Kerabie AI Assistant: How It Works and How to Get the Most From It

Kerabie's AI assistant is powered by Google Gemini and trained specifically on your organisation's knowledge base, not generic internet data. Here's what that means in practice.

How the AI learns your content

When you add articles to your knowledge base, Kerabie indexes them and makes them available to the AI. When a visitor asks a question, the AI:

1. Searches your knowledge base for relevant content
2. Uses Gemini to synthesise an answer from that content
3. Cites the source article in the response
4. Offers to connect the visitor to a human if uncertain

This is called Retrieval-Augmented Generation (RAG) — the AI uses your content, not generic training data, to answer.

What it can do

• Answer questions that are covered in your knowledge base
• Handle common variations of the same question ("how do I reset my password?", "forgot my password", "can't log in")
• Suggest quick replies to your agents based on conversation context
• Summarise long conversations for agents joining mid-conversation
• Detect intent and route conversations appropriately

What it can't do

• Make up information that isn't in your knowledge base
• Handle nuanced, emotional conversations well
• Replace a human for complex or escalated situations
• Access real-time data (your database, live order status) without custom integration

Configuring the AI for your team

Train it on good content

The AI is only as good as your knowledge base. Before enabling AI responses:

1. Write articles covering your top 20 support questions
2. Use the exact language your customers use
3. Keep articles focused — one topic per article works better than long mixed articles

Set the right escalation threshold

In Settings → AI Assistant, configure:

• Confidence threshold — how certain the AI needs to be before responding (default 0.7)
• Topics to never auto-respond to (e.g., billing disputes, refund requests)
• Always escalate conditions (e.g., customer says "cancel", "legal", "lawyer")

Review AI responses weekly

Check the AI conversations log. Where did it give wrong answers? Update the corresponding knowledge base articles.

Set up AI Assistant →