You're right, that can happen due to the limitations of AI. However, you can implement post-processing routines that can modify the raw response from the model to make it seem more human-like. You can add variability, correct any biases, and even modify the grammatical structure to make it read more naturally.