Many recommender systems generate personalized recommendations by observing users’ past behaviors. In this work, we conduct a lab experiment and survey to test whether a user’s awareness of their recommendation algorithm changes how they interact with the platform—a phenomenon that we refer to as user strategization. We use a model in which strategic users modify their behavior when they believe they will receive future recommendations based on current behavior (the Incentive condition) and adapt their behaviors to different algorithms (the Information condition). To conduct an online behavioral experiment with full control over the system and recommendations, we build a custom music player and randomize participants into different Information and Incentive conditions. We find strong evidence of strategization across outcome metrics. For example, in our Incentive condition, participants who are told their behavior is used to generate personalized recommendations gave explicit feedback to the platform 6.9 times more on average than participants who are only told that we are learning music preferences; similarly, in our Information condition, participants who are told the algorithm pays particular attention to their dwell time use the explicit “like” and “dislike” buttons 3.2 times less. We find these large effects despite minimal differences between the UI across treatments (in particular, the Incentive condition does not reveal information about the algorithm). In our post-experiment survey, participant responses corroborate our experimental results, with nearly half self-reporting that they strategize “in the wild.” Some respondents share that they ignore content that they like to avoid over-recommendation of that content in the future while others state that they strategize to help the algorithm. Together, our findings indicate that user strategization is common and accounting for it can improve personalization for both the user and platform.