Abstract
Algorithmic advice is prevalent in markets for complex goods, helping consumers compare alternatives, yet its effects remain ambiguous. We develop a simple model in which comparing complex products is cognitively costly and buyers may follow imperfect advice to economize on cognitive effort. We then conduct an experiment that varies product complexity and the availability and accuracy of advice. Consistent with the model, complexity reduces buyer choice accuracy and surplus when no advice is available, while perfect advice restores buyer performance to the level observed in markets with simple goods. Imperfect advice, however, does not improve buyer performance, even when buyers are informed of its inaccuracy. These results suggest that the effectiveness of algorithmic advice depends critically on its accuracy and that disclosing its limitations alone is insufficient to improve buyer performance.