It is critical to predict the motion of surrounding vehicles for self-drivingplanning, especially in a socially compliant and flexible way. However, futureprediction is challenging due to the interaction and uncertainty in drivingbehaviors. We propose planning-informed trajectory prediction (PiP) to tacklethe prediction problem in the multi-agent setting. Our approach isdifferentiated from the traditional manner of prediction, which is only basedon historical information and decoupled with planning. By informing theprediction process with the planning of ego vehicle, our method achieves thestate-of-the-art performance of multi-agent forecasting on highway datasets.Moreover, our approach enables a novel pipeline which couples the predictionand planning, by conditioning PiP on multiple candidate trajectories of the egovehicle, which is highly beneficial for autonomous driving in interactivescenarios.