This can be adjusted by moderation. Why not have the chatbot be compliant at first, and then increase the moderation level as the relationship grows and the lock-in effect occurs?