Fooling yourself: how narratives shape beliefs

Decision-makers usually receive information through narratives that combine diagnostic evidence with nondiagnostic details. In a laboratory experiment, we study how such nondiagnostic clues affect belief updating. Participants repeatedly report beliefs in a Bayesian inference task within a narrative context. Reduced-form estimates and subject-level classifications show that nondiagnostic narrative clues systematically induce belief revision toward maximal uncertainty, weakening previously accumulated diagnostic evidence. This effect is weaker in an equivalent abstract context and disappears in an identical narrative task when nondiagnostic clues are replaced by no–information messages. We assess the economic significance of such return to uncertainty by showing that it delays belief convergence, thereby increasing the amount of information agents require before making a decision.