Jesus Christ.
Silver's whole #AllDataPointsMatter nonsense is getting out of control. Adding obvious junk polls under the presumed assumption that an equally polarizing and junkish poll will come along later to balance out the overall result is silliness.
I disagree. He's using them to tack onto a blended sample of data for the state when he runs his projections. He's well aware of the MOE and the tendency for the poll in isolation to be suboptimal, especially with the tiny sample sizes.
The polls are instructive as data points when added to other data points and probably more instructive than Reuters/Ipsos and Washington Post/Survey Monkey as the data is over a week span as opposed to a 1-3 week span (Reuters), or 4 week span (SM).
You could recreate a more representative sample by compounding the surveys into 1, 2, 3 week surveys if you wanted to like Reuters and SM does.
You can likely extrapolate the Reuters/Ispos numbers over a week if you wanted to at this point, but solely if you had access to the underlying data. Reuters doesn't give us this. Google does, down to the individualized weights of the individualized survey respondent.