Well, messaging history is just an example; I'm not actually sure if that is something which apps are allowed to lift, but I figured I'd err on the extreme end.

Either way, discrete and easy-to-digest information like 'likes' are exactly what they wouldn't have access to from Google, and honestly it seems like automated sentiment analyses are still very error-prone; large corpuses of text don't seem ideal for this sort of targeting.

Graphs that connect users are not all the same. For example, it's much more common to follow relatives and people you know in real life on Facebook than on Twitter. (This is one of the reasons Google tried to build Google+: they were worried that not having a real social network would put them at a disadvantage.)

It's all privacy leakage but it's not at all the same data. Facebook's graph is still noisy but it's better or at least different data than other companies have.

Was the personality profile any good? Otherwise I could make that claim about extracting a personality profile out of tea leaves.

Seems to me they could make more fine grained permissions. E.g., I don't like that my javascript disabler Firefox plugin has permissions to 'Access data for all websites', 'Access browser tabs'.