Zulip is implementing the same bad architecture as google/ms/apple/fb's messengers (custom protocol, no federation), but at least it does it well (open source, usable, do-one-thing-well, no dark patterns, powerful and stable API).
But that said I don't think "using a standard protocol" is a value in and of itself.
Standard protocols are very valuable if you care about reach while avoiding vendor-lock-in. I want to be able to self-host my server (or pay someone to host it for me), but I also want to be able to communicate with people on other servers. It is that balance of control and networking that is really powerful. Someone else mentioned here about how FHIR uses Zulip as the group chat for their community. That is great! But, to participate, you need an account on their instance and then if you want to participate in multiple Zulip instances you either need multiple clients or some kind of multi-instance support in your client. And then you still have to manage separate profiles for each instance, etc, etc...
OTOH, a bunch of open source projects use Matrix instances for their community chat rooms. One of the huge benefits I see from that is that I can use my single Matrix account (on the server that I host) to join all the different communities that I am interested in. (And now with the new Spaces feature in Element, there is even better controls over being able to still retain a separation of concerns between all the different conversations I am a part of...)
That's what the X in XMPP is for.
Matrix.org is a great choice for an open federated messaging protocol, but the challenges of operating in that kind of ecosystem have certainly stunted their ability to add features and polish when compared to a closed system such as Zulip.
Is this question just opening myself up to a cacophony of "what was wrong with IRC / XMPP?"
Generally speaking, use the standard until you can't. Then just extend it and write up a proposal for your extension to make it into the next version of the official protocol.