A poll can be a valuable indicator of how an election will probably go. A poll can also be wrong. The day before the big election in 2000, Zogby said the New York Senate race was too close to call and predicted "I think we're looking at a one point race." The next day, Hillary Clinton beat her opponent by twelve percentage points.
News sources don't like to say that any race is over because they think you'll stop paying attention to them. So if there's any possible way to claim that a race is tightening, they say that. But if the poll has a four point margin of error, a shift of two points is meaningless. If someone is working for a party or a candidate, anything they say about their internal polling is especially meaningless.
The other thing you shouldn't fall for is the self-serving reason for victory or defeat. If I was interested in reducing the capital gains tax, I think I'd be out there now saying, "A Democratic win on Tuesday will be a mandate to reduce the capital gains tax." Matter of fact, I think I'll start spinning for something useful. You know, a Democratic victory on Tuesday means that America demands more Groo comics. I'll bet that if our publisher bought ad time on the cable news networks, I could get some pundit to echo that idea. Even if they don't, no one can prove me wrong. And it makes as much sense as some things we'll be hearing this week.