The overall result is probably not very accurate but if the methodology has not changed then you could try to gauge something based on change from poll to poll.
There was an interesting methodology I read about. Basically, they read 4 statements and asked how many (not which!) the interviewee agreed with.
Then they did the same thing with 5 statements, where the 5th was what they actually wanted to find out. With a large enough sample size on both and the power of math, they can essentially deduct test 1 from test 2 to find out how many people agree with the 5th statement without anybody outing themselves to the FSB.
It is called the "unmatched count technique" or "list experiment." It has a wider error range, so you need to poll more people, but you get honest answers.