A shift away from goal-directed, model-based behavior is commonly viewed to characterize alcohol use disorder (AUD). Previous research, however, has failed to demonstrate differences between individuals with and without AUD regarding goal-directed control, operationalized as model-based behavior. Instead, findings suggest associations between model-based behavior and alcohol consumption patterns, but mechanistic insights into the link between model-based behavioral and neural signatures and longitudinal, real-life control over alcohol intake remain elusive. Here, we investigated whether experimentally assessed model-based behavior can prospectively predict intentional reduction of alcohol consumption in daily life. Therefore, we related behavioral and neural markers of model-based behavior during a sequential decision-making task in 67 participants with AUD (20 women) to long-term smartphone-based ecological momentary assessments of daily alcohol intake and weekly alcohol consumption intentions over a period of up to one year. Model-based behavior and its neural signatures in bilateral hippocampus and ventral striatum moderated how well individuals succeeded in aligning their alcohol consumption with their drinking intentions during the following year. Specifically, AUD participants with higher model-based behavior and associated stronger hippocampal and weaker ventral striatal learning signals exhibited enhanced capacity to intentionally reduce their alcohol consumption in everyday life. These findings provide evidence for the ecological validity of computational concepts of goal-directed behavior and suggest specific treatment targets for individually tailored interventions to regain control over alcohol use.