Free-riding produces inequality in the prisoners' dilemma: cooperators suffer costs that defectors avoid, thus putting them at a material disadvantage to their anti-social peers. This inequality, accordingly, conveys information about a social partner's choices in past game play and raises the possibility that agents can use the aggregation of past payoffs-i.e. wealth-to identify a social partner who uses their same strategy. Building on these insights, we study a computational model in which agents can employ a strategy-when playing multiple one-shot prisoners' dilemma games per generation-in which they view other agents' summed payoffs from previous games, choose to enter a PD game with the agent whose summed payoffs most-closely approximate their own, and then always cooperate. Here we show that this strategy of wealth homophily-labelled COEQUALS ("CO-operate with EQUALS")-can both invade an incumbent population of defectors and resist invasion. The strategy succeeds because wealth homophily leads agents to direct cooperation disproportionately toward others of their own type-a phenomenon known as "positive assortment". These findings illuminate empirical evidence indicating that viewable inequality degrades cooperation and they show how a standard feature of evolutionary game models-viz. the aggregation of payoffs during a generation-can double as an information mechanism that facilitates positive assortment.