gather your real customer questions, write one clear entry per topic with the exact policy or answer, add product details and edge cases, test the chatbot with awkward questions, and review it monthly. Accurate, specific entries matter more than volume.
An AI agent is only as good as what it can look up. Most disappointing chatbots are not model problems, they are content problems. This guide shows you how to build a knowledge base that keeps answers accurate and on-brand.
Start from real questions
Do not write what you think customers ask. Export or scan your last two to three months of conversations and list the questions that repeat. Group them: delivery, pricing, returns, opening hours, how-to, account issues. Those groups become your first entries.
What each entry should contain
- A clear title phrased as the customer would ask it.
- The exact answer, including numbers, timeframes and conditions.
- Exceptions. “Refunds take 5 working days, except for international orders.”
- What to do if it does not apply: who to escalate to.
- A photo or link where visuals help, such as a size guide.
Structure that helps retrieval
- One topic per entry. Do not bury five policies in one long document.
- Use plain wording and the same terms customers use.
- Put the answer first, then the detail.
- Avoid contradictions between entries. If two entries disagree, the AI may pick either.
Define what the agent should not answer
Decide up front which topics always go to a human: legal complaints, refunds above a threshold, medical or financial advice, anything involving personal data changes. Configure escalation triggers for these. In Resolvenyx you set keywords and conditions that hand the conversation to an agent with an escalation message. See AI customer service agent.
Test before you go live
- Ask the ten most common questions and check each answer against your source.
- Ask them badly: typos, slang, two questions in one message, another language.
- Ask things that are not in the knowledge base and confirm it escalates rather than guesses.
- Try to make it break policy: “Can you give me a discount?”
Use the built-in test chat before you attach the agent to a live inbox.
Keep it current
- Review every escalated conversation. Each one shows a gap.
- Update entries whenever prices, policies or hours change.
- Track containment rate and fix the lowest-performing topics first.
- Assign one owner. A knowledge base with no owner decays.
Frequently asked questions
How big should an AI chatbot knowledge base be?
Start with your top 20 to 30 real questions and grow from escalations. Coverage of the common questions beats a large, uneven document dump.
Can the same knowledge base serve several channels?
Yes. In an omnichannel setup the same agent and knowledge base can be attached to more than one inbox. See omnichannel customer support.
What if the AI does not know an answer?
Configure it to escalate to a human with a clear handover message rather than guess.
See it working on your own conversations
Resolvenyx puts WhatsApp, email, Instagram, Facebook and web chat in one inbox, with tickets and AI agents built in. 7-day free trial, no credit card required.
Written by the Resolvenyx team
We build a WhatsApp and omnichannel customer support platform, and write these guides from what we see support teams run into. Product details are checked against the current app. About Resolvenyx