I'd guess that maybe 1% of connections in Ancestry trees are incorrect, later than 1800, and more before that.
A big part of the issue is that Ancestry suggests people add more generations to their tree from others. So once one person makes an error, then it gets copied by people who don't know how to research properly.
Not long ago a cousin of my father popped up as a match. I sent him a lengthy file based on proper research. He promptly ignored it went about adding a bunch of Ancestry suggestions to his tree, including a number of casual and catastrophic errors.
When you meet such people in life you may hear things such as "Oh, I've researched the family tree." When in reality they've just copied "research" without question. Other common lines are, "We go back to the 12th century", ... William the conqueror or Charlemagne... Maybe. But the paper trial is more likely to be dubious.
I am very careful with making connections with DNA, because it's easy to make connections where there is no overlap with trees and DNA, if you don't know what you are doing.
The worst one I have found in my tree is an ancestor Janet Hamilton was born c 1793 in Ayrshire, Scotland. Her father is given as Alexander Hamilton, b. 1786 in New York City, who is in turn the son of Alexander Hamilton, 1st Secretary of the Treasury of the US. This has been copied into many trees. But on her death certificate, which costs about £1.50 to order, the parents are listed as something else.
It's also common to see many trees with mad scrabbles back though the 16th-18th centuries. Maybe there is a baptism in Cornwall, the line then goes to Maryland, US, then Lincolnshire, and before ending in South Wales.
I also find that US immigrant lines back to the UK tend to be incredibly unreliable.
But from the perspective of DNA the trees are immensely useful. If Ancestry didn't have this vast repository of often shaky trees, I would be missing a few lines I have solved with DNA. You just need to know the pitfalls and what is possible with paper and DNA trials.
Personally I don't even use Thru-lines. To be sure of the information you can extract from DNA matches, you need to go through all of them down to 20cM and more of the shared matches below 20cM. If you have ancestors in a concentrated area, clusters of matches can become jumbled due to overlapping on more than one line. For a long time I thought that one of my lines was via an NPE. But when I systematically went through all matches I found that I had two lines where almost all of the matches shared 2+ lines of ancestry.
