AI Chatbot Building Revision
AI Chatbot Building Revision
Deploy a 24/7 Digital Assistant
Framework: Three Pillars (Knowledge Base, Persona, Guardrails)
The core idea
Every functional chatbot rests on Three Pillars: the Knowledge Base (what it actually knows — its source information), the Persona (who it is, defined using the same CO-STAR framework from Day 1), and Guardrails (what it explicitly won't do — its boundary rules). Skipping any one of the three produces a bot that's either uninformed, generic-sounding, or prone to going off-script.
Testing matters as much as building: a solid test set spans easy questions, edge cases, and deliberately out-of-scope questions, checking that the bot handles all three gracefully rather than only the easy ones. Once built, chatbots can be published multiple ways — a direct shareable chat link, a website embed, or integration into an existing channel like WhatsApp Business.
Key terms
| Term | Definition |
|---|---|
| Three Pillars | Knowledge Base, Persona, Guardrails -- the 3-part architecture underlying any chatbot |
| Knowledge Base | The source information a chatbot draws its answers from |
| Guardrails | Explicit rules defining what a chatbot will not do or discuss |
| CO-STAR Persona | Using the CO-STAR framework specifically to define a chatbot's personality and response style |
Common mistakes to avoid
- Building a chatbot's Knowledge Base without testing it against realistic edge-case questions
- Skipping Guardrails entirely, leaving the bot free to answer anything, including off-topic requests
- Writing a Persona prompt with only 1-2 CO-STAR elements instead of the full 6
- Publishing a chatbot without ever testing an out-of-scope question against it