A lot of people moved from MongoDB back to traditional RDBMSs like MySQL because most data is in fact... relational.
There's probably a good use-case for MongoDB, but what I saw in practice was that it was an excuse for poor developers to avoid learning some dialect of SQL.
It's the same story with ORMs. You will eventually need to learn to write SQL. For some reason developers try to avoid this almost as much as possible. Maybe more than other technologies.
At some point, you are going to want to query data. And eventually you will reinvent what RDBMSs do out of the box, because they're purpose built for this work.
Further, if you need to just dump JSON, an RDBMS is still a better choice for all the reasons above.
What does it mean, in your view, for data to be relational? I think there's some circular logic with this take that boils down to whether you believe you need to conceptualize your data model in normalized (storage optimized) or denormalized (optimized for reasoning). I dislike these terms like normalize/demoralize as much as relational since it hides so much intuitive understanding
Ignoring the academic point of view for a moment, if you have any sort of table of related data, surely, they have some sort of relationship or queryable attributes.
Find me all users with X... (SELECT * users WHERE X,...).
In practice, I've found that you don't really query documents. Can you? Sure, I suppose, but I found myself reinventing what those databases do for me.
Sorry, I know that's not a formal explanation. But it is a practical day-to-day one. It's easier for me, and I think others too albeit with practice, to simply write a line of SQL to get what I want.
It has a bit of a hokey syntax. The original syntax was eschewed in the 70s because it looked too technical. The hope was it could be used directly by non-developers. Of course, that never happened so we are left with this somewhat peculiar prosy compromise.
It is not difficult. It is just not in fashion to learn SQL directly. I can't tell you how many times I have interviewed "full stack" devs who cannot explain a simple join.
There's probably a good use-case for MongoDB, but what I saw in practice was that it was an excuse for poor developers to avoid learning some dialect of SQL.
It's the same story with ORMs. You will eventually need to learn to write SQL. For some reason developers try to avoid this almost as much as possible. Maybe more than other technologies.
At some point, you are going to want to query data. And eventually you will reinvent what RDBMSs do out of the box, because they're purpose built for this work.
Further, if you need to just dump JSON, an RDBMS is still a better choice for all the reasons above.