Standard rational learning models with information frictions can explain consumers’ heterogeneous macroeconomic beliefs but not their systematic bias. Using novel data on 47,000 consumers across 47 countries representing 90% of global GDP, we document novel facts that can inspire realistic adaptations of the rational-inattention framework. First, conditional on gathering economic information, most consumers use signals from local economic environments—utility bills, shopping, family & friends, and social media—that are not representative of the distribution of aggregate macroeconomic variables and lead to biased inflation, interest rate, and house price expectations. Second, information seeking is highly heterogeneous across countries: consumers in countries with higher historical inflation are more likely to seek more economic information but form more inaccurate expectations because they focus on signals from their local economic environments. Third, the cross-country setting allows revealing that sorting into local vs. aggregate sources relates to consumers’ trust in governments and central banks (the suppliers of aggregate economic information): trust in these institutions increases reliance on aggregate signals, resulting in smaller forecast errors. Our results inspire advances in models of belief formation and suggest an overlooked role for trust in economic institutions in driving the process of information seeking and beliefs formation by consumers.