I really think there is absolutely no evidence that "extremist content" is a serious problem that would warrant infrastructure to moderate or censor content. What if we try to quantify the problem? How many people are really affected? And in which direction?

I think a closer inspection would quickly reveal that it is not a particular video or text that drives people to the edges and has more to do with the general reaction to terrorism for example.

"Extremist content" could very well be the rock'n'roll of the seventies.

And wrong people are already getting banned. And not for being "extremists".

Seriously, if wiser people just decided to completely drop the topic immediately, the problem which still lacks quantification, would be nearly invisible.