The single most reliable predictor of chatbot answer quality is the quality of its knowledge sources. A bot cannot be more specific than its source material. If your pricing page says contact us for a quote, the bot will say contact us for a quote, no matter how sophisticated the model behind it. This is why the most productive chatbot work is usually website work.
Good sources share three properties. They are specific, containing actual numbers, areas, timeframes and conditions rather than marketing abstractions. They are current, because a bot repeating last year's opening hours is worse than no bot. And they are consistent, because when two pages disagree about a fact, the bot will confidently pick one and it will not always be the right one. Auditing for contradictions across your site is unglamorous and pays off immediately.
Clerkzo trains by crawling your website, which makes your published pages the source of truth. That has a useful side effect: every improvement you make to a page improves the bot at the same time, so content work and answer quality stop being separate projects.
Related terms
Browse all 142 terms- Knowledge baseA knowledge base is the organized collection of information a business, or an AI assistant working for it, draws answers from: website pages, documents, price lists, policies, and FAQs.
- Conversation RatingA conversation rating is a simple signal, usually a thumbs up or thumbs down, that a visitor gives to a chatbot reply to indicate whether it was helpful. It is the cheapest quality feedback a chat widget can collect.
- First-contact resolutionFirst-contact resolution (FCR) is the share of customer enquiries that are completely resolved during the first interaction, without the customer needing to follow up or be called back.
- Customer satisfaction scoreCustomer satisfaction score (CSAT) is a metric that measures how satisfied a customer was with a specific interaction, collected by asking them to rate the experience immediately afterwards.