Abstract: Simplex wh-questions have a default existential inference: "Who cheated?" typically implies that someone cheated. The status of this inference is a long-standing debate in the literature. Against the presuppositional view, some authors argue that negative answers (e.g., Nobody cheated) are also question alternatives. To address this debate, I investigate a variety of semantic and pragmatic contexts that influence the presence or absence of this existential inference. An inherent asymmetry in questions must be posited to make sense of the observed variations. These data suggest a duality of wh-words between a referential (or definite) reading and a non-referential (or free-choice) one. These two interpretations have different properties wrt. anaphora, weak islands, NPI licensing, and expectation of answers. Their relative likelihood explains why the existential inference is weak and volatile overall.