When the model produces reasonable results from one prompt, you could assume that it will also return reasonable results through the follow up prompts.