← All writing

Template

How to build a chatbot knowledge base from your website content

Build a website chatbot knowledge base from four sources: crawl the pages customers already read, upload useful documents, paste details that need a stable home, and type direct question-and-answer pairs for common gaps. Remove contradictions first, then re-crawl a source when its information changes.

Four groups of natural materials converge toward a shallow ceramic bowl on a workbench.

Build a website chatbot knowledge base from four sources: crawl the pages customers already read, upload useful documents, paste details that need a stable home, and type direct question-and-answer pairs for common gaps. Remove contradictions first, then re-crawl a source when its information changes.

The goal is not to give the chatbot the most material possible. It is to give it one clear, current source for each question a visitor is likely to ask.

What should go into a website chatbot knowledge base?

Start with information a customer needs before they can decide what to do next. Prices or price boundaries. What a service includes. Areas you cover. Opening hours. Delivery terms. Cancellation rules. The documents needed before an appointment.

Find the exact place where each answer lives. Chatterbox can build knowledge from a site crawl, a file upload, pasted text, or hand-typed question-and-answer pairs. Each method is useful, but adding the same fact four ways can create four places to maintain it.

Choose one primary source for each answer. If the website has a clear, current service-area page, use that page. If staff rely on a current document for a detailed policy, use the document. If the answer exists only in the owner's head, write it down before expecting the widget to use it.

When should you crawl pages or upload a document?

Crawl the public pages that already give visitors useful answers. A focused set of service, pricing, FAQ, contact, delivery, and policy pages is more useful than a large crawl filled with repeated introductions and old location pages.

Chatterbox's crawler is called ChatterboxBot and obeys robots.txt. Check that the pages you want included are available to it. A source can be re-crawled once a day, so a corrected price or policy can be picked up without rebuilding the knowledge base.

Use a file upload when the business maintains the source as a document. A rate card, care guide, course outline, or detailed terms document may be clearer there than on a short website page. Treat that file as a source you must replace when its facts change. Do not leave an old copy in place beside a new one and expect the chatbot to decide which is current.

How do pasted text and typed answers fill the gaps?

Pasted text is useful when the business has approved information that does not yet have a sensible public page. Keep it short and give each subject its own clear block. A collection of unrelated notes becomes difficult for the team to review later.

Hand-typed question-and-answer pairs are the direct route for recurring questions that staff answer every day. Write the question in the customer's words. Then write the answer the business can stand behind, including any condition that changes it.

For example, “Do you work on Sundays?” needs more than “Yes” if Sunday visits require advance booking. “Is the advance refundable?” needs the actual condition, not “contact us for details.” A useful answer is complete enough to prevent a second interpretation.

Do not create an answer for a custom quote, an exception, or a decision that genuinely needs a person. Chatterbox says it cannot find the answer rather than inventing one, then asks for contact details. For those questions, that is the correct entry in the process: no standard answer, followed by a human decision.

How do you keep chatbot answers current?

Give every important fact one owner and one source. When a price changes, update the primary page or document first. Remove the old version. Then re-crawl or replace the source.

Test the result with the same question in ordinary wording. Chatterbox shows the source passage behind each answer, so check that it points to the material you intended. Also ask an edge-case question the source does not settle. The widget should refuse and collect contact details instead of extending the rule.

Repeat this check whenever hours, coverage, pricing, or policies change. A knowledge base is not a one-time import. It is a small set of business facts that stays useful only when the source stays true.