Skip to main content

​
Timing

Give users time to orient themselves before offering help.Recommended delays:
  • Simple pages: 30-45 seconds
  • Complex forms: 45-90 seconds
  • Technical setup: 90-180 seconds
Distinguish between:
  • User reading carefully (good)
  • User stuck and frustrated (needs help)
Use multiple signals: time + inactivity + scroll position.
Exit intent (mouse leaving window) is powerful but can feel aggressive. Reserve for high-value pages like checkout.

​
Frequency

  • Max 2-3 triggers per session
  • Never more than 1 at a time
  • Wait 5+ minutes between triggers
If a user dismisses a trigger:
  • Don’t show the same trigger again that session
  • Consider a 24-hour cooldown for that event
  • Track dismissal patterns to improve targeting
Users who see too many triggers become blind to them. Quality over quantity - fewer, better-timed triggers perform better.

​
Messaging

❌ “Need help?” ✅ “Having trouble with payment? I can help!”❌ “Click here for support” ✅ “Stuck on this step? I can walk you through it.”
If your brand is:
  • Professional: “Need assistance with checkout?”
  • Friendly: “Hey! Can I help with that?”
  • Technical: “Encountering an issue? Let’s debug together.”
When users are refining code, use edit mode for targeted changes:
  • Toggle switch: Use the Chat/Edit/Plan toggle next to the send button
  • Per-message Edit button: Click the Edit action on a specific message to target it
  • Be specific: “Add input validation to the email field” works better than “make it better”
  • The agent analyzes your request and applies precise edits to workspace files
  • Workspace required: Edit mode requires the agent’s workspace setting to be enabled
When a task involves multiple tools or steps, use plan mode to review before executing:
  • Toggle switch: Use the Chat/Edit/Plan toggle next to the send button
  • Review first: The AI proposes a step-by-step plan — review it before approving
  • Modify plans: Reply with changes to refine the plan before execution
  • Approve to execute: Confirm the plan and the AI executes each step automatically
  • Plan mode works best for multi-tool workflows like searching, editing, and deploying
  • Workspace required: Plan mode requires the agent’s workspace setting to be enabled
✅ “Your cart is saved for 30 minutes” ❌ “Buy now or lose your cart!”✅ “Questions before you go?” ❌ “Don’t leave! We need you!”

​
Context

The more context your agent has, the better it can help:
// Good
boostgpt.trigger('checkout_help', {
  cart_value: '$149',
  items: 3,
  user_name: 'Sarah',
  time_on_page: 45,
  current_step: 'payment'
});

// Not as helpful
boostgpt.trigger('help', {});

{
  user: {
    id: 'user_123',
    name: 'Sarah Chen',
    email: 'sarah@example.com',
    plan: 'pro',
    account_age: '6 months'
  }
}
This enables personalized responses.
Include:
  • Current page/step
  • Previous pages visited
  • Actions already taken
  • Time spent

​
Trigger Types

SituationBest Type
General availabilityButton
Proactive outreachToast
Urgent/importantBanner
Contextual helpInline
Toasts feel personal and conversational. Use them when you want higher engagement. Stacked messages create a chat-like experience.
Buttons are less intrusive and always available. Good for “just in case” help without interrupting flow.

​
Training Your Agent

Your agent should know:
  • Product features and pricing
  • Common issues and solutions
  • Policies (shipping, returns, refunds)
  • Technical documentation
Templates should:
  • Describe the situation clearly
  • Include all relevant context
  • Provide instructions on how to help
  • Set the right tone
Before deploying, test:
  • Common user questions
  • Edge cases
  • Error scenarios
  • Various context combinations

​
Analytics & Optimization

MetricWhat It Tells You
Trigger RateHow often conditions are met
View RateHow often users see triggers
Engagement RateHow often users click
Resolution RateHow often chat helps
Conversion LiftImpact on desired outcomes
Test variations of:
  • Trigger timing
  • Message copy
  • Trigger type (toast vs button)
  • Position
  • Colors
  • High view, low click: Message isn’t compelling
  • High click, low resolution: Agent needs better training
  • Low view: Trigger timing/conditions need adjustment

​
Common Mistakes

Don’t do these:
  1. Triggering on page load - Let users settle first
  2. Too many triggers - Causes fatigue and annoyance
  3. Generic messages - “Need help?” is forgettable
  4. Blocking the page - Never prevent users from continuing
  5. No dismiss option - Always let users close triggers
  6. Same trigger repeatedly - Respect when users say no
  7. Ignoring mobile - Test on all devices
  8. No analytics - You can’t improve what you don’t measure

​
Checklist

Before launching triggers, verify:
  • Timing feels natural, not aggressive
  • Messages are specific and helpful
  • Agent is trained on relevant content
  • Dismissals are respected
  • Works on mobile devices
  • Analytics tracking is in place
  • Rate limits won’t be exceeded
  • Legal/privacy requirements are met

​
Next Steps

Quickstart

Set up your first trigger

Use Case Guides

Real-world implementations