I think data and metadata is a good analogy. They don't want to control the end content (the data), but want to route users to the content that is engaging (the

I think data and metadata is a good analogy. They don't want to control the end content (the data), but want to route users to the content that is engaging (the metadata, aka knowing the users prefs and knowing what content meets those prefs). Argued to the extreme, there could be infinite variations of content that you could route users to, and the routing algorithm is so good that it effectively 'controls the content' by selecting which exact content to route to, even though technically speaking they are not controlling the content. This kind of hairsplitting is basically the 'have cake, eat it too' situation.