A commonplace view is that only a semantic theory that interprets sentences of a language according to what their utterances intuitively say can be correct. The rationale is that only by requiring a tight connection between what a sentence means and what its users intuitively say can we explain why, normally, those linguistically competent with a language upon hearing its sentences uttered can discern what they say. More precisely, this approach ties the semantic content of a sentence to intuitions about “says that” reports. However, this approach has been forcefully attacked (e.g., Cappelen and Lepore 1997, 2004). Building on Stojnic and Lepore (2013), I argue for a direct way of identifying semantic content, one exploring the role of linguistic conventions, cashed out in terms of Lewis’ twin ideas of conversational record and coordination (Lewis, 1969,1975). Moreover, while most theorists assume that linguistic conventions alone underdetermine what is said, I show that they in fact deliver fully specified semantic content of an utterance in context.