Test chat and supplementary answers

AI customer service isn't something you configure once and forget. Try it before launch, and keep improving it after launch—your answer rate will keep rising.

Test chat

On the "Test chat" page, role-play as a customer and talk to the AI to check whether its answers are accurate. You can:

• Enter a customer ID to simulate a logged-in customer and test order lookup (this ID will be used to call the order API you configured).
• Enter a customer IP to test language detection based on location. For example, enter a German IP and see whether the widget defaults to German.
• Say "I want to talk to a human" to test the handoff flow.

Test chats count toward usage just like real conversations.

Pending questions

Questions the AI can't find answers to in your materials are automatically recorded under "Pending questions." For each one, you can:

• Add an answer: After you save it, a Q&A entry is generated automatically so the AI can answer similar questions next time.
• View conversation: See how the customer asked the question at the time.
• Ignore: Ignore questions unrelated to your store.

We recommend checking once a week and prioritizing the questions asked most often.

Common causes

When the AI gives poor answers, it's usually one of these situations:

• Missing from your materials: Add Q&A entries or documents.
• Materials are too vague: Spell out conditions and numbers clearly.
• Old and new materials conflict: Delete the outdated version.
• The question needs real-time data (orders, inventory): Connect order lookup or sync products.