We investigate a specific subtype of implicitly abusive language, focusing on non-negative sentences about identity groups (e.g.Women make good cooks).We introduce a novel dataset comprising such utterances.It not only profiles abusive sentences but also includes various semantic variants of the same characteristic attributed to an identity group, allowing us to systematically examine the impact of different degrees of generalization and perspective framing.Thus we demonstrate that specific variants significantly intensify the perception of abusiveness.By switching identity groups, we highlight how the characteristics often described in stereotypes are not inherently abusive.We also report on classification experiments.