Disclaimer: used to work in that sector.
If a poll is biased towards political inclination they can exclude that from the selection criteria to avoid result bias.
Also based on experience there are "shill traps" in the surveys as well.
This is because the non-YouGov polls are quite slow to aggregate data. For example Ipsos Mori is mainly scheduled physical panels and not online (some of it is still paper collection!?!) so any shift in behaviour isn't immediately noticed. It can take a week or so for anything to become apparent.
Opinion changes very rapidly when there are political items on the table. Sometimes within hours. YouGov wins here as it can constantly obtain data almost in a stream-like format and adjust rapidly. They had a thing called BrandIndex that did that a few years ago as well with brand opinions based on advertising and press scraping.
Usually the last minute swings are because there has been a lot of focus on this subject recently in the press which means people are likely to commit to firmer decisions based on research, actually decide to communicate their decisions or are relatively conservative at making decisions.
If you look at how these things are worded, they are always in the form of NO to the status quo and YES to change as well. This means that people less conservative with their opinions will be apparent up front and people more conservative (and usually more intelligent) will lag behind.
The nature of the vote depends on which group goes which way.
Watch these stats over the next few major political events - it's really interesting!