This paper presents AccessiBot, a tactile paving detection robot developed to address urban accessibility challenges. Beyond collecting real-time data for evidence-based governance, this study explores AccessiBot's potential to encourage citizen engagement. Using the Wizard-of-Oz methodology, we conducted an in-the-wild study to simulate real-world interactions and observe how people respond to the robot. Our observation findings show that people do help robots, and their engagement with the robot varied significantly, ranging from passive acknowledgement to active assistance. Besides, post-hoc interviews indicated that participants recognized the robot's social value.