Points-of-Interest (POI) search, the task of retrieving and ranking geographically situated entities in response to a query issued from a specific location, underpins major mapping and navigation services. We present POI Search Suite (POISS), the first large-scale dataset designed for this task, pairing over 300K geo-anchored natural language queries across 7 languages and 14 locales with more than 4M graded relevance labels across over 90M candidate POIs drawn from the 72M-entity Overture Maps Places index, jointly capturing semantic match, geographic proximity, and entity quality. Our experiments assess the utility of the proposed dataset for advancing research on POI search.