It is tempting to introduce an AI assistant by giving it everything: every brochure, every page, every old proposal. More material can feel like more preparation. In practice, a small collection of clear, current answers is often a better starting point.

Your first knowledge base should help the assistant answer the questions customers actually ask, using information your business is comfortable sharing. It should also be easy for your team to maintain.

Begin with real questions

Look at recent support emails and conversations. Write down the recurring questions in the customer's own language. A useful first set might cover what you offer, how someone gets started, where you operate, your published prices, and your cancellation or return policy.

Do not start by writing a company encyclopedia. Start with the ten questions that interrupt your team most often. A short answer to a specific question is easier to verify than a long document trying to cover every possible situation.

Separate facts from behavior

“Our standard delivery takes three to five business days” is a fact about your business. “Keep replies warm and concise” is guidance about behavior. Store the fact in Knowledge and the behavior guidance in your assistant settings.

That separation helps your team know where to make a change. If delivery times change, update the policy source. If the replies feel too formal, adjust the tone. Avoid hiding essential business facts in a long personality prompt.

Write answers that can stand alone

A knowledge source may be retrieved in smaller passages. Make each passage understandable without requiring the reader to guess what “it,” “that plan,” or “the process above” refers to.

  • Name the product or service the answer applies to.
  • State the conditions and exceptions close to the main answer.
  • Use a currency whenever you mention a price.
  • Say whether a time estimate uses business days or calendar days.
  • Explain when a teammate needs to confirm an individual case.

For example, “Returns are accepted” leaves too much open. A more useful source states the relevant time window, item condition, exclusions, and how the customer should start a return.

Leave private material out

Only add material approved for customer-facing answers. Internal passwords, supplier negotiation notes, employee records, and customer-specific private information do not belong in a general support knowledge base.

When a question requires private account data or a personal judgment, describe the next step the assistant should offer: ask the customer to contact the appropriate team, or hand the conversation to a person. Do not create a public fact sheet that accidentally contains private cases.

Test the edges, not only the obvious answer

Once a source is saved, test it with a normal question and a question that falls just outside the policy. If you publish a standard delivery time, ask whether the business can guarantee delivery by tomorrow. If a service is available in one location, ask about a location you do not serve.

The useful result is not always an answer. Sometimes it is a clear statement that the assistant does not have enough approved information, followed by a human handoff. In Botinto, knowledge-only mode is a sensible starting point while you learn where the gaps are.

Give every source an owner

Decide who updates each policy and when it should be checked again. Review sources after a product launch, a pricing change, or a change to your support process. Delete superseded information instead of leaving contradictory copies for the assistant to choose between.

Botinto keeps source versions and lets you edit and re-index knowledge. Use a descriptive title, such as “Standard delivery — domestic orders,” so a source reference is meaningful when a teammate reviews a reply.

A good knowledge base is a maintained part of your customer experience. Begin small, review the questions that reach your inbox, and expand when real conversations reveal what is missing.

Ready to put your first sources together? Explore Botinto’s AI website chatbot, or follow the setup steps. For customer email, explore the Botinto shared email inbox.

← Back to the journal