Coalition theories assume that coalition behavior is shaped by anticipating how others want to distribute resources. Therefore, coalition outcomes depend on players anticipating others’ demands and choosing to join those who offer favorable terms. But how do players anticipate others' demands? Current theories assume (but have not tested) that the static features of the game (e.g., players’ resources at the start of the negotiation) elicit universal allocation norms (rules about fair distribution). We propose that while static features likely serve as cues to anticipate others’ norms, these are context-dependent rather than universal. Consequently, we argue that coalition formation requires perspective-taking to infer others' endorsed norms. We further argue that the role of perspective-taking depends on the predictability of these norms: it is crucial when allocation norms are harder to anticipate, but less important when norms are obvious. We tested this hypothesis in a non-registered experiment (N = 851), using a real-time three-person coalition paradigm, in which allocation norms were either clear (High-Predictable condition) or unclear (Low-Predictable condition). As expected, we found that perspective-taking predicted greater inclusion in winning coalitions and higher final payoffs only in the Low-Predictable condition, but not in the High-Predictable condition. We aim to replicate these findings in a registered report. These findings demonstrate that psychological processes, such as perspective-taking, play a relevant role in coalition formation and challenge existing theories in the field.