Conventional wisdom posits that smog suppresses outdoor activity while shifting peoples’ activities indoors. Using anonymized Mobile Phone Data Provider Records fused with Point-of-Interest (POI) data sourced from the Gaode (Amap) open database for Beijing (2–22 February 2015), we test this substitution hypothesis at an hourly resolution across 12 POI-defined activity categories. We estimate the adjusted population density (APD) from mobile phone data via usage-bias calibration, interpolate city-wide AQI (Air Quality Index) and PM2.5 fields, and identify associations with a two-way fixed-effects design (Voronoi polygon (VP), day × hour model. We also handle time-invariant POI activities, while factoring in weather and day types. We find a dual suppression of both outdoor and indoor physical activities: worsening air quality is associated with lower participation in most outdoor and indoor activities. Effects are heterogeneous across categories and hours; shopping shows all-day negative marginal effects, whereas a few categories (e.g., sightseeing) display positive correlations in select afternoon hours consistent with congestion-avoidance rather than health-driven indoor substitution. Quantitatively, a 100-point AQI increase is associated with an order of 1–5 persons/km2 decline at peak hours for most activities. A Comprehensive Impact Index (CII) summarizes the spatial heterogeneity across the city. POI venue operators should anticipate city-wide activity reduction both indoors and outdoors under heavy pollution, rather than plan solely for outdoor-to-indoor activity shifts.