Extreme heat is a growing public health threat. In the United States (US), the National Weather Service (NWS) issues alerts ahead of forecast periods of extreme heat. However, the behavioral impact of official heat alerts remains poorly understood at actionable scales, and their real-world effectiveness has been difficult to quantify. We used anonymized, county-level Google search data on the proportion of searches classified into eight heat-associated categories, aggregated daily across 2,581 U.S. counties from May to September 2023. We implemented a time-stratified case-crossover method using conditional Poisson models to quantify the association between the proportion of heat-related internet searches and (a) daily maximum temperature and (b) NWS heat alerts, adjusting for same-day county-specific temperature percentiles. We further evaluated how associations varied spatially and temporally within the season. Across all counties, searches for heat stroke/exhaustion were 3.60 (95% CI, 3.38–3.85) times higher when comparing the 95th percentile of daily maximum temperatures to the 1st percentile. Air-conditioning searches were 2.47 (2.43–2.51) times higher. Exposure–response curves rose steeply above the 80th percentile except for public swimming and cooling center queries. On heat alert days, heat-related searches were 1.27 (1.26, 1.28) times higher relative to matched non-alert days. Results varied by region. However, effect modification was pronounced: early-summer alerts (May–June) elicited stronger responses than late-summer alerts (July-September) in all heat-related search categories except cooling centers. Our findings suggest that heat alerts trigger meaningful, real-time behavioral responses during heatwaves, particularly in early summer and historically cooler regions. High-resolution internet search data offer a promising tool for evaluating public engagement with risk communication and can provide local officials guidance for further optimizing messaging strategies.