These shifts were not outside the margins of error in the Reuters/Ipsos and Echelon Insights surveys, while the Times’s pollsters said that they could not calculate such a margin for their recontacting survey. Nevertheless, the fact that the same shift was recorded across three different surveys lends credence to its validity.

There’s two interesting things the article points out.

1. These were recontact polls. So these were people who were already polled, presumably a small percentage changing their minds.
2. These were multiple pills across different organizations running them. So while they can be within the margin of error, they all saw the same trend.

Obviously you can’t go by polls, and there’s a ton of time between now and Nov, but it’s interesting nonetheless.