You could implement the recommendation backend so that it can be accessed from feed readers, server-side frameworks, client-side widgets, and browser extensions. This way you can also gather data from all these sources to base the recommendations on. Perhaps there's a revenue-sharing scheme to motivate third parties.
I guess finding similar blogs would take more long-term data whereas finding similar posts should react very quickly. I'm specifically thinking of the case where two blog posts are both linked to by a third post: those two could then be considered similar. Comparing the frequency of some key words mentioned in the posts would be another way to get started, as would be looking at the audiences.
Whatever the method, it should provide results that users find meaningful in meeting the need they have.