How to Set up Guardrails for Your AI Agent

Quack gives you full control over how your AI Agent interacts with your customers. Guardrails help you define what the AI Agent can and cannot say or do, ensuring every interaction aligns with your brand, values, and policies. You can set these parameters using briefs and instructions, and the custom tone prompt.

What Are Guardrails?

Guardrails are rules, tone settings, and limits that shape the AI Agent's behavior. They ensure your AI Agent stays on-brand, handles sensitive topics properly, and knows when to escalate.

Three Key Ways to Set Guardrails

1. Briefs

A brief is a written set of knowledge and logic that tells Quack how to respond to a specific topic.

Use briefs to:

  • Highlight key messaging priorities for a topic

  • Guide the structure or flow of answers

  • Clarify how specific features, plans, or processes work

  • Outline key points to always cover in a response

2. Instructions

Instructions are short, direct commands that tell Quack what to always or never do in a specific topic. They’re perfect for enforcing quick, repeatable rules.

Use instructions to:

  • Enforce boundaries (e.g., "Never provide pricing over chat")

  • Guide behavior (e.g., "Always confirm the user’s plan")

  • Define escalation triggers (e.g., "Escalate if refund is mentioned")

  • Restrict actions (e.g., "Never ask for contact information")

3. Custom Tone Prompt

The Custom Tone Prompt defines how your AI Agent should communicate and sets persistent guidelines for how it should behave across every case, topic, or use this agent handles.

Use the custom tone promt to:

  • Define your brand voice (e.g., “Sound confident and friendly”)

  • Maintain tone consistency across all topics

  • Set guardrails that apply to every interaction (e.g., “Avoid slang or casual language”)

  • Align the AI Agent answers with your support or brand standards

Fallback Logic

Fallback logic defines how your AI Agent responds when it cannot confidently answer a question. Common fallback actions include:

  • Escalating to a live agent

  • Providing a neutral handoff message

  • Linking to a help article

💡 Pro Tip: For Ticket Instances, you can enable Strict Mode to further tighten guardrails. In Strict Mode, your AI Agent will only respond to questions it has been explicitly trained on, everything else is deferred or escalated. This is ideal for teams operating in sensitive or regulated environments.

How to Set Guardrails in Quack AI

  1. Go to your Quack Training Center

  2. Select the relevant Agent

  3. Add the Custom Tone Prompt in Configurations

  4. Add or edit your Brief and Instructions under specific topics

  5. Save and test in sample conversations before releasing

Best Practices

  • Keep it simple: Clear, concise language makes guidance easier for your AI Agent to follow

  • Update regularly: Review briefs and instructions as your product or policies evolve

  • Monitor and refine: Use Simulations to see where guardrails help or need tweaking

Why Guardrails Matter

Well-defined guardrails let your AI Agent operate safely and effectively, giving you confidence that every automated interaction meets your standards.