One important aspect of conceptual engineering concerns the evaluation of concepts for usefulness, which lets us decide whether to keep those concepts or to revise them. Among the concepts that pose a particular challenge for that project are empty concepts. These are concepts like ‘unicorn’, or ‘phlogiston’, that do not refer to any actual objects. We use such concepts quite often—but how could they be useful when they do not refer to anything, and how could they come about in the first place? In earlier work, we proposed an account of concept utility that treated concepts as groupings of objects, implying that empty concepts would have zero utility, and leaving no explanation for how they could arise. While we believe they indeed have no utility in natural contexts once known to be empty, here we extend our theory in two ways to account for empty concepts. We show how they can first arise—as hypothetical groupings of features rather than objects; and we show how they can have a measurable utility—in scientific theorizing, and in fiction.