Many decision problems can be divided into three parts: Generating a set of options to consider, evaluating them, and choosing the best. Prior models often assume that the ``consideration set'' is established in a single step prior to evaluation. Alternatively, people may dynamically and continually assess whether to expand the consideration set based on the quality of the actions considered so far. We use modeling to derive a signature property of dynamic consideration set construction and then demonstrate it in two experiments on human participants.