Skip to main content

Knowledge & Guidelines

Knowledge and Guidelines work together to create intelligent, contextual AI agents. Knowledge provides the information your agent can access, while Guidelines shape how it communicates and behaves.

Knowledge base​

A knowledge base is your agent's collection of information that it relies on to answer questions and provide support. When a user asks a question, the agent searches connected knowledge bases, retrieves the most relevant fragments, and uses this information along with guidelines to generate a response.

This RAG (Retrieval Augmented Generation) approach ensures accurate, grounded responses based on your actual content.

For setup instructions, see Knowledge agent quick start.

Guidelines​

Guidelines are instructions that define your AI agent's personality, behavior, and response patterns.

Writing effective guidelines​

The quality of your guidelines directly impacts agent performance. Here's how to write guidelines that work:

1. Be specific and clear​

When users ask about pricing, always mention:

1. Starting price is $29/month
2. 14-day free trial available
3. Annual billing saves 20%
4. Enterprise plans available for 50+ users

Format prices with currency symbol first (e.g., $29 not 29$).

2. Handle objections strategically​

Price Objection Response Framework:
1. Acknowledge the concern: "I understand budget is an important consideration."
2. Highlight value: "Our customers typically save 10+ hours per week."
3. Reduce risk: "That's why we offer a 14-day free trial."
4. Provide options: "We also have a starter plan at $19/month."

Never dismiss price concerns or say "it's worth it" without explanation.

3. Set clear boundaries​

Topics to avoid:
- Political opinions or commentary
- Medical or legal advice
- Competitor pricing without context
- Promises about future features not yet released

If asked about these topics, politely redirect:
"I'm focused on helping you with [your product/service]. For [topic],
I'd recommend consulting with a qualified [professional type]."

4. Include examples​

When explaining our return policy:

User: "Can I return this after 30 days?"
Response: "Our standard return window is 30 days from purchase. If you're
slightly past that date, I can check if we can make an exception. May I
have your order number?"

Always offer to help even when the standard policy doesn't apply.

Predefined fields by agent type​

Each agent type has different predefined guideline fields optimized for its use case. Fill out the Goal field first to establish your agent's primary objective, then add supporting context in the other fields.

Customer Support agents use Objections and Custom instructions fields.

Custom instructions:

Deliver clear and knowledgeable assistance in every interaction.
Always begin by acknowledging the customer's concern and showing understanding of their situation.
Guide them through solutions in a step-by-step manner, ensuring they feel supported throughout the process.
When possible, offer educational resources, FAQs, or tutorials that empower customers to solve issues independently in the future.
Turn challenges into positive experiences that strengthen trust and loyalty.

Objections:

## I've already tried that and it didn't work
I'm sorry that didn't resolve your issue. Let me look into this more closely and find an alternative solution for you.

Adding custom guidelines​

In addition to the predefined fields, you can add custom guidelines for specific topics or scenarios. Each custom guideline has a title and body.

To add a custom guideline:

  1. Navigate to your AI Agent's Knowledge tab
  2. Click Add guideline
  3. Enter a descriptive title (e.g., "Return Policy", "Pricing Questions")
  4. Write the guideline content in the body

Advanced guideline techniques​

Dynamic personalization​

Use context variables to personalize responses:

If the user's account type is "Premium":
- Mention premium-only features
- Provide priority support language
- Reference their dedicated account manager

If the user is a new customer (account age < 30 days):
- Be extra helpful and patient
- Proactively offer onboarding resources
- Check in on their progress

Conditional responses​

Create branching logic in your guidelines:

For technical questions:
- If basic user: Provide simple, non-technical explanation
- If power user: Include advanced options and shortcuts
- If developer: Reference API docs and code examples

Identify user type by their question complexity and terminology used.

Multi-language considerations​

When responding in languages other than English:
- Maintain the same tone and professionalism
- Use formal address unless the culture prefers informal
- Be aware of cultural sensitivities
- Localize examples (currency, dates, measurements)

Testing and iterating guidelines​

Your guidelines should evolve based on real interactions:

  1. Start simple - Begin with core guidelines and add complexity gradually
  2. Monitor conversations - Review actual interactions to identify gaps
  3. A/B test variations - Try different guideline approaches
  4. Gather feedback - Ask users if responses were helpful
  5. Update regularly - Refine based on new products, policies, or learnings
Pro tip

Use the Insights feature to get AI-powered suggestions for improving your guidelines based on actual conversation patterns.

Common guideline mistakes to avoid​

MistakeWhy it's problematicBetter approach
Too vague"Be helpful" doesn't guide specific behavior"When users ask for help, first clarify their specific need, then provide step-by-step guidance"
Contradictory rulesConfuses the agent and creates inconsistent responsesReview all guidelines together to ensure alignment
Over-scriptingMakes responses robotic and inflexibleProvide frameworks and principles, not word-for-word scripts
Ignoring edge casesAgent struggles with unusual requestsInclude "If unsure..." fallback instructions
No personalityCreates generic, unmemorable interactionsDefine specific tone, style, and even catchphrases

Combining knowledge and guidelines​

The magic happens when knowledge and guidelines work together:

Example scenario: User asks "How do I reset my password?"

  1. Knowledge provides: The actual reset process from your documentation
  2. Guidelines shape: How to present this information (tone, detail level, additional offers)

Result: A response that's both accurate (from knowledge) and on-brand (from guidelines).

Best practices summary​

✅ Do:

  • Write guidelines as if training a new team member
  • Use specific examples and scenarios
  • Update guidelines based on real conversations
  • Test different approaches and measure results
  • Keep guidelines organized and non-conflicting

❌ Don't:

  • Write vague or generic instructions
  • Create overly rigid scripts
  • Ignore cultural and language considerations
  • Forget to handle edge cases
  • Neglect regular updates and improvements

Next steps​

Last updated on