AI
Test & Optimize AI
💬Get free consultation

Test & optimize AI

How to test AI assistant

AI test zone helps simulate customer interactions before going live, ensuring your AI responds appropriately to customer inquiries.

Before AI assistant test

Can I test AI assistant without enabling it?

You can still test AI assistant with data sources that you added without enabling AI assistant.

Go to test zone

Go to AI agent Go to Test AI

You can also click Test AI in the top right of any AI agent page. It opens the same test chat in a side panel, so you can try a change without leaving the page you are editing.

Chat with your AI

In chatbox, ask and interact with AI assistant as a customer. You can use question examples to ask AI. Pick a topic from the dropdown above the examples — Store Info, Product Recommendation, Order Issue, Shipping, Return/Refund, or FAQ — to load a different set of questions. Click the reset icon in the header to clear the test conversation and start over.

Tips for better testing:

  • Test one topic at a time
  • Ask the same question in different ways
  • Make sure responses aren't too long or too short

Check the responses and make improvements

What to check in AI response:

  • Is the information correct?
  • Is the tone friendly but professional?
  • Are product details accurate?
  • Are prices and availability correct?

You can also review what data sources your AI uses to answer customer questions in test zone.

AI completion score

The AI completion score tells you how well your AI handles customer questions, so you can see where to improve. It is calculated from the past 30 days of conversations only.

AI Completion Score = (0.7 × RR + 0.3 × MRR) × 100%
  • RR (resolution rate) = AI resolved conversations / total AI conversations × 100
  • MRR (merchant review rate) = unresolved questions reviewed / total unresolved questions × 100. With zero unresolved questions, MRR is 100%.
ScoreBadge
Under 50%Red bar, Need optimize
50–80%Orange bar, Good
Over 80%Green bar, Optimal

Two ways to raise it: resolve more conversations — mark AI-handled chats as Resolved in the inbox, or turn on SettingsAutomationAutomatic resolution — and review the unresolved questions below.

How to optimize AI responses

A. Unresolved questions

Unresolved questions are customer inquiries that your AI Assistant couldn't answer effectively.

The page counts them for the past 30 days and splits them by reason, so you can see what is really going wrong:

ReasonWhat it means
Missing knowledgeYour training data has no answer for the question
Customer request humanThe customer asked to speak to a person
Negative sentimentThe customer was frustrated or angry
Configured handoverOne of your own handover rules sent the chat to your team

We only show unresolved questions in the past 30 days

How to optimize unresolved questions

Go to Unresolved questions

Check unresolved questions

What you should do with unresolved questions:

  1. Look for patterns in unresolved questions to identify gaps in your training data
  2. Group similar unresolved questions to create comprehensive answers that address various phrasings of the same inquiry
  3. Use unresolved questions to identify new FAQs to add to your help center

Add answers to unresolved questions

Select a question from the unresolved list and click Add answer → Enter your answer.

Questions and answers will be added to data sources. (in Question list)

Test again

Go to AI agentTest AI and ask the question to ensure that AI assistant can handle that question now.

With duplicate or spam questions, you can click Ignore to delete the question from the list.

B. AI responses review

Each AI response now includes a "Review sources" button to see what information was used

This helps you understand why the AI responded the way it did.

Use this feature to:

  • Add missing product information the AI should have included
  • Correct inaccurate policy details
  • Expand on partial answers that need more detail
  • Add variations of common questions that weren't properly understood

What you should do

  • Identify and correct inaccurate responses
    • When you notice incorrect or incomplete answers, click "Add relevant data"
    • You'll be redirected to the data sources page to add the appropriate information
  • Watch for patterns in corrections
    • If you're frequently adding the same type of information, consider creating comprehensive entries about that topic
    • Group similar corrections into comprehensive answers to improve overall AI performance