Learning From False Stories

False information often shapes beliefs even after retraction or correction. Using incentivized online experiments, we document a qualitative residue in learning from false information: False quantitative signals generate little to no belief updating, whereas false stories lead to substantial residual belief impact. This effect is robust across a range of design variants. Replacing evaluative stories with more neutral variants eliminates the residue, indicating that the valence of the qualitative information plays a central role. We provide direct evidence that false stories increase mental simulation, measured via self-reports and valence extracted from speech recordings. Respondents are also more confident in beliefs formed after false stories than after false quantitative information, despite being further from the rational benchmark. Additional experiments suggest that the effect also appears among respondents who report not being influenced by the story, consistent with stories partly shaping beliefs below conscious awareness.