AI Filters
Intelligently filter feedback using AI-powered natural language prompts before it reaches your destination
Overview
The AI Filter feature allows you to intelligently filter feedback before it reaches your destination. Using natural language prompts, you can instruct an AI model to analyze feedback and decide whether it should be forwarded or blocked.
Primary engine
Under Primary engine, choose how the filter decides:
- Jev — Fast, and free while it is experimental. Jev returns a probability rather than an explanation. A Jev run does not use an AI credit, but credits must still be available. You can turn on Encatch AI fallback. A below-cutoff or failed Jev evaluation then uses 1 AI credit.
- Encatch AI — Explains each decision and uses 1 AI credit per evaluation.
How It Works
Configure Your Prompt
Write a clear instruction that tells the AI what feedback to forward or block. Be specific about your criteria.
Select Context Data
Choose which feedback data fields to include in the analysis:
- Questions and answers
- Device information
- Geographic information
- User information
Test Your Filter
Use the test feature to see how your filter responds to sample feedback. Review the decision, reason, and confidence level.
Save & Enable
Once satisfied with test results, save your configuration and enable the filter to start filtering feedback automatically.
Key Features
- Natural Language Prompts: Write instructions in plain English - no complex syntax required
- Context-Aware Analysis: Include relevant feedback data for better filtering decisions
- Default Forward on Error: Configure behavior when AI processing fails
- Test Before Deploy: Preview filter decisions with sample feedback data
Writing Effective Prompts
To create effective prompts, follow these guidelines:
- Be Specific: Clearly state what feedback should be forwarded or blocked
- Define Criteria: Specify the criteria for your decision (sentiment, topic, quality, etc.)
- Use Examples: Include examples of feedback that should be forwarded/blocked (optional but helpful)
- Reference Context: Mention the available data fields in your prompt when relevant
A well-structured prompt typically includes:
- Clear instruction on what the AI should do
- Specific criteria for forwarding/blocking
- Examples or scenarios (optional)
Example Prompts
Example 1: Forward Only Positive Feedback
Use this prompt to forward only positive or neutral feedback:
Analyze the feedback and determine if it should be forwarded to the destination.
Forward the feedback ONLY if:
- The overall sentiment is positive or neutral
- The feedback contains constructive suggestions or praise
- The rating (if available) is 3 stars or higher
Do NOT forward if:
- The feedback is primarily negative or critical
- The feedback contains complaints without constructive elements
- The rating is below 3 stars
Respond with a clear decision: FORWARD or DO NOT FORWARD, along with a brief reason.Example 2: Filter by Topic Relevance
Use this prompt to filter feedback based on topic relevance:
Review the feedback and determine if it's relevant to our product features.
Forward the feedback if it discusses:
- Product features, functionality, or usability
- User experience improvements
- Feature requests or bug reports
Do NOT forward if it discusses:
- Pricing or billing questions
- Account management issues
- Spam or irrelevant content
Provide your decision with a brief explanation.Example 3: Quality-Based Filtering
Use this prompt to filter based on feedback quality:
Evaluate the feedback quality and completeness.
Forward the feedback if:
- It contains substantive content (not just "good" or "bad")
- It includes specific details or examples
- It provides actionable insights
Do NOT forward if:
- The feedback is too brief or lacks substance
- It contains only generic responses
- It appears to be spam or automated
Explain your decision based on the quality and completeness of the feedback.Context Data Fields
When configuring your filter, you can include these data fields in the analysis:
- Questions: The questionnaire structure and questions asked
- Answers: The user's responses to questions
- Device Info: Device type, OS, browser, timezone, theme, etc.
- Geo Info: Location data (country, region, city, postcode)
- User Info: User-related information (if available)
Select only relevant fields to optimize performance and reduce token usage. At least one field must be selected.
Settings
Default Forward on AI Filter Error
When enabled, feedback will be automatically forwarded if AI filter processing fails. This ensures no feedback is lost due to technical issues.
Recommended: Enable this setting to prevent data loss, unless you have strict requirements to block feedback when AI processing fails.
Why AI Filters Fail
AI filters may fail to process feedback for several reasons. Understanding these scenarios helps you configure your settings appropriately and troubleshoot issues effectively.
Your AI filter requires available AI Credits to process feedback. Jev does not consume a credit, but it still requires credits to be available. If your account has exhausted all AI Credits, the filter will fail to execute.
Solution: Buy an AI Credits add-on from the billing screen (Team and Growth; add-on credits expire at the end of the cycle), upgrade from Free, or wait for your monthly allowance to reset. Check your billing dashboard to monitor credit usage.
The combined size of your prompt and selected context data fields may exceed the AI model's token limit. This happens when:
- Your prompt is extremely long
- You've selected too many context fields
- The feedback data itself is very large
Solution: Simplify your prompt, reduce the number of context fields selected, or break down complex filtering logic into multiple simpler filters.
Occasionally, internal system errors or processing issues may cause the AI filter to fail. These are temporary issues that typically resolve automatically.
Solution: Retry the operation after a few moments. If the issue persists, check system status or contact support. Enable "Default Forward on AI Filter Error" to ensure feedback isn't lost during these incidents.
The underlying AI service provider may experience downtime or service interruptions, preventing the filter from processing feedback.
Solution: Enable "Default Forward on AI Filter Error" to automatically forward feedback when the AI service is unavailable. Monitor service status and retry once the service is restored.
To prevent data loss when AI filters fail, always enable "Default Forward on AI Filter Error" unless you have strict requirements to block feedback during failures.
Testing Your Filter
To test your AI filter:
- Click "Test AI Filter" (always enabled by default)
- Optionally check "Save Configuration" to save your changes
- Optionally check "Enable AI Filter" to activate the filter
- Click "Execute Process" to test with sample feedback
The test results will show:
- Decision: Whether the feedback would be FORWARDED or NOT FORWARDED
- Reason: The AI's explanation for the decision
- Confidence: How confident the AI is in its decision
- Token Usage: How many tokens were consumed
An Encatch AI test uses 1 AI credit. A Jev test does not, but credits must still be available.
Tips for Success
- Start Simple: Begin with a basic prompt and refine based on test results
- Test Thoroughly: Test with various types of feedback to ensure it works as expected
- Be Specific: Vague prompts lead to inconsistent results - be as specific as possible
- Use Context Wisely: Only include context fields that are relevant to your filtering logic
- Iterate: Adjust your prompt based on test results and real-world performance
Common Use Cases
- Quality Control: Filter out low-quality or spam feedback
- Topic Filtering: Only forward feedback relevant to specific topics
- Sentiment Analysis: Forward only positive or constructive feedback
- Compliance: Filter out inappropriate or non-compliant content
Troubleshooting
Make your criteria less strict or add more conditions for forwarding. Review your prompt and ensure it's not blocking valid feedback.
Add more specific criteria or tighten your requirements. Be more explicit about what should be blocked.
Make your prompt more explicit with clear examples and criteria. Test with multiple feedback samples to identify patterns.
Reduce the number of context fields selected or simplify your prompt. Only include fields that are essential for your filtering logic.
The Execute button is disabled when the prompt is empty. Enter a prompt to enable execution.
Was this page helpful?