The problem is that a survey result is a pretty meaningless number if it can be changed by anyone you put on television with a fancy suit and a pie chart. What you really want to know from a survey is a) what the people who feel strongly think about it, i.e. the people who can't be swayed by an ad buy or a newscast, and b) what portion of the population are those people and what portion are the ones who can be easily swayed. But if people are overestimating how strongly they're committed to their answers then they're contaminating both results: The second directly and the first indirectly by inserting the uninformed answers of the disinterested into the sample that ostensibly measures those with strong feelings.
This may be a semantics thing, but surveys are extremely effective - just not as a tool for understanding opinions.