Of course they don't record and listen, that would be slow and take huge stores of data. They save transcripts and read/parse it instead. Much faster and less data to store.