There is a certain beauty to the PRISM idea though. You're right that a capable adversary will already have been avoiding U.S. nets for obvious reasons.
But if you're going to recruit new soldiers, bomb-makers, suicide bombers, etc., you eventually have to go to where the people themselves are. So someone in the terrorist group will have to dip their toes into those dangerous waters.
If a cell is smart that one person will be the only one who does so, but that is still one link into the overall terrorist network, if the NSA can find it, and that can help lead to IP tracking, which can lead to drone surveillance, which can lead to a boots-on-ground raid to grab the computer (and maybe even the recruiter), and go from there.
Obviously the hypothetical "capable adversary" can be resistant to a lot of tracking techniques, but that doesn't mean you give up. They may very well make a mistake eventually and then you've got them.
The NSA's precursor once even cracked part of a one-time pad because the Soviets reused individual pages from their one-time pad codebooks. Breaking the parts of the code they did took lots of time but by 1946 revealed the existence of Soviet spies in many high-level organizations, including the Manhattan Project, which was a pretty significant coup of its own for U.S. counter-intel.
And all that by pursuing an enemy using theoretically unbreakable crypto...