Building footprint data are essential for spatial planning, disaster management and hydrological analysis, yet in many Indonesian cities they are still produced by manual digitization or remain incomplete. This study evaluates how well three open sources detect buildings in Binjai City, North Sumatra (94.19 km²): the Google Open Buildings version 3 polygons (confidence 0.65–1.0), OpenStreetMap (OSM) building polygons downloaded in May 2025, and a built-up mask derived from the Normalized Difference Built-up Index (NDBI) of a 2023–2024 Sentinel-2 composite, all processed in Google Earth Engine. No new model was trained. Accuracy was assessed at 440 reference points (220 building and 220 non-building) placed manually on Google Satellite imagery tiles downloaded in February 2025. Google Open Buildings detected 116,913 buildings (1,241 per km²) and agreed best with the reference points (overall accuracy 97.27%, F1 97.22%), followed by OSM, with 86,605 buildings (91.14%, F1 90.32%), and NDBI (77.95%, F1 79.66%). OSM produced almost no false positives but missed many buildings (recall 82.73%), whereas NDBI produced 67 false positives, mostly on roads and rivers. Raising the Open Buildings confidence threshold did not improve precision and reduced recall sharply above 0.80. Because the reference points were placed manually, building points were restricted to clearly visible roofs, only one city was studied and the data dates differ, the accuracy values are descriptive, probably optimistic for recall, and should not be extrapolated. Open Buildings appears to be a promising starting dataset for building inventories in comparable Indonesian cities when verified locally, and the Earth Engine script is openly available for replication