Haslanger (Noûs 34(1):31–55, 2000) has argued that we should ameliorate concepts of race or gender to better capture existing structural inequalities. Her analysis was criticized by Simion (Inquiry 61(8):914–928, 2018a), who argued that a concept should be ameliorated only if doing so preserves epistemic accuracy. But, as I argue, this criticism misses Haslanger’s target. In response, Podosky (Inquiry 1–15, 2018) and McKenna (Logos Episteme 9(3):335–342, 2018) have argued that conceptual revisions need not preserve “epistemic accuracy” since concepts can “shape reality”, not just represent it. Here I argue that social kind concepts need not “shape” the corresponding social kinds: neither causally nor constitutively. This raises doubts about the function of ameliorative projects for social kind concepts as means to change social reality.