LoL - I did the same and tested an AI-chatbot. Same results you published here.

From the perspective of the human questioner, if he expects to be impressed, he does not bother to ask the same question in repetition. He might tend to take the answer given as true whenever he is not able to find out the answer himself. I agree, you have to ask for the specifics and the path, the chatbot took to come to a result.

To provide my own answer to the Sally question:

Brother 1: 3 sisters
Brother 2: 3 sisters
Brother 3: 3 sisters
Brother 4: 3 sisters

Answer: Sally has 2 sisters.