I think the languages you selected in your final remark sum it up for me. If one is truly taken by functional programming, much of their mental model starts to revolve around algebraic (inductively defined) data types and structural recursion (aided by pattern matching) over them - such that mentally reducing the set of candidate languages really does become an implicit process of questioning: "does X have ergonomic, statically-typed, discriminated sums?".

Lots of mainstream languages simply fail this test and make it feel like intellectual poverty or that there's extreme, turgid, boilerplate required for a weak imitation of the features (see the idiomatic class-hierarchy encoding of ADTs in large projects - such as LLVM - in C++, for example).

It's absolutely no surprise that more mainstream languages are picking up a match-like construct and lighter encodings of discriminated sums. So, it really comes to what else you wish to be burdened with when compiling an OCaml-like mental model to X in your head: Tagged unions a-la C? Class hierarchies for ADTs in C++? The travesties of std::variant? Monad transformers in Haskell? Lazy evaluation? No static typing at all? Caring about memory management and ownership? Box and Arc-ing recursive components of ADTs? Writing your own arena allocator? Compiling to the JVM? Spotty TCO support?

OCaml is just a nice, fairly simple (at its core, at least), language that captures the essence of the ML family, compiles to native (and bytecode and, transitively, JavaScript), has great tooling (opam, dune, ocamllex, memhir, etc.), great libraries (official LLVM bindings, for example), and a great community. Lots of OCamlers are well aware of other potential languages that somewhat suit their style of programming, they just don't want to be burdened by the other stuff.