My prompting is conservative to err on the side of playing an ad if there is a chance it might be part of the actual content, not really getting false positives at all yet. That being said while still in development I haven't reached the stage of running on a huge collection of podcasts to get more representative statistics.