Conceptual surveys like the FCI and CSEM are common, and course reforms often have the goal of improving student gains on these surveys. There exist various approaches to improving said gains, and there is occasionally concern that such methods "teach to the test" excessively. To our knowledge, however, there has been little direct experimentation on whether teaching to the test, even intentionally, has the expected result. In this paper, we report on a simple two-semester experiment involving ~900 students where we tried two different approaches to improving CSEM gains in an introductory E&M class. In the first trial, we gave students many of the questions from the CSEM as Peer Instruction-style clicker questions in lecture. In the second, we redeveloped parts of our Studio physics curriculum to target CSEM concepts without replicating CSEM questions. Comparing the CSEM gains in the experimental sections to the previous year's sections, we find that the first trial resulted in significant (~0.20) shifts in normalized gains on the relevant questions, while the second trial resulted in minimal or no shifts.