Before you start
Your AI agent must be created (AI Agents > Create Agent)
Your knowledge base should have initial content — guidelines control behavior, not knowledge
What behavioral guidelines do
Behavioral guidelines are free-text instructions that shape how your AI voice agent talks, what it can and can't say, and when it should hand off to a human. Think of them as the training manual for a new hire — except the AI follows them on every single call.
You'll find them in AI Agents > select your agent > Behavioral Guidelines field.
Writing effective guidelines
Tone and personality
No need to be overly specific on how the AI Should sound, by default it will mimic the 'vibes' from your knowledge base content.
You might want to specify certain characteristics specific to your client base, such as - Speak slowly since you are speaking to elderly people
Boundaries — what the AI should never do
These are the most important guidelines. List specific things the AI must avoid:
- Never promise a specific delivery date. Say "typically ships within 2–3 business days" not "it'll arrive Friday."
- Never offer discounts beyond what's listed in the knowledge base.
- Never make up information. If you don't know the answer, say "Let me connect you with our team" and transfer.
- Never discuss competitor products or pricing.
- Never share internal information (employee names, supplier names, cost prices).
- Do not offer callbacks within the next hour — say "our team will follow up within 24 hours."
Escalation rules — when to transfer
Be explicit about when the AI should stop trying to resolve and hand off to a human:
Transfer to a live agent when:
- The customer explicitly asks to speak with a person or a manager
- The customer mentions a chargeback, dispute, BBB complaint, or legal action
- The customer's issue involves modifying a subscription
- The issue requires accessing systems the AI doesn't have
- The customer has asked the same question 3 times without getting a satisfactory answer
- The customer sounds frustrated or their tone escalates
>
Before transferring, always say: "Let me connect you with a team member who can help. I'll share everything we've discussed so they can pick up right where we left off."
Product-specific instructions
Add context for products or scenarios that need special handling:
- For the "Pro Bundle" — always mention the 30-day trial period and the money-back guarantee before quoting the price.
- For sizing questions on footwear — recommend going half a size up based on customer reviews. Direct them to the size chart at [URL].
- For wholesale inquiries — this is not something we handle by phone. Transfer to the sales team.
Template to start from
Copy and customize this template:
TONE:- Be friendly and helpful. Keep responses concise.- Match the customer's communication style.BOUNDARIES:- Never promise specific delivery dates.- Never offer discounts not in the knowledge base.- Never make up information — transfer if unsure.- Do not discuss competitors.ESCALATION — TRANSFER WHEN:- Customer asks for a person or manager.- Customer mentions chargeback, dispute, or complaint.- Customer is frustrated or repeating themselves.- Issue requires human judgment or system access the AI doesn't have.TRANSFER SCRIPT:"Let me connect you with our team. I'll share everything we've discussed so you won't have to repeat yourself."PRODUCT NOTES:- [Add product-specific guidance here]
Testing your guidelines
After updating guidelines:
Open Test Agent in the AI Agents page
Try to trick the AI — ask for a discount, ask for a delivery date promise, pretend to be angry
Verify the AI follows your boundaries
Test escalation — say "I want to speak with a manager" and confirm it offers to transfer
Adjust and re-test until the behavior matches your expectations
Common issues
AI gives wrong answers despite guidelines: Guidelines control behavior, not knowledge. If the AI states incorrect facts, update the knowledge base, not the guidelines. Makes sure to avoid duplicate and contradictory knowledge
AI never transfers: Your escalation rules might be too narrow. Add more trigger phrases or situations.
AI transfers too much: Your boundaries might be too strict. Loosen "transfer if unsure" to "try once, then transfer if the customer isn't satisfied."
AI ignores tone instructions: Make tone instructions specific and actionable. "Be professional" is ignored; "Use contractions, avoid exclamation marks, and acknowledge frustration before solving" gets followed.
Maintenance
Review your behavioral guidelines monthly:
Are there new products or policies that need specific handling?
Has customer feedback revealed scenarios the guidelines don't cover?
Are there common transfer reasons that the AI could handle with better guidelines?