How to suppress Request creation

ClearFeed uses AI-powered filtering to automatically suppress request creation for certain types of messages that don't require support attention. This helps reduce noise and ensures your team focuses on messages that actually need responses.

How AI Filtering Works

ClearFeed uses GPT-4.1-mini to analyze incoming messages and determine if they should be ignored based on predefined categories. Messages that fall entirely within these categories are automatically filtered out and don't create requests.

Default AI Filtering Categories

By default, ClearFeed ignores messages that fall into these categories:

• Greetings: Simple greetings and wishes (e.g., "Good morning", "Happy birthday")

• Acknowledgments: Messages showing understanding or confirmation (e.g., "Ok thanks", "Got it", "Makes sense")

• Appreciation: Thank you messages and praise (e.g., "Great work", "Thanks for helping out")

• Resolved: Messages indicating an issue is solved (e.g., "It's working now", "Got the answer thanks")

Customizing AI Filtering

You can customize the AI filtering behavior by modifying the prompt that ClearFeed uses to classify messages. This allows you to:

• Add new categories of messages to ignore

• Modify existing categories to better match your team's communication patterns

• Adjust the filtering criteria to be more or less strict

Current AI Filtering Prompt

Below is the current prompt used for AI filtering. You can copy and modify this prompt to customize the filtering behavior:

{
  "task": "Given a user's query, check if the query falls under the categories defined below. Our task is to return a JSON - where the `is_ig` attribute is returned as True if the entire query belongs to one OR more of the categories below. Else, if ANY part of the query DOES NOT belong to one of the categories below - then return `is_ig` as False.",
  
  "categories": {
    "greeting": {
      "definition": "User queries containing ONLY greetings and wishes",
      "examples": ["Good morning", "Happy birthday", "Good luck"]
    },
    "acknowledgment": {
      "definition": "User queries which show acknowledgment",
      "examples": ["ok thanks", "Hmm got it", "makes sense", "cool", "no issues", "thanks for letting me know", "thanks for a prompt reply"]
    },
    "appreciation": {
      "definition": "queries including appreciation expressions",
      "examples": ["Great work", "thanks for helping out", "Man awesome guys!", "woot woot", "thank you will"]
    },
    "resolved": {
      "definition": "Queries indicating resolution/completion of a task or an issue. The issue should not be under investigation. There should be no further action required",
      "examples": ["It is working now", "Resolved", "Got the answer thanks", "Thanks for resolving this issue", "Ticket has been closed", "This solution works"]
    }
  },
  
  "detailed_instructions": [
    "Read user queries carefully and if the entire user query fall under one OR more of the above categories, then return the flag `is_ig` as True, else return it as False.",
    "Also provide a clear but short explanation for the same in a field named `exp`. The explanation should explain which categories from above the query belongs to if is_ig is returned as True, or which part of the query did not match any of the categories above, if is_ig is returned as False.",
    "The explanation should not be more than 200 characters.",
    "The classification should be very high precision",
    "The `is_ig` flag will be used to filter out the queries which are not important. So it is critical to return it as TRUE only when Absolutely Sure that ALL of it belongs to one or more of the categories above.",
    "If you are not sure, return the is_ig flag as False.",
    "The output should be in JSON format with two keys: \"is_ig\" and \"exp\".",
    "Do not include additional tags like '\\t', '\\n' etc in the output."
  ],
  
  "examples": [
    {
      "query": "Thanks, I got the information I needed and it's working now!",
      "output": {
        "exp": "The statement \"I got the information I needed\" indicates acknowledgment and \"it's working now\" indicates resolved - hence is_ig is True",
        "is_ig": true
      }
    },
    {
      "query": "This is working on our end. Please check",
      "output": {
        "exp": "Contains request for verification which does not fit any of the categories",
        "is_ig": false
      }
    },
    {
      "query": "Good morning! Thanks for the update.",
      "output": {
        "exp": "Combines greeting and \"Thanks for the update\" indicates acknowledgment - hence is_ig is True",
        "is_ig": true
      }
    },
    {
      "query": "Hmm, interesting point!",
      "output": {
        "exp": "Doesn't fit any category",
        "is_ig": false
      }
    }
  ]
}

Testing Your Custom Prompt

You can test your modified prompt using the OpenAI playground at: https://ai-sdk.dev/playground/openai:gpt-4.1-mini

This allows you to validate that your custom categories and examples work as expected before implementing them.

Implementing Custom AI Filtering

To implement custom AI filtering for your account:

1. Copy the prompt above and modify it according to your needs

2. Add new categories or modify existing ones to match your team's communication patterns

3. Update the examples to reflect your specific use cases

4. Contact ClearFeed support via Slack or email at [email protected] with your modified prompt

Important Limitations:

• AI filtering can only suppress message types (prevent request creation)

• It cannot be used to prioritize or highlight important messages

• The filtering is designed for high precision - when in doubt, messages will create requests rather than be filtered out

Best Practices

1. Start Conservative: Begin with the default categories and gradually add more specific ones based on your team's patterns

2. Test Thoroughly: Use the OpenAI playground to test your prompt with real examples from your channels

3. Monitor Results: After implementation, monitor to ensure important messages aren't being filtered out

4. Regular Review: Periodically review and adjust your filtering criteria as communication patterns evolve

Common Use Cases

• Internal Team Channels: Filter out casual conversations, and status updates

• Customer Support: Focus on actual issues while filtering acknowledgments and thank-you messages

• Project Updates: Suppress routine status confirmations while preserving actual questions or blockers

• Announcement Channels: Filter acknowledgments to announcements