In this paper, we analyze a principal's optimal feedback policy in tournaments. We close a gap in the literature by assuming the principal to be unable to commit to a certain policy at the beginning of the tournament. Our analysis shows that in equilibrium the principal reveals in-termediate information regarding the agents' previous performances if these performances are not too different. Moreover, we investigate a situation where the principal is not able to credi-bly communicate her information. Having presented our formal analysis, we test these results using data from laboratory experiments. The experimental findings provide some support for the model.