In this article, the data-based output consensus of discrete-time multiagent systems under switching topology (ST) is studied via reinforcement learning. Due to the existence of ST, the kernel matrix of value function is switching-varying, which cannot be applied to existing algorithms. To overcome the inapplicability of varying kernel matrix, a two-layer reinforcement learning algorithm is proposed in this article. To further implement the proposed algorithm, a data-based distributed control policy is presented, which is applicable to both fixed topology and ST. Besides, the proposed method does not need assumptions on the eigenvalues of leader's dynamic matrix, it avoids the assumptions in the previous method. Subsequently, the convergence of algorithm is analyzed. Finally, three simulation examples are provided to verify the proposed algorithm.