GH et al:
Polls are handicapped. It is costly to conduct a poll that is broad enough to use its statistics to infer to a population. There are a few organizations that do large scale polling. But the cost of doing this cannot be easily borne by small organizations, and some of the media reported polls fall into this category.
Next, a representative sampling is a huge challenge. Polling in a blue state will say less than enough about an election where the match up of the two parties is the subject. There is also an assumption that the individuals being polled know enough, but perhaps not too much about the compare and contrast between the candidates. There are many who respond to pollsters by telling them what they wish to hear, or simply echoing what they have been hearing within their circles of communication. That might not reliably project how that same individual would behave in the voting booth.
In summary, the non-scientific polls are completely meaningless. Even the larger ones do not have the needed reliability to be predictors. They feed discussion, but talking about sports is at least as useful.