The vision of XML was a semantic web. Nowadays everybody knows that semantic is 'em' and non-semantic is 'b' or 'i'. This is simple, but wrong. In fact a notation is semantic when you can make all the distinctions you care about and (this is important) do not have to make distinctions you do not care about. In this case every distinction means something and thus is semantic.
How do you apply this to documents? They are so different. XML gives the answer: you INVENT a notation that suits just your case and use it. This way you perfectly solve the enigma of semantic.
OK, fine, but what to do with my invented notation? Nobody understands it. Well, that is OK. You want to render it as HTML; HTML has no idea about you notation, but is (was) also a kind of XML, so you write a transformation from your notation to HTML. Now you want to render it for printed media: here is XSL-FO, go ahead. Or maybe you want to let blind people read your document too; here is (a non-existent) AUDIO-ML, just add a transformation into this format. In fact there could be lots of different notations for different purposes (search, for instance) and they are all within a single transformation step.
And for that transformation we give you a tool: XSLT.
(I remember a piece discussed here; it was about different languages and one of examples of very simple languages was XSLT. It is my impression as well; XSLT is unconventional, but otherwise very simple.)
Of course you do not have to invent a new notation each time. It's equally fine to invent small specific notations and mix them with yours.
For example, imagine a specific chess notation. It allows you to describe positions and a sequence of moves, giving you either a party or a composition. You write about chess and add snippets in this notation. First, it can be very expressive; referring to a position should take no more than:
Given the party is described this can render the whole board. Or you can refer to a sequence of moves:
and this can be rendered in any chess move notation.
And then imagine a specific search engine that crawls the web, indexes parties and compositions and then can search, for example, for other pages that discuss this party, or for similar positions, or for matching sequences of moves.
XML even had a foundation to incorporate other notations. XML itself is, indeed, verbose (although this can be lessened with a good design, which is rare), but starting from v1.0 it has a way to formally indicate that contents of an element are written in a specific notation. If that direction was followed it could lead to things like:
... ...
all in the same document.
The vision of XML was federated web. Lots of notations, big and small, evolving and mixing. It was dismissed on the premise it was too strict. I myself think it was too free.