How to check for incorrect answers in the knowledge base before deploying Text's AI customer service?
In the AI Agent Playground, teams can use their own knowledge sources to ask the robot questions, check which materials the answers cite, and then decide whether to make it available to customers.

Choose one from real pre-sales questions, such as whether the same product can be shipped to two different countries. Ask the question using the names of both countries and compare whether the answers refer to the correct shipping policy. Then remove one detail from the policy and observe whether the bot continues to make promises when information is insufficient.
Write down the questions you answered incorrectly, along with their sources. After updating your knowledge, retest using the original questions, and then review again with a different phrasing. This process is suitable for different language sites, as return policies may vary from market to market.


