AI Auto-Respond Setup

Updated September 5, 2026

ChattyPie's AI auto-respond feature uses retrieval-augmented generation (RAG) to automatically answer customer questions based on your knowledge base articles, past conversations, and custom data sources.

How It Works

  1. A customer sends a message
  2. The AI searches your knowledge base and conversation history for relevant content
  3. If a confident answer is found, the AI drafts a response
  4. Depending on your response mode, the answer is sent automatically or queued for agent review

Enabling AI Auto-Respond

  1. Go to Settings > AI > Auto-Respond
  2. Toggle Enable Auto-Respond
  3. Configure the settings below
  4. Click Save

Data Sources

The AI draws answers from these sources:

  • Knowledge base articles — Published articles in your help center
  • Past conversations — Resolved conversations (anonymized)
  • Custom data — Upload additional documents (PDF, TXT, MD) in Settings > AI > Data Sources

The AI automatically re-indexes when you publish or update articles.

Confidence Threshold

The confidence threshold controls how sure the AI must be before responding:

Threshold Behavior
High (0.85+) Only responds when very confident. Fewer responses but higher accuracy.
Medium (0.7) Balanced. Good starting point for most teams.
Low (0.5) Responds more often but may give less accurate answers.

Start with Medium and adjust based on customer feedback.

Response Modes

Mode Description
Auto-send AI sends the response directly to the customer
Suggest AI drafts a response for the agent to review and send
After-hours only Auto-send outside business hours, suggest during hours

Handoff to Human

The AI automatically hands off to a human agent when:

  • Confidence is below the threshold
  • The customer explicitly asks for a human ("talk to a person", "speak to agent")
  • The conversation has gone back and forth 3+ times without resolution
  • The customer expresses frustration (detected via sentiment analysis)

When handoff occurs, the AI creates an internal note summarizing the conversation so the agent has context.

Monitoring Performance

Track AI performance in Reports > AI:

  • Auto-resolved rate — Percentage of conversations fully handled by AI
  • Handoff rate — How often AI transfers to a human
  • Customer satisfaction — CSAT scores for AI-handled conversations
  • Accuracy — Based on agent corrections to AI suggestions

Did this answer your question?