Skip to main content

The Problem

Traditional support channels create friction:
  • Live chat requires staffing and has limited hours
  • Ticket systems are slow and impersonal
  • Knowledge bases require users to search for answers
  • Users abandon when they can’t get immediate help

The Solution

Deploy the BoostGPT chat widget as an always-on AI support agent that answers instantly, 24/7, trained on your product knowledge.

Implementation

Step 1: Train Your Agent

Before deploying the widget, train your agent on your support content:
  1. Go to Memory in your agent dashboard
  2. Upload your knowledge base:
    • Product documentation
    • FAQ pages
    • Pricing information
    • Policies (shipping, returns, refunds)
    • Common troubleshooting guides
Start with your top 20 most-asked support questions. You can always add more content later.

Step 2: Configure Agent Instructions

Set up your agent’s system instruction to handle support conversations:
You are a friendly customer support agent for [Your Company].

Your job is to:
- Answer product questions accurately using your knowledge base
- Help users troubleshoot common issues
- Guide users through setup and configuration
- Provide pricing and plan information
- Escalate complex issues to the human support team

Rules:
- Be friendly, professional, and concise
- If you don't know the answer, say so — don't make things up
- For billing disputes or account issues, direct users to support@yourcompany.com
- Always confirm you've resolved the user's question before ending

Step 3: Install the Widget

Add one line to your website:
<script src="https://embed.boostgpt.co/widget.js" data-bot-id="YOUR_BOT_UUID"></script>

Step 4: Configure Greeting Messages

Set up welcoming greetings in Branding → Content → Message Greetings:
"Hi! I'm your AI support assistant. Ask me anything about [Product]."
"Need help getting started? I can walk you through setup."
"Questions about pricing or features? I'm here to help!"
These appear as stacked cards above the bubble, inviting users to start a conversation. If your site has user authentication, connect SSO to identify users:
// On your backend: generate a signed JWT
const token = jwt.sign(
  { email: user.email, name: user.name },
  'YOUR_SSO_SECRET',
  { expiresIn: '24h' }
);

// Pass to widget
boostgpt.init({
  botId: 'YOUR_BOT_UUID',
  userToken: token
});
Benefits:
  • Agent greets users by name
  • Conversation history persists across sessions
  • Support team can identify who needs help

Step 6: Add Proactive Triggers

Combine the widget with triggers to proactively offer help at friction points:
// Help page visitors who seem stuck
setTimeout(() => {
  boostgpt.trigger('help_page_stuck', {
    page: window.location.pathname,
    time_on_page: 60
  }, {
    text: 'Looking for something specific? I can help you find it.'
  });
}, 60000);

// Pricing page assistance
if (window.location.pathname === '/pricing') {
  setTimeout(() => {
    boostgpt.trigger('pricing_help', {
      current_plan: getUserPlan()
    }, {
      text: 'Questions about our plans? I can help you choose the right one.',
      delay: 30000
    });
  });
}

Widget vs Triggers — When to Use Each

ApproachBest ForExample
Widget (always visible)General support availability”Chat with us” bubble on every page
Triggers (proactive)Specific friction points”Stuck on checkout?” when user hesitates
Both (recommended)Maximum coverageAlways-on support + proactive help
The widget and triggers work together — the bubble is always available, and triggers add contextual messages when they detect friction.

Measuring Success

Track these metrics to evaluate your support widget:
MetricTargetHow to Measure
Resolution rate80%+Conversations resolved without human handoff
Response accuracy90%+Spot-check agent answers weekly
User satisfaction4/5+Add rating prompt after conversations
Support ticket reduction30-50% decreaseCompare ticket volume before/after
First response timeUnder 5 secondsAI responds instantly

Optimization Tips

Check the Conversations tab to find:
  • Questions the agent couldn’t answer (gaps in knowledge)
  • Incorrect or confusing responses (needs instruction refinement)
  • Common topics (may need dedicated documentation)
Start small, then add content based on real conversations:
  1. Launch with core FAQ content
  2. Review unanswered questions after 1 week
  3. Add missing content to memory
  4. Repeat monthly
Enable web access so your agent can fetch live data:
  • Current status page information
  • Latest pricing from your website
  • Product changelog and release notes
Not everything should be handled by AI. Configure clear escalation:

Next Steps

Widget Configuration

Customize appearance and behavior

SSO Setup

Identify your users automatically

Event Triggers

Add proactive messages

Best Practices

Tips for effective deployment