The extent of reward generalization varies with access to explicit associative memory representations

Memories tend to form networks that reflect relationships between experiences. These networks, or cognitive graphs, have been used to study how structural features of memory can bias decisions—analogous to how reward shapes choices by prioritizing reward-associated items in memory. However, how reward interacts with the structure of existing cognitive graphs to influence subsequent value-based decisions remains unclear. Across three experiments, participants learned a community-structured cognitive graph, reconstructed it, learned which item in the graph is associated with reward, and then made value-based decisions. Results indicated a limited retroactive spread of reward, contingent on access to an explicit representation of either the cognitive graph or reward-item associations. In their absence, participants’ choices revealed a reorganization of the cognitive graph structure. These findings suggest that explicit memory representations enable reward to spread broadly across association space; without them, reward actively reshapes associative memory structures to support value-based decision making.