Answer Filters Overview

Answer Filters are a powerful feature in PLai Framework that allow you to define specific response patterns for your agents. They enable you to control what your agent should or shouldn’t say when users ask certain types of questions, ensuring consistent, accurate, and brand-appropriate responses.

What are Answer Filters?

Answer Filters act as intelligent guards for your agent’s responses. They work by:

Detecting specific query patterns that match your defined triggers
Evaluating the similarity between user queries and your filter definitions
Guiding the agent to use good responses or avoid bad responses

Think of Answer Filters as training wheels that help your agent stay on track, especially for critical topics where consistency and accuracy matter most.

Control Responses

Define exactly what your agent should say for specific queries

Prevent Mistakes

Block inappropriate or incorrect responses before they reach users

Brand Consistency

Ensure responses align with your brand voice and policies

Quality Assurance

Maintain high-quality, accurate information delivery

How Answer Filters Work

Answer Filters operate using a similarity-based matching system:

Key Components

Queries
Good Responses
Bad Responses
Trigger Threshold

Definition: Example questions or topics that trigger this filter.Purpose: Help the system identify when to apply the filter based on user input similarity.Example:

- "What are your refund policies?"
- "Can I return this product?"
- "How do I get my money back?"

These query examples train the filter to recognize similar questions from users.

Definition: Approved, preferred responses for the matched queries.Purpose: Guide the agent toward giving these specific answers when the filter is triggered.Example:

- "We offer a 30-day money-back guarantee on all products."
- "You can initiate a return through your account dashboard."
- "Refunds are processed within 5-7 business days."

The agent will be encouraged to incorporate this information in its response.

Definition: Responses that should be avoided or blocked.Purpose: Prevent the agent from saying specific things that could be harmful, incorrect, or off-brand.Example:

- "We don't offer refunds." (if you do offer them)
- "Just contact support." (too vague)
- "I don't know our refund policy." (unhelpful)

The agent will actively avoid these response patterns.

Use Cases

Answer Filters excel in scenarios where consistency and accuracy are crucial:

1. Policy & Compliance

Ensure your agent always provides accurate information about company policies, legal requirements, or compliance matters.

Example: Return Policy Filter

Queries:
  - "What is your return policy?"
  - "Can I return items?"
  - "How do returns work?"

Good Responses:
  - "We accept returns within 30 days of purchase with original packaging."
  - "Items must be unused and in original condition."
  - "Refunds are issued to the original payment method within 5-7 business days."

Bad Responses:
  - "All sales are final." (incorrect)
  - "I'm not sure about our return policy." (unhelpful)
  - "You'll have to ask someone else." (deflecting)

2. Pricing & Billing

Critical for preventing misinformation about costs, pricing tiers, or billing processes.

Example: Pricing Filter

Queries:
  - "How much does it cost?"
  - "What are your prices?"
  - "Do you have a free plan?"

Good Responses:
  - "Our Basic plan starts at $29/month."
  - "We offer three tiers: Basic ($29), Pro ($79), and Enterprise (custom pricing)."
  - "Yes, we have a 14-day free trial with no credit card required."

Bad Responses:
  - "It's free." (if not accurate)
  - "Prices vary." (too vague)
  - "I don't know the pricing." (unhelpful)

3. Brand Voice & Messaging

Maintain consistent brand personality and messaging across all interactions.

Example: Brand Tone Filter

Queries:
  - "Who are you?"
  - "Tell me about your company"
  - "What makes you different?"

Good Responses:
  - "We're a customer-first company dedicated to making AI accessible to everyone."
  - "Our mission is to empower businesses with intelligent automation."
  - "We combine cutting-edge technology with exceptional support."

Bad Responses:
  - "We're just another AI company." (weak positioning)
  - "I don't know much about the company." (not brand-aligned)
  - Generic corporate speak that doesn't match your brand

4. Technical Support

Example: Troubleshooting Filter

Queries:
  - "My account won't log in"
  - "I forgot my password"
  - "Can't access my account"

Good Responses:
  - "Click 'Forgot Password' on the login page to reset your password."
  - "Check your email for a password reset link (it may take a few minutes)."
  - "If you don't receive the email, check your spam folder."

Bad Responses:
  - "Your account is probably locked." (assumption)
  - "Just try again later." (not helpful)
  - "Contact IT support." (deflecting when self-service is available)

5. Sensitive Topics

Handle sensitive or controversial topics with predefined, carefully crafted responses.

Example: Data Privacy Filter

Queries:
  - "Do you sell my data?"
  - "How do you use my information?"
  - "Is my data safe?"

Good Responses:
  - "We never sell your personal data to third parties."
  - "Your data is encrypted and stored securely in compliance with GDPR."
  - "You can view our full privacy policy at [link]."

Bad Responses:
  - "We might share some data." (vague and concerning)
  - "Don't worry about it." (dismissive)
  - "I can't answer privacy questions." (evasive)

Benefits of Answer Filters

Consistency

Ensure uniform responses across all user interactions for the same questions.

Same information every time
No contradictory answers
Reliable user experience

Accuracy

Guarantee correct information for critical topics.

Prevent hallucinations on important topics
Override model uncertainties
Maintain factual accuracy

Compliance

Meet legal and regulatory requirements with consistent messaging.

Legal disclaimer compliance
GDPR/privacy law adherence
Industry-specific regulations

Brand Protection

Protect your brand reputation by controlling sensitive responses.

Prevent off-brand messages
Avoid PR issues
Maintain professional tone

Quality Control

Improve response quality for common questions.

Better than generic AI responses
Incorporate domain expertise
Reduce need for human intervention

When to Use Answer Filters

✅ Good Use Cases
❌ When Not to Use
⚖️ Consider Carefully

Answer Filters are ideal for:

📋 Company policies (returns, privacy, terms)
💰 Pricing and billing information
🔒 Security and compliance topics
🏢 Brand messaging and positioning
🆘 Critical troubleshooting steps
📞 Contact information and escalation paths
⚖️ Legal disclaimers and requirements
🎯 FAQs with specific approved answers

Characteristics:

High-stakes information
Needs to be consistent
Cannot be “creative” or varied
Subject to legal/compliance review

Answer Filters vs. Other Approaches

Understanding when to use Answer Filters versus other PLai Framework features:

Feature	Best For	Answer Filters	System Prompt	Datasources
Specific FAQs	Exact, consistent answers	✅ Ideal	⚠️ Can drift	⚠️ May vary
General Knowledge	Broad information	❌ Too rigid	✅ Ideal	✅ Ideal
Policies	Legal/compliance	✅ Ideal	⚠️ May forget	✅ Good
Dynamic Data	Real-time info	❌ Static	❌ No access	✅ Ideal
Brand Voice	Tone and style	⚠️ Limited	✅ Ideal	❌ Wrong tool
Prohibited Content	Block specific responses	✅ Ideal	⚠️ Can be overridden	❌ Wrong tool

Best Practice: Combine multiple approaches for optimal results. Use Answer Filters for critical, specific content, System Prompts for general behavior, and Datasources for knowledge retrieval.

Limitations & Considerations

Important Limitations to Understand:

Similarity Matching

Answer Filters use semantic similarity, not exact matching:

User queries don’t need to match exactly
Similar phrasings will trigger the same filter
Very creative phrasings might not trigger

Not a Hard Block

Answer Filters guide the agent but don’t completely override its behavior:

Agents can still deviate slightly from good responses
Answer Filters influence but don’t guarantee exact wording
For absolute control, consider using Structured Output mode

Maintenance Required

Keep filters updated as policies change
Review and refine based on actual usage
Remove outdated filters promptly

Performance Impact

Too many filters can slow response time slightly
Each filter adds processing overhead
Aim for 10-20 filters per agent (not a hard limit)

Best Practices

Start Small

Begin with your most critical topics

Identify 3-5 critical topics first
Test thoroughly before adding more
Expand based on actual user interactions

Be Specific

Clear, detailed definitions work best

Provide multiple query examples (3-5 per filter)
Write concrete good/bad responses
Avoid vague or ambiguous language

Test Thoroughly

Validate filter behavior before deployment

Test with various phrasings
Check edge cases
Monitor initial interactions closely

Monitor & Iterate

Continuously improve your filters

Review agent conversations regularly
Add new query examples as you discover them
Refine responses based on user feedback

Document Your Answer Filters

Keep track of why filters exist

Note the business reason for each filter
Document expected behavior
Set reminders to review periodically

Getting Started

Ready to create your first Answer Filter?

Identify Critical Topics

List questions where consistency is crucial

Draft Responses

Write approved responses for each topic

Create Filter

Navigate to the Answer Filters tab and click “Create Filter”

Test Behavior

Test with multiple query variations

Monitor & Adjust

Watch real conversations and refine as needed

Next Steps

Creating Answer Filters

Step-by-step guide to creating your first filter

Managing Answer Filters

Learn to edit, update, and delete filters

Agent Settings

Configure other agent features

Analytics

Monitor how filters affect conversations

Common Questions

How many Answer Filters can I create per agent?

There’s no hard limit, but we recommend keeping it under 50 filters per agent for optimal performance. Focus on quality over quantity – a few well-crafted filters are more effective than many vague ones.

Can filters conflict with each other?

If multiple filters match the same query, the system will consider all applicable filters. This usually works well, but you should test to ensure filters complement rather than contradict each other.

Do Answer Filters work in all languages?

Yes, Answer Filters work with multilingual agents. However, you’ll need to create separate filters for each language or use the Language Detection feature to maintain consistency across languages.

Can I export/import filters between agents?

Currently, filters are agent-specific and cannot be bulk exported/imported. However, you can manually recreate filters on different agents. Bulk management features are on our roadmap.

What's the difference between Answer Filters and Guardrails?

Answer Filters guide specific responses to specific queries. Guardrails enforce broader safety and compliance rules across all interactions. Use both for comprehensive control.

​Answer Filters Overview

​What are Answer Filters?

Control Responses

Prevent Mistakes

Brand Consistency

Quality Assurance

​How Answer Filters Work

​Key Components

​Use Cases

​1. Policy & Compliance

​2. Pricing & Billing

​3. Brand Voice & Messaging

​4. Technical Support

​5. Sensitive Topics

​Benefits of Answer Filters

​When to Use Answer Filters

​Answer Filters vs. Other Approaches

​Limitations & Considerations

​Similarity Matching

​Not a Hard Block

​Maintenance Required

​Performance Impact

​Best Practices

​Getting Started

​Next Steps

Creating Answer Filters

Managing Answer Filters

Agent Settings

Analytics

​Common Questions

Answer Filters Overview

What are Answer Filters?

How Answer Filters Work

Key Components

Use Cases

1. Policy & Compliance

2. Pricing & Billing

3. Brand Voice & Messaging

4. Technical Support

5. Sensitive Topics

Benefits of Answer Filters

When to Use Answer Filters

Answer Filters vs. Other Approaches

Limitations & Considerations

Similarity Matching

Not a Hard Block

Maintenance Required

Performance Impact

Best Practices

Getting Started

Next Steps

Common Questions