Evaluate AI knowledge sources in Text Playground before launch
Ask questions, inspect the knowledge behind replies and identify gaps before opening multilingual AI support to visitors.

Choose a shipping policy with exceptions. Ask about one covered destination, one explicitly excluded destination and one the policy never mentions. Write an acceptable answer for each first, then check the AI’s source and any delivery promises.
Keep the question, language, expected conditions, actual answer and source version. If an old notice was used, update it and rerun the same cases. Test different return windows by market to catch a policy applied too broadly.
Fix mistakes affecting payment, delivery or refunds before live traffic. Use recurring unanswered questions to prioritize new help articles, without duplicating a document for every wording variation.


