Do students who use AI and then humanize produce better or worse final work than students who don't use AI at all?

genuine pedagogical question not a rhetorical one

i’m trying to form an evidence-based view on this. the policy question (should students use AI) is separate from the outcome question (what does AI use do to the quality of student learning and work). i’m more interested in the second one.

does the process of generating AI output and then editing it toward human-sounding prose develop any useful skills? or does it mostly produce students who are good at editing and poor at generating original ideas? has anyone seen or studied this?