Earlier quoted context omitted.
Maybe not more data, but perhaps more important data? "Things you've bought before or looked into buying before" seems like much more useful data to sell ads than "non-shopping things you've searched for." It's definitely a higher signal to noise ratio.
If your Amazon receipts go to Gmail, Google collects not just all of that data, but from every other site you've bought stuff from: https://www.cnbc.com/2019/05/17/google-gmail-tracks-purchase...
https://www.nytimes.com/2017/06/23/technology/gmail-ads.html
Any aggregation or segmentation of emails by ML/AI algorithms, as your article talks about, is analogous to having smarter filters in your inbox. It's tech illiteracy to confuse these perception NNs with data mining for business purposes.