I'm pretty sure I remember that one of the original leaks showed that they keep metadata on Americans roughly indefinitely, and that they can keep full data (i.e. not metadata) for up to 7 days in order to determine whether or not the individual or their actions can be considered a threat. The reason I remember this is that I immediately thought that building a google-like index for each individual, and storing that index indefinitely, would be equivalent to what was stated in the leak. In fact, given all the subsequent leaks, I'm pretty sure that's exactly what they are doing.