This paper develops a framework for analyzing how economic agents form and use simplified causal narratives to understand and forecast macroeconomic dynamics, with specific application to monetary policy. We introduce data-driven narratives that satisfy internal consistency through cross-equation restrictions while potentially remaining misspecified. Applying this approach to a Fisherian model with average inflation targeting (AIT), we decompose belief distortions into extrinsic bias, intrinsic bias, and model-selection bias. We demonstrate that policy transparency fundamentally influences equilibrium determination: transparent AIT tends toward unique equilibria, while intentionally ambiguous AIT can generate multiple equilibria with different dominant narratives. The effectiveness of AIT critically depends on the persistence structure of underlying shocks and which narratives prevail. Our empirical analysis of U.S. data (1960-2019) reveals that demandfocused narratives have historically dominated. This helps explain why the Federal Reserve’s 2020 adoption of AIT may not generate the anticipated “make-up” dynamics essential to the policy’s success.