Well, I don't know. A typical HN reader should not be surprised by this. You know technology, you know how Google operates, it should be fairly obvious that they store everything, forever.
When I saw the article, my reaction was "hey, thanks for letting me know what the link actually is". I was not surprised at all by it. I sort of expected it.
I assumed they did something to improve matching, but I never imagined they stored the 'raw' (as in actual voice, albeit probably compressed) recordings.
You might know something when thinking about it but never have been aware of it. "Oh yeah, I guess all this time they must have been doing that."