Modern email spam detectors definitely read the contents of the email and even try to extract context through certain words appearing together.

Scanning to apply a filter (Machine-taught, or user-defined) definitely does.

What does it mean to extract the content?

To filter spam, you need to make enough sense of the email to identify it as spammy, according to some algorithm. To serve an ad, you need to make enough sense of the email to indicate it's about something one or more of your advertisers cares about. I can't think of a way to distinguish between the spam robot's access vs the ad robot's.