What to put in your knowledge base
- Product and service descriptions
- Frequently asked questions
- Business hours and service areas
- Process explanations
- Shipping or delivery guidance
- Return / cancellation policies where relevant
- Qualification criteria
- Escalation and human handoff guidance
What not to put in without controls
- Frequently changing stock values unless connected to a live source
- Negotiable pricing presented as fixed
- Internal confidential information
- Unreviewed legal or medical claims
- Any promise your team cannot reliably honour
Write knowledge for retrieval, not marketing
Keep each answer direct and specific. “Delivery varies” is less useful than an approved rule that explains where delivery is available and what information is needed to confirm a date. Organize similar topics together and remove outdated duplicates.
Create an answer boundary
Every knowledge system needs a “what if we do not know?” rule. The assistant should ask a clarifying question, say the information needs confirmation, or hand off to a person. It should not fill gaps with confident guesses.
Improve the knowledge base continuously
1. Review unanswered or escalated conversations weekly.
2. Identify questions customers ask repeatedly.
3. Add or improve approved answers.
4. Remove outdated information.
5. Test with real phrasing customers use — including short messages, typos and Hinglish.
A better knowledge base improves both customer experience and AI safety
The most useful AI system is not the one with the largest prompt. It is the one with clear, current information and predictable boundaries.