A notion of truth as applicable to events of assertoric use (utterances) of a sentence token is arguably presupposed and required by our evaluative practices of the use of language. The truth of an utterance seems clearly to depend on what the utterance says. This fundamental dependence seems in turn to be captured by the schema that, if an utterance u says that P, then u is true iff P. Such a schema may thus be thought to constitute a suitable basis for an adequate theory of utterance truth, so much so that it seems straightforwardly to avoid the problems arising from context dependence and the semantic paradoxes which notoriously beset theories of utterance truth based on a simple disquotational schema. The paper argues that appearances are deceptive in both cases. On the one hand, the schema cannot allow for plausible if not uncontroversial non-indexical forms of context dependence, arising from the possibility that what an utterance says can be the case or not relative to different situations and that the truth of an utterance u of a sentence ϕ arguably depends on the truth of ϕ at the situation “associated” with u. On the other hand, a quantified utterance-truth variation on the liar paradox shows that the schema entails some consequence ϕ and at the same time the untruth of any utterance of ϕ; moreover, a resilient quantified propositional variation on the contingent liar paradox is offered, which only relies on resources usually employed by theories of utterance truth based on the schema.