We propose and study the notion of survey rationalizability. In the simplest case of dichotomous attitudinal surveys, survey rationalizability means that both questions and ideal views of individuals can be located in the real line in a way that individuals endorse questions corresponding to the upper contour set of their strictly quasi-concave utility. We show how the relative location of questions can be learnt using a revelation mechanism that involves pairs of individuals and triplets of questions, and that the acyclicity of these revelations suffices for rationalizability. We then show that our analysis readily extends to the cases of polytomous surveys and of probabilistic data. We further study the identification of the main parameters of these models and show that, in an exponential version of the probabilistic model, even the cardinal locations can be fully determined. We conclude by studying an alternative model of survey responses for aptitudes.