HTTP is named Hypertext Transfer Protocol after all. Exchanging schema-less JSONs is conceptually pretty lame. It's plain old RPC architecture. That's why we end up needing to write API docs to line up the frontend and backend, or otherwise inspect backend code to figure out what's being handed over.
Browsers are built to understand HTML. Whereas in the modern world we largely ignore that and write double the code we'd have to write otherwise passing arbitrary messages and it introduces a whole new set of problems making sure those codebases line up properly. Whereas HTML produced by the backend (properly) contains all the information someone could need.