This paper studies how an optimal menu chosen by a social planner depends on whether agents receive imperfect signals about their true tastes (imperfect self-knowledge) or the properties of available alternatives (imperfect information). Under imperfect self-knowledge, it is not optimal to offer fewer alternatives than the number of different tastes present in the population, unless noise is infinite (agents have no clue about their true preferences).
As noise increases, the social planner offers menu items that are closer together (more similar). However, under imperfect information, as noise increases, it could be optimal to construct a menu with more distinct alternatives, restrict the number of options, or, for some finite noise, offer a single item.