> ## Documentation Index
> Fetch the complete documentation index at: https://docs.ticketcord.com/llms.txt
> Use this file to discover all available pages before exploring further.
# Feedback & Learning
> Help your bot learn and improve over time
The feedback system lets users rate AI suggestions, and optionally allows your bot to learn and improve based on those ratings.
## How Feedback Works
When your bot suggests an answer, users see vote buttons:
```
💡 Suggestion
Here's how to reset your password...
[👍 Helpful] [👎 Not Helpful]
```
When users click:
* **👍 Helpful** - Records positive feedback
* **👎 Not Helpful** - Records negative feedback
This data helps you understand which content works well and which needs improvement.
## Enabling Feedback
1. Go to **AI Features** → **Knowledge Base** → **Feedback**
2. Toggle **Enable Feedback** on
3. Save your settings
Feedback is enabled by default for new Knowledge Base setups.
## Feedback Settings
### Staff Only Voting
When enabled, only staff members can vote on suggestions.
**When to use:**
* Staff opinions are more valuable
* Prevent spam voting from users
* Train the system based on expert judgment
**When to disable:**
* All feedback is valuable
* Users know when answers help them
* You want broader feedback data
### Self-Learning
When enabled, the bot automatically adjusts content ranking based on votes.
**How it works:**
1. Content gets votes over time
2. After reaching the minimum votes threshold (default: 3)
3. High-rated content gets boosted (shown more often)
4. Low-rated content gets penalized (shown less often)
**Settings you can adjust:**
| Setting | Default | Description |
| ---------------------- | ------- | ------------------------------------ |
| Min Votes for Learning | 3 | Votes needed before learning applies |
| Good Content Threshold | 70% | Success rate to trigger boost |
| Poor Content Threshold | 40% | Success rate to trigger penalty |
| Boost Percentage | 15% | How much to boost good content |
| Penalty Percentage | 15% | How much to penalize poor content |
Start with defaults. Adjust if you notice the system learning too aggressively or too slowly.
### AI Disclaimer
When enabled, suggestions include a note that the response was AI-generated:
> 🤖 *This response was generated by AI and may not be 100% accurate.*
**When to enable:**
* Transparency is important to your users
* Users should verify important information
* You're in an industry that requires disclaimers
**When to disable:**
* Cleaner appearance
* Users don't need the reminder
* You've validated all your sources
## Auto-Escalation
When the AI doesn't have a confident answer, it can automatically alert your team.
### How It Works
1. User asks a question
2. AI searches Knowledge Base
3. If confidence is too low, AI triggers escalation
4. Staff gets notified
5. User sees a message that human help is coming
### Settings
**Auto-Escalate on Low Confidence**
Toggle this on to enable escalation when AI can't find good answers.
**Ping Staff on Escalation**
When enabled, the bot will ping your staff role when escalating.
Make sure you have a staff role configured in your server settings.
**Escalation Batch Window**
Wait time before sending escalation notification (in seconds):
* **0** - Immediate notification
* **30** - Wait 30 seconds (default, allows user to provide more context)
* **Up to 300** - Wait up to 5 minutes
### Human Assistance Detection
The bot can detect when users specifically ask for human help:
**Phrases that trigger immediate escalation:**
* "I want to talk to a human"
* "Can I speak to a real person?"
* "Human please"
* "Real support"
* Similar variations
When detected, the bot immediately escalates without trying to answer.
**Confidence Threshold**
How confident the AI must be that the user wants human help (default: 80%).
* Higher = fewer false positives, might miss some requests
* Lower = catches more requests, might escalate unnecessarily
## Viewing Feedback Data
In the dashboard, you can see:
### Overview Stats
* Total suggestions given
* Total votes received
* Thumbs up percentage
* Thumbs down percentage
### Source Performance
See how each source performs:
* Which sources get the most positive feedback
* Which sources have problems
* Suggested actions (deprecate, update, etc.)
### Problem Areas
Identify content that needs attention:
* Low success rate chunks
* Common feedback reasons (incorrect, outdated, not relevant)
* Suggestions for improvement
## Improving Based on Feedback
### Content With Low Ratings
If content consistently gets thumbs down:
1. **Check the source** - Is the documentation accurate and up-to-date?
2. **Read the feedback** - What are users saying is wrong?
3. **Update or remove** - Fix the content or delete the source
### Content With High Ratings
If content consistently gets thumbs up:
1. **Add similar content** - What makes this content work well?
2. **Use as a template** - Follow the same format for new content
3. **Consider priority boost** - Increase the source priority
### Persistent Issues
If the same questions keep getting bad ratings:
1. **Check if answer exists** - Maybe you need new documentation
2. **Improve the content** - Make it clearer or more complete
3. **Consider categories** - Maybe the wrong source is being used
## Best Practices
### Review Feedback Regularly
Check your feedback stats weekly to catch issues early:
* Sudden drops in success rate
* New problem areas
* Sources that need updates
### Enable Self-Learning Carefully
Self-learning works best with:
* High-quality documentation
* Enough feedback data (at least 50+ votes total)
* Regular monitoring
Start with it disabled, enable once you're confident in your sources.
### Use Escalation for Complex Topics
Some questions shouldn't have AI answers:
* Billing disputes
* Account security issues
* Complex technical problems
Set up category restrictions so AI doesn't answer sensitive tickets, and enable escalation for those categories.
Get instant support via private DM