Agent-Specific Instructions

Customize AI behavior differently for each live chat agent on your team. This allows personalized AI assistance that adapts to each agent's style, expertise, and preferences.

How to get there: Go to AI agents in the left menu → click your chatbot → Instructions in the sidebar.

To hide private contact details or old prices, see Hide private contact details and old prices.

Skills and Approval Gating

Skills (previously called Actions) are the discrete abilities you enable for a bot, things like sending an email, booking a meeting, or calling a webhook tool. Each skill is opt-in per bot.

Per-tool approval gating means sensitive tools can require your confirmation before the AI executes them, which is useful for anything that writes or sends. When a gated tool is triggered, the request pauses until an agent approves or rejects it in the inbox.

Per-action logs are viewable in the tool logs view, so you can audit exactly what the bot ran, with what arguments, and what came back.

What Are Agent-Specific Instructions?

While your chatbot has global custom instructions that apply to all AI responses, agent-specific instructions override or supplement these when a conversation is assigned to a particular agent.

Think of it as: Global instructions = Company voice, Specific instructions = Individual agent personality

How It Works

Without Agent-Specific Instructions:

All conversations use the same AI behavior:
  "You are a helpful support assistant for [Company].
   Always be professional and concise."

Every agent gets AI assistance with identical behavior.

With Agent-Specific Instructions:

When assigned to Sarah (Sales):
  "Focus on product benefits and ROI.
   Highlight enterprise features. Ask for budget and timeline."

When assigned to Mike (Technical Support):
  "Provide detailed technical explanations.
   Ask for error messages and logs. Reference documentation."

When assigned to Lisa (Customer Success):
  "Be empathetic and solution-focused.
   Emphasize long-term relationship. Offer training resources."

Each agent gets AI assistance tailored to their role and style.

Use Cases

By Department

Sales Agents:

- Focus on value proposition and ROI
- Ask qualifying questions (budget, decision maker, timeline)
- Suggest scheduling a demo
- Highlight case studies and success stories
- Be persuasive and confident

Support Agents:

- Prioritize solving the immediate problem
- Ask for detailed error information
- Reference help documentation
- Be patient and thorough
- Offer step-by-step guidance

Customer Success:

- Focus on customer goals and outcomes
- Suggest best practices and optimization
- Offer training and education
- Build long-term relationships
- Proactive problem-solving

By Expertise Level

Senior Agents:

- Provide advanced technical details
- Assume customer has basic knowledge
- Use industry terminology
- Offer architecture-level guidance

Junior Agents:

- Provide more context and explanation
- Use simpler language
- Include links to documentation
- Suggest when to escalate to senior agent

By Language/Region

English-speaking agents:

- Use American English spelling
- Reference USD pricing
- Mention US business hours

Spanish-speaking agents:

- Respond in Spanish
- Reference EUR/MXN pricing
- Adapt to regional customs

By Communication Style

Agent #1 (Formal):

- Use professional, formal tone
- Complete sentences
- Proper grammar and punctuation

Agent #2 (Casual):

- Friendly, conversational tone
- Use emojis occasionally
- Be approachable and warm

Setting Up Agent-Specific Instructions

Step 1: Access Agent Settings

Go to Live Chat → Agents
Click on an individual agent
Find "Specific Instructions" or "Custom AI Behavior"

Step 2: Write Instructions

Write clear, specific instructions for how AI should behave for this agent:

Good examples:

"When assisting this agent:
- Focus on technical accuracy
- Always include code examples when relevant
- Reference the API documentation
- Ask for programming language being used
- Be detailed and thorough"

"This agent handles enterprise sales:
- Ask about company size and use case
- Emphasize security and compliance features
- Suggest scheduling a call with their manager
- Be confident and consultative
- Mention SLA and premium support"

Bad examples:

"Help the customer"  // Too vague

"Be different"  // Doesn't explain how

"Sales"  // Not enough detail

Step 3: Test

Assign a test conversation to this agent
Verify AI behavior matches your instructions
Refine as needed

Step 4: Repeat for Each Agent

Set up specific instructions for each agent on your team, or just the ones that need customization (others will use global instructions).

Relationship to Global Instructions

Agent-specific instructions work in addition to global instructions:

Global Instructions (apply to all):

"You are a helpful assistant for Acme Corp.
 Always be professional and respectful.
 Never share pricing without qualifying the lead."

Agent-Specific (for Sales Agent):

"Focus on ROI and business value.
 Ask about budget and timeline.
 Suggest booking a demo."

Result for Sales Agent:
The AI follows BOTH sets of instructions - being professional and respectful (global) while also focusing on ROI and demos (specific).

Priority When Conflicting

If specific instructions contradict global:

Agent-specific instructions take precedence
This allows customization per agent

Example:

Global: "Keep responses under 50 words"
Specific: "Provide detailed, thorough explanations"
Result: This agent gets detailed responses (specific wins)

Best Practices

1. Start with Global Instructions

Define your company-wide voice and rules first:

Brand tone
Key policies (don't share pricing, don't make promises, etc.)
Basic behavior

Then add agent-specific only where needed.

2. Keep It Focused

Each agent's instructions should be:

Specific: "Ask for error logs" not "be helpful"
Actionable: "Suggest scheduling demo" not "be sales-oriented"
Concise: 3-5 key points, not a novel

3. Align with Agent Skills

Match instructions to actual agent expertise:

Don't tell AI to offer technical help for non-technical agents
Don't push sales tactics on support agents

4. Update Based on Feedback

Ask agents:

"Is the AI helpful when assisting you?"
"What should it do differently?"
"What does it do wrong?"

Refine instructions based on their input.

5. Test Edge Cases

Check that instructions work for:

Simple questions
Complex questions
Angry customers
Upsell opportunities

6. Document Your Setup

Keep notes on why each agent has specific instructions:

Sarah - Sales focus (hired specifically for enterprise)
Mike - Technical (Senior engineer, handles complex issues)
Lisa - Onboarding (New hire, needs more guidance from AI)

Advanced Tips

Include Agent Name

Reference the agent's name in instructions:

"When helping Lisa, remember she's new to the team.
 Provide extra context and suggest when to ask manager for help."

Conditional Logic

Use instructions that adapt:

"If customer mentions integration:
  - Ask which platform they use
  - Reference our API docs
  - Offer to schedule technical call

 If customer mentions pricing:
  - Qualify their company size first
  - Don't share exact pricing
  - Suggest demo to discuss custom pricing"

Knowledge Boundaries

Define what the AI should/shouldn't attempt:

"This agent handles:
 - Product questions
 - Basic troubleshooting
 - Feature requests

 This agent does NOT handle:
 - Billing issues (escalate to billing team)
 - Legal questions (escalate to manager)
 - Custom development (escalate to solutions team)"

Response Templates

Include suggested structures:

"When answering technical questions, follow this structure:
 1. Acknowledge the question
 2. Provide step-by-step solution
 3. Ask if they need clarification
 4. Offer link to documentation"

Examples by Industry

E-commerce

Regular Support Agent:

"Focus on order status, shipping, and returns.
 Always ask for order number.
 Be empathetic about delays.
 Offer discount codes for issues (10% max).
 Escalate refund requests > $100."

VIP Support Agent:

"Serve high-value customers (LTV > $1000).
 Be extra accommodating.
 Offer expedited shipping for free.
 Can approve refunds up to $500.
 Emphasize exclusive perks."

SaaS

Trial Support Agent:

"Help users get value from trial quickly.
 Focus on core features and quick wins.
 Ask about their use case and goals.
 Suggest relevant tutorials.
 Gently remind of trial expiration."

Enterprise Support Agent:

"Handle complex enterprise accounts.
 Focus on integration and API questions.
 Emphasize security and compliance.
 Offer architecture review.
 Can escalate to solutions engineer."

Professional Services

Consultation Scheduler:

"Goal: Book consultation calls.
 Ask about their current challenges.
 Emphasize expertise and track record.
 Suggest specific time slots.
 Collect detailed contact info."

Project Manager:

"Handle existing client questions.
 Focus on project progress and next steps.
 Be proactive about potential issues.
 Offer status updates.
 Coordinate with other team members."

Troubleshooting

AI isn't following specific instructions

Check:

Instructions are clear and specific
No conflicting global instructions
Agent is actually assigned to conversation
Instructions saved properly

Fix:

Rewrite more explicitly
Test with simple questions first
Check for typos or formatting issues

Instructions work inconsistently

Cause: Instructions might be ambiguous

Fix:

Use more specific, directive language
Include examples of desired behavior
Remove vague terms ("sometimes", "maybe")

Different agents getting same behavior

Cause: Specific instructions not set

Fix:

Verify each agent has unique instructions
Check that conversations are assigned to correct agent
Ensure specific instructions aren't blank

Quick Start Template

Copy and customize for each agent:

## Agent: [NAME]
## Role: [Sales/Support/Success]
## Expertise: [Junior/Senior]

### Primary Focus:
- [Key responsibility 1]
- [Key responsibility 2]
- [Key responsibility 3]

### Communication Style:
- Tone: [Formal/Casual/Friendly]
- Detail Level: [Concise/Detailed]
- Language: [English/Spanish/etc]

### Do's:
- ✅ [Specific action 1]
- ✅ [Specific action 2]
- ✅ [Specific action 3]

### Don'ts:
- ❌ [What to avoid 1]
- ❌ [What to avoid 2]

### Escalation Criteria:
- [When to escalate to manager]
- [When to route to another team]

Fill this out for each agent, then paste the result into their specific instructions field.

Measuring Success

Track whether agent-specific instructions help:

Metrics to monitor:

Agent satisfaction with AI assistance
Time to resolution (does AI speed things up?)
Customer satisfaction scores
Escalation rate (fewer escalations = AI helping more)

Ask agents monthly:

"Does the AI help or hinder you?"
"What should the AI do differently?"
"Rate AI helpfulness 1-10"

Continuously refine based on feedback.