http://fivethirtyeight.com/features/is-a-50-state-poll-as-good-as-50-state-polls/As Nate Silver explained before, state-specific poll is better than a huge national poll that surveyed all 50 states and gives the state result separately.
One potential problem being:
Oregon whites are more liberal than the average whites in America.
Mississippi whites are more conservative than the average whites.
If you try to weigh Google Survey to match the national average white's probability of voting R and D, it would likely overestimate Mississippi's Democrats share and underestimate Oregon's Democrats share. But these kinds of things cancel each other in a holistic national sample.
State-specific polls are useful in the sense that they can focus on that one specific state's race, age, gender, and education.
That's probably why we see some odd things like Hillary being up in red states and Trump being up in blue states.