> ## Documentation Index
> Fetch the complete documentation index at: https://docs.plaisolutions.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Creating Answer Filters

> Step-by-step guide to creating effective Answer Filters for your agents

# Creating Answer Filters

Learn how to create Answer Filters that guide your agent's responses for specific queries. This guide will walk you through the entire process, from planning to implementation and testing.

## Prerequisites

Before creating Answer Filters, ensure you have:

<Check>**Agent Created**: You need an existing agent to add filters to</Check>
<Check>**Admin Access**: Only project Admins and Owners can create filters</Check>
<Check>**Clear Objectives**: Know what topics need filtered responses</Check>

<Info>
  Answer Filters are configured per agent in the **Answer Filters** tab of the agent dashboard.
</Info>

***

## Planning Your Filter

Before creating a filter, take time to plan its structure. A well-planned filter is more effective and easier to maintain.

### Step 1: Identify the Topic

What specific topic or question type needs consistent responses?

<Tabs>
  <Tab title="Examples">
    **Good Topics for Answer Filters:**

    * Return policy questions
    * Pricing inquiries
    * Contact information requests
    * Specific product features
    * Account troubleshooting steps

    **Too Broad:**

    * "Customer questions" (too vague)
    * "Product information" (too general)
    * "Help requests" (too open-ended)
  </Tab>

  <Tab title="Checklist">
    Ask yourself:

    * ✅ Is this a frequently asked question?
    * ✅ Does it require a consistent answer?
    * ✅ Is the information stable (not changing often)?
    * ✅ Are there specific phrases I want to avoid?
    * ✅ Do I have approved responses ready?
  </Tab>
</Tabs>

***

### Step 2: Collect Query Examples

Gather 3-5 different ways users might ask about this topic.

<CodeGroup>
  ```text Example: Refund Policy theme={null}
  "What is your refund policy?"
  "Can I get a refund?"
  "How do I return something for a refund?"
  "Do you offer money back guarantees?"
  "What's your return and refund process?"
  ```

  ```text Example: Account Login theme={null}
  "I can't log in to my account"
  "My password doesn't work"
  "How do I reset my password?"
  "I forgot my login credentials"
  "Can't access my account"
  ```

  ```text Example: Shipping Times theme={null}
  "How long does shipping take?"
  "When will my order arrive?"
  "What are your delivery times?"
  "How fast is shipping?"
  "Do you offer expedited delivery?"
  ```
</CodeGroup>

<Note>
  **Pro Tip**: Include variations in phrasing, formality, and wording. The more diverse your examples, the better the filter will match real user queries.
</Note>

***

### Step 3: Define Good Responses

Write 2-4 approved responses that you want the agent to use or incorporate.

**Characteristics of Good Responses:**

* ✅ Accurate and up-to-date
* ✅ Clear and specific
* ✅ Helpful and actionable
* ✅ Brand-aligned
* ✅ Complete (answers the full question)

<Tabs>
  <Tab title="Example 1">
    **Topic**: Refund Policy

    **Good Responses**:

    ```
    1. "We offer a 30-day money-back guarantee on all products."

    2. "To request a refund, log in to your account and go to 
       'Orders' > 'Request Return' for the item you want to return."

    3. "Refunds are processed within 5-7 business days after we 
       receive your returned item."

    4. "Items must be in original condition with tags attached 
       to qualify for a full refund."
    ```
  </Tab>

  <Tab title="Example 2">
    **Topic**: Technical Support

    **Good Responses**:

    ```
    1. "Click 'Forgot Password' on the login page to reset your 
       password via email."

    2. "Password reset emails arrive within 5 minutes. Check your 
       spam folder if you don't see it."

    3. "For security, password reset links expire after 24 hours. 
       Request a new one if yours has expired."
    ```
  </Tab>

  <Tab title="Example 3">
    **Topic**: Pricing

    **Good Responses**:

    ```
    1. "Our Basic plan starts at $29/month and includes up to 
       10 users and 5GB storage."

    2. "We offer three tiers: Basic ($29/mo), Professional ($79/mo), 
       and Enterprise (custom pricing)."

    3. "All plans include a 14-day free trial with no credit card 
       required."

    4. "Annual billing saves you 20% compared to monthly billing."
    ```
  </Tab>
</Tabs>

***

### Step 4: Identify Bad Responses

List 2-3 responses that should be avoided.

**Types of Bad Responses:**

* ❌ Incorrect information
* ❌ Vague or unhelpful answers
* ❌ Off-brand language
* ❌ Deflecting or dismissive responses
* ❌ Outdated information

<Tabs>
  <Tab title="Example 1">
    **Topic**: Refund Policy

    **Bad Responses**:

    ```
    1. "We don't offer refunds." 
       ❌ Incorrect if you do offer refunds

    2. "Just contact customer service."
       ❌ Too vague, not helpful

    3. "I don't know our refund policy."
       ❌ Unhelpful and unprofessional

    4. "All sales are final."
       ❌ Incorrect and creates poor experience
    ```
  </Tab>

  <Tab title="Example 2">
    **Topic**: Pricing

    **Bad Responses**:

    ```
    1. "It's free!"
       ❌ Misleading if not actually free

    2. "Prices vary depending on many factors."
       ❌ Too vague when specific info exists

    3. "Our pricing is complicated."
       ❌ Discouraging and unhelpful

    4. "I can't discuss pricing."
       ❌ Deflecting when pricing is public
    ```
  </Tab>

  <Tab title="Example 3">
    **Topic**: Data Privacy

    **Bad Responses**:

    ```
    1. "We might sell some of your data."
       ❌ Concerning and likely incorrect

    2. "Don't worry about it."
       ❌ Dismissive of legitimate concern

    3. "Your data is probably safe."
       ❌ "Probably" undermines confidence

    4. "I can't answer privacy questions."
       ❌ Evasive on important topic
    ```
  </Tab>
</Tabs>

***

## Creating the Filter

Now that you've planned your filter, let's create it in the platform.

### Step-by-Step Process

<Steps>
  <Step title="Navigate to Answer Filters Tab">
    1. Go to your **Project Dashboard**
    2. Select your **Agent**
    3. Click on the **Answer Filters** tab

    <Info>
      You should see a table listing any existing Answer Filters for this agent.
    </Info>
  </Step>

  <Step title="Open Create Filter Dialog">
    Click the **"+ Create Filter"** or **"Add Filter"** button.

    A dialog or sheet will open with the Answer Filter creation form.
  </Step>

  <Step title="Add Queries">
    In the **Queries** section:

    1. Enter your first query example
    2. Click **"+ Add"** or press Enter to add more
    3. Add 3-5 query variations

    **Tips:**

    * Include different phrasings
    * Vary the formality level
    * Consider typos or common misspellings
    * Think about how real users actually ask

    <Frame>
      ```
      Example Queries:
      [1] What is your refund policy?          [x]
      [2] Can I get my money back?             [x]
      [3] How do returns work?                 [x]
      [4] Do you offer refunds?                [x]
      [+] Add another query...
      ```
    </Frame>
  </Step>

  <Step title="Add Good Responses">
    In the **Good Responses** section:

    1. Enter your first approved response
    2. Click **"+ Add"** to add more
    3. Add 2-4 good responses

    **Tips:**

    * Be specific and actionable
    * Include all necessary details
    * Use your brand voice
    * Ensure accuracy

    <Frame>
      ```
      Good Responses:
      [1] We offer a 30-day money-back guarantee...  [x]
      [2] Refunds are processed within 5-7 days...   [x]
      [3] Items must be in original condition...     [x]
      [+] Add another response...
      ```
    </Frame>
  </Step>

  <Step title="Add Bad Responses">
    In the **Bad Responses** section:

    1. Enter responses to avoid
    2. Click **"+ Add"** to add more
    3. Add 2-3 bad responses

    **Tips:**

    * Include common mistakes
    * List any outdated information
    * Note off-brand phrasings
    * Add deflecting responses

    <Frame>
      ```
      Bad Responses:
      [1] We don't offer refunds.                    [x]
      [2] Just contact support.                      [x]
      [3] I don't know our refund policy.            [x]
      [+] Add another response...
      ```
    </Frame>
  </Step>

  <Step title="Review and Create">
    1. **Review all fields** for accuracy
    2. **Check for typos** in all text
    3. Click **"Create Filter"** button
    4. Wait for confirmation message

    <Check>
      You should see a success message and the filter will appear in your filters list.
    </Check>
  </Step>
</Steps>

***

## Complete Examples

### Example 1: E-commerce Returns

<CodeGroup>
  ```yaml Complete Filter theme={null}
  Filter Name: Returns & Refunds

  Queries:
    - "What is your return policy?"
    - "Can I return this product?"
    - "How do I get a refund?"
    - "Do you accept returns?"
    - "What's your refund process?"

  Good Responses:
    - "We accept returns within 30 days of purchase for a full refund."
    - "Items must be unused and in original packaging with all tags attached."
    - "To start a return, log in to your account and select 'Request Return' from your order history."
    - "Refunds are processed within 5-7 business days after we receive and inspect your return."
    - "Return shipping is free for defective items or our mistakes. For other returns, a $5.99 shipping fee applies."

  Bad Responses:
    - "All sales are final."
    - "We don't accept returns."
    - "Just contact customer service."
    - "I don't know the return policy."
  ```
</CodeGroup>

***

### Example 2: SaaS Pricing

<CodeGroup>
  ```yaml Complete Filter theme={null}
  Filter Name: Pricing & Plans

  Queries:
    - "How much does it cost?"
    - "What are your pricing plans?"
    - "Do you have a free plan?"
    - "How much is the subscription?"
    - "What's included in each plan?"

  Good Responses:
    - "We offer three plans: Starter ($29/month), Professional ($79/month), and Enterprise (custom pricing)."
    - "The Starter plan includes 10 users, 5GB storage, and basic features."
    - "Professional adds unlimited users, 50GB storage, advanced analytics, and priority support."
    - "All plans come with a 14-day free trial, no credit card required."
    - "Annual billing saves you 20% compared to monthly billing."

  Bad Responses:
    - "It's completely free."
    - "Pricing is complex and varies."
    - "I can't discuss pricing details."
    - "You'll have to contact sales for pricing."
  ```
</CodeGroup>

***

### Example 3: Technical Support

<CodeGroup>
  ```yaml Complete Filter theme={null}
  Filter Name: Password Reset

  Queries:
    - "I forgot my password"
    - "How do I reset my password?"
    - "Can't log in"
    - "My password doesn't work"
    - "Need to change my password"

  Good Responses:
    - "Click 'Forgot Password' on the login page to receive a reset link via email."
    - "The password reset email arrives within 5 minutes. Check your spam folder if you don't see it."
    - "For security, reset links expire after 24 hours. Request a new one if yours expired."
    - "After resetting, your password must be at least 8 characters with a mix of letters and numbers."

  Bad Responses:
    - "Your account is probably locked."
    - "Just try again later."
    - "Contact IT support."
    - "I can't help with password issues."
  ```
</CodeGroup>

***

## Testing Your Filter

After creating a filter, thorough testing is crucial.

### Testing Checklist

<AccordionGroup>
  <Accordion title="1. Test Exact Matches">
    Try the exact queries you defined:

    ```
    User: "What is your return policy?"

    Expected: Agent should incorporate your good responses
    Verify: ✅ Mentions 30-day policy
            ✅ Explains process
            ✅ Avoids bad responses
    ```
  </Accordion>

  <Accordion title="2. Test Variations">
    Try similar but different phrasings:

    ```
    User: "Can I send something back for a refund?"
    User: "What if I want to return an item?"
    User: "Tell me about your refund policy"

    Expected: Filter should still trigger
    Verify: ✅ Consistent information
            ✅ Good responses used
    ```
  </Accordion>

  <Accordion title="3. Test Edge Cases">
    Try queries that are somewhat related but different:

    ```
    User: "What if the product is damaged?"
    User: "Can I exchange instead of refund?"
    User: "Do you have a warranty?"

    Expected: May or may not trigger depending on similarity
    Verify: ✅ Response is still helpful
            ✅ If triggered, info is relevant
    ```
  </Accordion>

  <Accordion title="4. Test Negative Cases">
    Verify bad responses are avoided:

    ```
    Review several responses to the filtered queries.

    Verify: ❌ None of the bad responses appear
            ❌ No deflecting or vague answers
            ❌ No incorrect information
    ```
  </Accordion>
</AccordionGroup>

### Testing in Chat

<Steps>
  <Step title="Open Chat Interface">
    Navigate to your agent's chat interface or test environment
  </Step>

  <Step title="Ask Test Questions">
    Type each query variation and review responses
  </Step>

  <Step title="Document Results">
    Note which queries trigger the filter and response quality
  </Step>

  <Step title="Refine if Needed">
    If responses aren't as expected, edit the filter (see Managing Answer Filters guide)
  </Step>
</Steps>

***

## Best Practices

<CardGroup cols={2}>
  <Card title="Quality over Quantity" icon="badge-check">
    **Start with 5-10 critical filters**

    Focus on topics that:

    * Are asked frequently
    * Need consistent answers
    * Have compliance requirements
    * Protect brand reputation
  </Card>

  <Card title="Be Specific" icon="bullseye">
    **Detailed is better than vague**

    Instead of:
    ❌ "Our return policy is flexible"

    Use:
    ✅ "We accept returns within 30 days with original packaging for a full refund"
  </Card>

  <Card title="Include Context" icon="book-open">
    **Provide complete information**

    Good responses should:

    * Answer the full question
    * Include next steps
    * Provide relevant details
    * Link to more info if needed
  </Card>

  <Card title="Plan for Maintenance" icon="wrench">
    **Keep filters current**

    * Review monthly
    * Update when policies change
    * Remove outdated filters
    * Add new ones based on user questions
  </Card>
</CardGroup>

***

## Common Mistakes to Avoid

<Warning>
  Watch out for these common pitfalls:
</Warning>

<AccordionGroup>
  <Accordion title="❌ Too Few Query Examples">
    **Problem**: Only 1-2 query examples

    **Impact**: Filter won't trigger for many similar questions

    **Solution**: Add 3-5 diverse query variations
  </Accordion>

  <Accordion title="❌ Vague Good Responses">
    **Problem**: "We have a good return policy"

    **Impact**: Agent can't provide specific, helpful information

    **Solution**: Include specific details, numbers, timeframes
  </Accordion>

  <Accordion title="❌ No Bad Responses">
    **Problem**: Only defining what to say, not what to avoid

    **Impact**: Agent might still make known mistakes

    **Solution**: List 2-3 common mistakes or outdated info
  </Accordion>

  <Accordion title="❌ Overlapping Answer Filters">
    **Problem**: Multiple filters cover similar topics

    **Impact**: Inconsistent responses, confusion

    **Solution**: Keep filters distinct or combine related ones
  </Accordion>

  <Accordion title="❌ Outdated Information">
    **Problem**: Filter contains old policy or pricing

    **Impact**: Agent gives wrong information

    **Solution**: Set calendar reminders to review filters
  </Accordion>
</AccordionGroup>

***

## Trigger Threshold (Advanced)

<Note>
  **Default Setting**: 0.5 (50% similarity)

  Most users should stick with the default. Adjust only if you're experiencing issues with filter triggering.
</Note>

The trigger threshold determines how similar a user's query must be to your examples for the filter to activate.

### When to Adjust

<Tabs>
  <Tab title="Lower Threshold (0.3-0.4)">
    **Use when:**

    * Filter isn't triggering often enough
    * Topic is commonly asked in many different ways
    * You want broad matching

    **Example**:

    ```
    Filter: Pricing Information
    Threshold: 0.4

    Will trigger for:
    ✅ "How much?"
    ✅ "What's the cost?"
    ✅ "Is it expensive?"
    ✅ "Price?"
    ```

    <Warning>
      Risk of over-triggering (false positives)
    </Warning>
  </Tab>

  <Tab title="Higher Threshold (0.6-0.7)">
    **Use when:**

    * Filter triggers too often
    * You need precise matching
    * Topic is very specific

    **Example**:

    ```
    Filter: Enterprise Pricing
    Threshold: 0.7

    Will trigger for:
    ✅ "Enterprise pricing information"
    ✅ "How much for enterprise plan?"
    ❌ "How much?" (too general)
    ❌ "Tell me about plans" (too broad)
    ```

    <Warning>
      Risk of under-triggering (false negatives)
    </Warning>
  </Tab>

  <Tab title="Default (0.5)">
    **Recommended for most cases**

    Balanced approach that works well for:

    * Standard FAQ topics
    * Clear, distinct topics
    * General use cases

    **Behavior**:

    * Triggers for reasonably similar queries
    * Low false positive rate
    * Low false negative rate
  </Tab>
</Tabs>

***

## Next Steps

<Steps>
  <Step title="Create Your First Filter">
    Follow this guide to create 1-2 filters for your most critical topics
  </Step>

  <Step title="Test Thoroughly">
    Spend time testing with various query types
  </Step>

  <Step title="Monitor Usage">
    Check analytics to see how often filters trigger
  </Step>

  <Step title="Iterate">
    Refine based on real user interactions
  </Step>

  <Step title="Scale Up">
    Add more filters for additional topics
  </Step>
</Steps>

<CardGroup cols={2}>
  <Card title="Managing Answer Filters" icon="list-check" href="/agents/answer-filters/managing-answer-filters">
    Learn to edit, update, and delete filters
  </Card>

  <Card title="Overview" icon="book-open" href="/agents/answer-filters/overview">
    Learn more about Answer Filters concepts
  </Card>

  <Card title="Analytics" icon="chart-line" href="/agents/analytics/overview">
    Monitor filter effectiveness
  </Card>

  <Card title="Agent Settings" icon="sliders" href="/agents/settings">
    Configure other agent features
  </Card>
</CardGroup>

***

## Troubleshooting

<AccordionGroup>
  <Accordion title="Filter isn't triggering">
    **Possible causes:**

    * Query examples are too different from user queries
    * Threshold is too high
    * Not enough query examples

    **Solutions:**

    * Add more diverse query examples
    * Lower the threshold slightly (0.4)
    * Test with actual user phrasings
  </Accordion>

  <Accordion title="Filter triggers too often">
    **Possible causes:**

    * Query examples are too generic
    * Threshold is too low
    * Filter topic overlaps with others

    **Solutions:**

    * Make query examples more specific
    * Raise the threshold (0.6-0.7)
    * Review and consolidate overlapping filters
  </Accordion>

  <Accordion title="Agent ignores good responses">
    **Possible causes:**

    * Good responses are too vague
    * Agent's system prompt overrides filter
    * Model doesn't support filters well

    **Solutions:**

    * Make good responses more specific and directive
    * Review system prompt for conflicts
    * Ensure filter is actually triggering (test with exact queries)
  </Accordion>

  <Accordion title="Can't create filter (permission error)">
    **Possible causes:**

    * Insufficient permissions
    * Not an Admin or Owner

    **Solutions:**

    * Contact project Admin or Owner
    * Request elevated permissions
    * Check project membership role
  </Accordion>
</AccordionGroup>