Build an FAQ Library Before You Build a Chatbot

A customer asks whether you work Saturdays. One team member says yes, an old document says no, and the chatbot combines both into a confident answer nobody approved. The missing piece is not a better chat interface. It is a maintained answer.

Start with repeated questions Review the types of questions arriving through your inbox categories. Pick five that repeatedly slow people down: service area, booking steps, preparation, estimates, and what happens next. Remove private customer details before turning a conversation into a reusable example. Use the FTC’s data-protection guidance when deciding what to retain.

Create a complete FAQ entry Question: Can you give a final price from a photo? Approved answer: Photos help us assess the request. A final quote requires our stated review of scope and details. Source: the current estimating policy. Owner: the person responsible for estimates. Reviewed on: [date]. Escalate when: the request falls outside that policy. Adapt the example to your actual business. Our quote-preparation guide shows which missing details an answer should acknowledge.

Keep one rule behind similar questions “Do you work weekends?” and “Can I book Saturday?” may depend on the same policy. Reference that policy once rather than maintaining conflicting copies. Link booking answers to your confirmation process. Use Google’s version-history guidance if the source is a shared document. Name a review owner who can update it when the business changes.

Let AI help with wording, not policy Practice prompt: “Rewrite this approved answer in plain language. Preserve every condition and limitation. Do not add guarantees, prices, or availability. Flag anything unclear instead of filling it in.” Use OpenAI’s prompting documentation to refine that instruction. Route changed policy wording through the approval process.

Test the gaps deliberately Ask a teammate to find answers to ten fictional questions. Include one that is not covered. The right result is an escalation, not the nearest answer presented as fact. Record the outcome using the testing format. Document how staff search, answer, and escalate as a short operating procedure. Your first milestone is five answers people can rely on. If a searchable page solves the problem, a chatbot is optional—not the finish line.