Traditional polling questions that ask about individuals’ voting intentions often fall short in predicting election outcomes. In contrast, wisdom-of-crowds questions, such as asking people to estimate how their social contacts or fellow state residents will vote, consistently outperform their intentions across European and U.S. elections. These questions tap into individuals’ social knowledge and informal observations, improving representativeness and revealing hidden voter preferences. Building on this, we demonstrate that integrating these diverse sources of information using a Bayesian bootstrap framework further enhances forecast accuracy. Among the components, social-circle expectations contribute the most, capturing information that is otherwise underrepresented or distorted in traditional polls. Our findings underscore the importance of leveraging social expectations to enhance the accuracy of election forecasts.