AI Test

AI Test is where you check your assistant's answers by asking it sample customer questions. Whenever you add new information or change behavior settings, you can try the results here.

What Is It For?

You can inspect whether an FAQ is used correctly, whether product information is sufficient, and whether the assistant answers in the tone you want. Based on the test result, edit the relevant information or instruction and try the same question again.

AI Test is not a separate publishing approval

AI Test conversations are for experimentation and are not treated the same as real customer conversations. However, the training and behavior settings you save can also affect your live assistant.

Getting Started

Prepare your information

Add the business information the assistant needs and complete the training steps.

Open the AI Test section

Open the AI Test section in the panel.

Check your test allowance

Check your remaining test allowance.

Send a question

Send a question a customer might write.

Review the answer

Check the answer's information, language, tone and any missing conditions.

If the message area is unavailable because training is not complete, return to the AI Training section. Simply uploading a file may not be enough; make sure the prepared information has been reviewed and saved.

Test Allowance and Message Limit

A separate monthly allowance is provided for AI Test. Test messages are tracked against the allowance reserved for this area; they should not be confused with your normal customer message quota. You can see used and remaining allowances from the counters on the screen.

Your test allowance depends on your plan and account rights. Allowances renew monthly. A single test message can be at most 1,000 characters.

Resetting the test conversation does not mean renewing your monthly allowance.

Which Questions Should I Try?

Real customer questions: Ask using the phrasings customers actually use.
Different wordings: Try the same question again with different words.
Follow-up questions: After the first answer, ask a related second question.
Missing information: Ask for information not in your records; observe how the assistant behaves.
Conditions and exceptions: Ask about details in price, campaigns, delivery and service scope.
Language: Try first messages in different languages and explicit language-change requests.
Human support: Test the live agent request separately from a request for a future appointment.

You can also try these scenarios using sample information instead of real customer data.

Interpreting the Results

Evaluate indicators such as average response time, knowledge base usage and session test count on the screen together with the answer text. An item appearing to have been used does not mean the entire answer is correct.

If the answer is insufficient, fix the relevant record or instruction, save it, and ask the same question again. If needed, reset the conversation to also check the first-message behavior separately.