I don't agree with the conclusion; 3% difference is basically within the error margins for these types of things. I typically assume a ~5% error margin, which is nice round number I just pulled out of my ass, but it seems roughly accurate.
No, you're misreading too - see my other comment.
But in that case you need to add a control for both questions without the prior four questions. How do you know the difference in phrasing doesn't matter?
This proves the idea that you can get a result that points the way you want it to. But it doesn't quite prove that the leading questions have any effect.
To do that they would need to ask 2 more sets of questions with the "oppose"/"support" questions inverted. Otherwise the effect could just be due to the different wording of the final question.