An AI chatbot knowledge base checklist for useful customer answers

A chatbot cannot reliably explain a delivery policy that your own team describes three different ways. Before adding more instructions, make the business information clear, current and internally consistent.

This guide is published by the team behind Flas CRM. We sell one of the products in this category, so treat it as a vendor's checklist — the criteria are written to be useful whichever tool you end up choosing.

Start with the questions customers actually ask

Review a sample of recent conversations with personal details removed. Group the recurring questions into products, pricing, service area, working hours and the next step for exceptions. The first knowledge set should cover those ordinary needs rather than every document the company has ever produced.

Write each answer so a new employee could understand it. If the answer depends on a condition, state that condition beside the answer. Do not expect the chatbot to infer a policy from a collection of unrelated marketing paragraphs.

Give each fact an owner

Knowledge area	Owner's responsibility
Products	Keep specifications and names consistent
Prices	State which prices are fixed and which need a quotation
Hours	Update normal and exceptional availability
Service coverage	Define where the business can actually fulfil
Escalation	Identify who handles questions beyond the approved answer

A knowledge base without maintenance ownership becomes stale even if it is accurate at launch. Keep a review date in your internal source records and revisit the affected answers whenever the business changes an offer or policy.

Separate current facts from historical examples

An old quotation can show how the business writes, but it may contain an expired price or a customer-specific exception. Do not mix such documents into general answering material without checking what the system will use them for.

Use approved, customer-facing information wherever possible. Avoid supplying confidential records merely to make the knowledge set larger. More text is not automatically better guidance.

Test the configured behaviour

Flas describes AI answers from business knowledge and human handoff on its feature page. Configure a small set, then ask normal questions, incomplete questions and questions outside the scope. Check the actual answers rather than assuming the presence of a source document guarantees a correct reply.

A good fallback might ask for a missing detail or hand the conversation to a person. It should not invent an answer to avoid admitting uncertainty.

Improve the source before lengthening the prompt

When an answer is wrong, find the cause. The source may be outdated, contradictory or missing the condition that matters. Repairing that source is usually more understandable than adding a long instruction telling the chatbot never to make the same mistake again.

Use the pricing answer guide for commercially sensitive facts and the test plan before broadening the rollout. Start with a knowledge set your team can maintain and defend.