We utilize region-day variation in actual vs. forecasted wet bulb globe temperatures (i.e. forecasting errors) to investigate the effects of the first comprehensive heat-health warning system in Japan. We find that heat alerts led to an increase in heat stroke counts of 17%. An analysis of mechanisms utilizing several datasets suggests the effect is due to increased reporting, as opposed to potential “adverse” behaviors or substitution in health diagnoses. We further find that four times as many heat strokes are detected in low-income neighborhoods compared to high-income neighborhoods, highlighting severe environmental inequalities in health reporting behavior.