This is like saying we shouldn't use TCP/IP because it's not encrypted. How it actually works is that encryption is enforced by the application - indeed the only place you can reasonably enforce it. See for example the gradual phasing out of HTTP in browsers by various means.
What this means in practice is that you shouldn't focus on whether XMPP (or Matrix, or whatever) protocols are encrypted, but whether the applications enforce it. Just as there are many web browsers to choose from, there are many messaging apps. Use (and recommend) apps that enforce encryption if that's what you want.
I'm not sure I agree, particularly given that there's some incentive for us to get our relatives using these messenger protocols and clients. The Web made it work because everyone came together and gathered consensus (well, modulo some details) that enforcing HTTPS is, ultimately, a good idea given the context.
So far, I'm not seeing that same consensus from the XSF and client vendors. If the capital investment can be made to encourage that same culture, the comparison can perhaps be a little closer.
The consensus comes from the people using the clients, not from the standards bodies. It's the same for HTTPs, where the users (in this case the server admins) decided it would be a good idea to use encryption.
There are even apps like Quicksy which have a more familiar onboarding experience using the mobile phone number as the username, while still being federated with other standard compliant servers. There is little reason to use walled garden apps like Signal these days.
This is not true, it depends on the client. Conversations has OMEMO enabled per default.