75 Million More Addresses: Expanding Geocoding Precision

Inaccurate geocoding breaks the user experience. If your query returns a street-level guess instead of a rooftop coordinate, a user might be on the right street but the wrong block, or a delivery driver gets routed to an approximate centroid instead of a front door. The coordinates are "close enough" until they aren't, and your application takes the blame.

We continuously update our geocoding dataset as new address and POI data becomes available. This release is our largest expansion yet: 75 million new addresses, 3 million new points of interest, improved intent recognition, and a new way to query with structured address fields.

Rooftop-Level Accuracy at Scale

The bulk of this expansion comes from OpenAddresses, one of the largest open datasets of address-level coordinates. These aren't interpolated guesses along a road segment. They're rooftop-level points tied to individual parcels, which matters for any application that depends on precise placement: routing, last-mile logistics, property lookup.

The gains aren't spread evenly. France and the US lead with roughly 25 million new addresses each, followed by Germany with 3 million and notable expansions across Brazil, Australia, and Mexico. If you've been hitting street-level fallbacks in these regions, you should see noticeably better results.

Fresher Points of Interest

POI data goes stale quickly. A restaurant opens, a shop closes, a business moves across the street, and suddenly your search results are sending users to the wrong place. We've added 3 million POIs from Foursquare OS Places, bringing our dataset more in line with what's actually on the ground. Better coverage for the kinds of searches your users actually run: finding a specific restaurant, locating the nearest pharmacy, or navigating to a business that opened last month.

Smarter Search, Fewer Misses

Beyond raw data, we've improved how our /search APIs interpret what users actually mean. The changes focus on intent recognition: better handling of queries for larger geographic areas, complex or compound place names, and cases where the most obvious textual match isn't the right result.

Your users should find more of what they're looking for on the first try.

Structured Address Input, Now on v1

For developers working with structured address data, this release adds house_number, street, and unit as first-class fields on the geocoding API. Previously, structured queries supported country, region, and postal code, but the address itself arrived as a single text field that we had to parse before running a match.

If your dataset is already structured that way, you can now pass it through directly. This is common for shipping, CRM, and property data. Queries with the new fields skip the parsing step entirely, which typically means better match quality and faster responses.

The free-text address field isn't going anywhere. If you only have a single string, we'll still parse it for you. The new fields are available on v1 today. Adopt them at your own pace.

No Tracking Required

A common assumption in the industry is that better geocoding requires more user data: search history, device location, behavioral signals. We don't work that way. These improvements come from better source data and smarter algorithms, not from profiling your end users. Your users' queries are yours, not ours.

As with everything we build, these updates are available through standard APIs and SDKs, including MapLibre and Leaflet. Technical precision, fair pricing, and privacy are baked in from the start.

Get Started

The expanded dataset, structured address fields, and search improvements are live now for all Stadia Maps customers. New here? Create a free account and start geocoding.