AI Optimization
AI optimization analyzes your assistant’s real conversations and generates actionable recommendations to improve performance. It’s like having a consultant reviewing your assistant 24/7.
How it works
AI optimization analyzes your assistant’s conversation history and generates recommendations in five categories:
- System prompt - Rewrites your assistant’s prompt to better handle the questions visitors actually ask
- Tone adjustment - Suggests changes to the 4-axis tone settings (formality, friendliness, verbosity, creativity)
- Knowledge gaps - Identifies topics visitors ask about but the assistant can’t answer well
- FAQ generation - Creates ready-made FAQ items based on the most common questions
- Welcome message - Suggests improved welcome messages based on visitor behavior
Plan availability
| Plan | AI optimization frequency |
|---|---|
| Starter | - |
| Pro | Weekly |
| Business | Daily |
Reviewing recommendations
When AI optimization generates new recommendations:
- Go to your assistant’s AI Optimization tab
- Review each recommendation - you’ll see the current value and the proposed improvement
- Accept to apply the change, or Dismiss to skip it
- Changes take effect immediately after acceptance
Important: AI optimization never changes anything automatically. You always review and approve recommendations before any changes take effect.
Manual trigger
On supported plans, you can also trigger AI optimization manually:
- Go to the AI Optimization tab
- Click Run optimization
- Wait for the analysis to complete (usually 10-30 seconds)
- Review the new recommendations
Getting the most out of AI optimization
- Wait for data - AI optimization works best with at least 20-30 conversations to analyze
- Accept gradually - Apply one recommendation at a time and monitor the impact
- Pay attention to knowledge gaps - These are direct signals about what content to add to your knowledge base
- Review FAQ suggestions - They come directly from what your visitors are asking
How recommendations are generated
AI optimization uses an efficient language model to analyze conversation patterns:
- Which questions get good answers vs. poor ones
- Common topics and themes
- Visitor satisfaction signals (follow-up questions, conversation length)
- Gaps between what visitors ask and what’s in the knowledge base
The AI then generates specific, actionable suggestions with clear before/after comparisons.