My current theory on this is that companies such as Google probably generate vast amounts of metadata (there must be a fair amount of it behind Google Now), pri

My current theory on this is that companies such as Google probably generate vast amounts of metadata (there must be a fair amount of it behind Google Now), primarily for advertising purposes, and that it's this which PRISM has "direct access" to under some sort of gagging order.