The thing that really annoys me is when the question doesn't match the way the data's used. My ISP asks for a rating out of 10 for "how likely would you be to recommend <ISP> to family and friends" after every support call, but the phone operators explicitly tell you that they're rated on the score that you give them. Those are two different questions - I'd like to rate the phone operator highly, the back-end technical administrators as average (they once assigned the same IP address to myself and another customer - it's disappointing that this is even a manual process!) and the ISP as a whole as 'could be improved' due to their lack of upstream bandwidth at peak times.