We examine how candidate selection into the supply of policy information determines its electoral effects. In a nationwide debate initiative designed to solicit and rebroadcast policy promises from Liberian legislative candidates, we randomized the encouragement of debate participation across districts. The intervention substantially increased the debate participation of leading candidates but led to uneven electoral returns for these candidates, with incumbents benefiting at the expense of challengers. These results are driven by differences in compliance: complying incumbents, but not challengers, positively selected into debate participation based on the congruence of their policy priorities with those of their constituents.