The assistant is more careful about what it claims to know, and more useful when a message asks for two things at once.
- When a customer asks about a specific topic such as refunds or make-up classes, the assistant answers directly from your own written policies only if you have actually written about that topic. If you have not, it looks the answer up instead of guessing.
- A general question such as "what are your rules?" is still answered straight away from your written material.
- Rules you wrote for one specific group stay out of general answers.
- A message that mixes a question with a request, such as asking about the cancellation policy and also asking to cancel a class, now gets both: the answer and the cancellation.
- When you ask why a customer did not receive a notification, the assistant now states plainly that no matching attempt was recorded, that this alone does not explain why the notification was never created, and separately whether that customer can receive messages right now - without guessing at a cause.
- Wording about LINE and app connections now matches the words used on screen, including for app devices.
