A lack of direct electoral checks on government bureaucrats challenges norms of democratic accountability. One proposed solution is to increase the president's control over federal agencies. It is, however, an open question as to whether voters will attribute responsibility to the president even when in charge of agencies. A key empirical challenge has been that presidential control is not randomly assigned across agencies. To overcome this issue, I compare two agencies that enforce the same policy but differ in insulation from presidential control. I examine a large, unique dataset of news coverage, showing that news coverage of the presidentially-controlled agency features more politicized content that ties the agency to the president. I then demonstrate experimentally that this political content increases attribution of control to the president. The results support theories that claim agency design moderates voter attribution of responsibility to the president. This paper broadly adds to the literature on institutional design and the determinants of agency discretion.