I’m working on a setup for auto-responding to comments, e.g., on YouTube. It doesn’t matter what platform—Instagram, LinkedIn, etc.—the goal is the same: connect incoming questions to a Custom GPT model and a knowledge base.
Here’s the challenge: how do I ensure the system differentiates clearly between “found an answer” and “didn’t find an answer”? If no answer exists, I’d like it to trigger an event (e.g., an email notification with a link to the unanswered question) so it can be manually addressed or added to a ticket system.
So it’s not important what kind of platform… what I care about is how to handle the strict if/else logic the best. That it does not hallucinate and answer stuff from its own knowledge. I mean we can set this via the system prompt sure but is there a better way to do this? Similar to how platforms like Voiceflow do it with their chatbot builder?
n8n version: 1.69.2
Database (default: SQLite): QLite
n8n EXECUTIONS_PROCESS setting (default: own, main): own, main
Running n8n via (Docker, npm, n8n cloud, desktop app): self-hosted in google cloud
The AI agent you (are to) use can be instructed to not halucinate and not make up the answers. If no answer is found do something specific. And yes, different models can produce different results. You need to experiment which one works best for you.
If you use Question and Answer Chain node as your AI assistant connected to your KB the default prompt seems to be resonable. It states
Use the following pieces of context to answer the users question.
If you don’t know the answer, just say that you don’t know, don’t try to make up an answer.