With the increasing popularity of intelligent assistants (IAs), IA quality assessment is becoming an increasingly active area of research. This paper identifies and quantifies the impact of feedback, a new component in IA user interactions: how the capabilities and limitations of IA affect user behavior over time. First, we show that nonhelpful IA responses cause a delay or reduction of subsequent reactions in the short term through an observational study. Next, we extend the time horizon to examine behavioral changes and show that as users discover the limitations of IA’s understanding and functional capabilities, they learn to modify the scope and wording of their requests to increase the likelihood of receiving a beneficial response from the IA. Our findings highlight the influence of the feedback effect at both the micro and meso levels. We also discuss its consequences on a macro level: unsatisfactory interactions consistently reduce the likelihood and variety of future user interactions in the feedback loop.