Transport planning for earthquakes concentrates on damaged supply, yet in most events the network keeps running, and demand determines how it is used. This paper examines who stopped using urban rail after a strongly felt but largely undamaging earthquake, and what explains the spatial pattern of the loss. The analysis uses open smart-card counts by station, day and age group for more than 340 rail stations in Istanbul in April–May 2024 and 2025, around the moment magnitude 6.2 earthquake of 23 April 2025. It struck on a public holiday and was followed by two days of school and university closures. An event-study difference-in-differences design with a same-date 2024 control separates the earthquake from the holiday, bridge-day placebos bound the long-weekend effect, and station-level regressions for 270 stations relate losses to modelled shaking, user age and neighbourhood socio-economic status. Ridership fell by about one-third on the day after the earthquake and was within 5% of normal within a week, although a small residual shortfall cannot be ruled out. The loss was largest among young users, as the closures imply. However, users aged 30–60, whom the closures affected only indirectly, still travelled about one-quarter less. Shaking intensity does not explain which stations lost most; losses were larger where stations served more young users and where neighbourhood status was higher. Overall, the demand shock of this moderate earthquake was large, mostly short-lived and socially graded, shaped more by institutional closures and the social composition of station catchments than by the strength of shaking.