The consumer search literature mostly considers independently distributed products. In contrast, I study a model of directed search with infinitely many products whose valuations are correlated through shared attributes. I propose a tractable, systematic, history-dependent scoring system based on nests of correlated products that leverages the predictability of the optimal search process along different paths. This scoring system generates an optimal search policy conceptually equivalent to the familiar optimal policy with independently distributed search products. The policy instructs the consumer to inspect unrelated products until an attribute the realization of which surpasses the added informational value of inspecting two new attributes is found. The search paths emerging from this policy match recent evidence of consumer learning through search, and can rationalize backtracking to a previously abandoned attribute.