media types / content types are far less "how do i read this" (data serialization) and a lot more "what do i do with this" (processing model) and "what do keywords mean" (semantic model)

when you say application/json your processing model is "json object/array", but when you say application/activity+json your processing model is (or should be) "activity streams". if you say application/ld+json the processing model is "linked data". if you fit multiple processing models, signal the intended one