Summary We explored (i) whether narratives can influence viewers' evaluations toward alcohol through evaluative learning and (ii) compared predictions from dual‐process and single‐process models of evaluative learning. In study 1, participants had to read vignettes, while they were exposed to TV show excerpts in study 2. Both studies ( n study1 = 147; n study2 = 150) followed a 2 (valence: positive vs. negative) × 2 (drinking consequences: yes vs. no) study design. Automatic associations and propositional beliefs were then measured by an Implicit Association Test (IAT) and a Relational Responding Task (RRT) respectively. A multilevel meta‐regression was conducted to provide cumulative evidence for our hypotheses. Our first study did not yield robust significant results in the direction of associative or propositional processes. Conversely, the results of study 2 and meta‐analytic findings showed stronger evidence for (i) an effect of exposure to narratives on alcohol‐related evaluations and (ii) in favor of propositional models. Simply presenting a stimulus within a valenced context had no effect on the IAT or RRT. We conclude that these results are more in line with inferential propositional models of evaluative learning than with dual‐process models.