I think it's important to consider scale when discussing metadata vs raw content

Getting high precision is difficult when done on "all content", especially considering the multitude of languages and dialects in Europe. At a certain point, more data does not result in better output, sometimes it's actually detrimental

The patterns that arise from metadata are much more generalisable and there's less of it, so it's easier to search through

Metadata interesting lets you hone in on what you want to inspect full content on. It's definitely an upgrade

The article is a bit older, but metadata analytics can be used to "unveil" the key conspirators that led to what we today know as the "American Revolution"[1] - please note how little data they use to find out who calls the shots, and whom you might want to disappear if you wanted to keep the colonies closely aligned with George III.

Widespread enough, this can be used to find dissidents today, suppressing ethically legitimate uprisings against injustice.

[1] https://blog.wolfram.com/2017/06/29/analyzing-social-network...

Metadata is highly structured, standardized, and easily understandable by machines. Content is not.

Because processing is easier for metadata but analysis of it can answer many questions.

Direct communication is only interesting for direct surveillance.