This is very much my experience too. There is also a very high variance in quality of discourse in face-to-face situations.
I think it's slightly easier to have moderate-to-high quality discussions in asynchronous online writing (assuming that's what the participants want), because you can treat stuff-you-can-Google-easily as an assumed baseline of knowledge and competence.
A silly idea I have is to model the quality of conversation as a random walk. With no boundary, you will almost-surely sink below the YouTube Comment Event Horizon as time passes. But if you have Wikipedia as a lower bound, the average quality of discussions will tend to increase over time.
When I was younger, I thought that conversations in real life were much more likely to promote true beliefs and meaningful changes than conversations online, because people in real life were only willing/able to cite evidence they were actually confident in, while those online were able to easily search for arguments favoring their position.
While this is obviously wrong—the concept that people in real life only cite evidence they are justifiably confident in is comically false—I do think the dichotomy illustrated there is interesting. One thing I've noticed is that in general the "rigor" of discussions online is higher (in terms of citations, links to external content, etc.), but that conversations in real life seem still much more likely to actually change people's minds.
I have noticed this effect in both myself and others—what do you think is going on here, and how do you think we might circumvent it? If online discussions could be made more effective at causing people to actually change their minds, this could potentially prove extremely useful.