What Is Geocoding?

Geocoding is the conversion between addresses and coordinates. Forward geocoding goes address → coordinate (e.g. "1600 Pennsylvania Ave NW" → 38.8977°N, −77.0365°W); reverse geocoding goes coordinate → nearest known address within a search radius.

Geocoding is the bridge between two ways of locating a place: the postal address that humans write on letters and the coordinate that GPS receivers, web maps and spatial databases need to do anything with the location. The Coordinately tools /tools/address-to-coordinates and /tools/coordinates-to-address implement both directions; the underlying coordinates are stored in WGS-84 per /learn/wgs84-explained. Forward and reverse geocoding are foundational to ride-hailing, food delivery, real-estate search, emergency dispatch, and almost every map-based consumer application. This pillar explains how the conversion works, what the major providers offer, the accuracy and confidence model that every production geocoder publishes, and the standardisation steps that turn "1600 Pennsylvania Ave NW" into something the database can match.

Forward vs reverse geocoding

The two directions of geocoding are operationally distinct but use the same underlying reference data.

Forward geocoding is one-to-many in nature: "Springfield" alone matches dozens of cities; even a complete street address can match multiple buildings sharing the address (in different countries, or duplicate ranges within a city). Production geocoders return a ranked list of candidates with confidence scores rather than a single answer.

How forward geocoding works

A modern geocoder runs three stages: address standardisation, candidate lookup against the reference database, and ranking by confidence.

The accuracy ceiling is set by the reference data, not the algorithm. A geocoder backed by parcel-level data (each address mapped to its exact building rooftop) reaches metre-level accuracy — comparable to civilian GPS, per /learn/gps-accuracy-explained. A geocoder falling back to address-range interpolation (the street is divided into linear segments and the address is placed proportionally along the segment) reaches tens-of-metres accuracy. A geocoder with no local data and only place-name information falls back to centroid of the smallest known administrative region — potentially several kilometres off.

Match-code confidence: what the score means

Every production geocoder returns a confidence indicator alongside the coordinate. Mapbox v6 publishes a match_code.confidence field that illustrates the convention.

The Coordinately tools (per architectural decision 19.1) treat anything below "high" as a candidate to surface in the UI rather than auto-select. Auto-selecting a low-confidence match silently can put a ride-share driver at the wrong intersection, a delivery at the wrong building, or an emergency responder at the wrong street — all incidents that have happened in production geocoders without confidence-aware logic.

The major geocoding providers

The choice depends on coverage area, accuracy needs, licensing constraints (Mapbox response-retention restriction is a hard operational constraint for many use cases), and budget.

Address standardisation: the unsung hero

Most of a geocoder's real work is parsing the input. Human addresses vary widely: capitalisation, abbreviation, punctuation, component order, language — all of which the standardiser must normalise before lookup.

The USPS Address Information System (AIS) is the canonical US reference; equivalent national postal authorities exist in most countries. ISO 19112:2019 standardises the concept of "address as geographic identifier" but leaves the actual string format to national standards (which differ widely).

Reverse geocoding: from coordinate to address

Reverse geocoding is conceptually simpler but operationally tricky. The query is "what address is closest to this (lat, lon)?" and the answer depends on what kind of address-objects are stored in the database.

Most production reverse geocoders return multiple levels in a single response, letting the consumer pick the granularity they want. A delivery app picks rooftop or address-point; a weather app picks city; a country-detection feature picks country.

Geocoding accuracy in real numbers

How accurate is geocoding in practice? Independent benchmarks on US urban addresses give the following typical numbers across providers.

The "geocoding gap" between urban and rural addresses is one of the quieter inequalities in location services: a downtown ride-share in New York or London gets under 10 m accuracy; a delivery to a rural address in rural Africa or parts of India gets several-km accuracy — sometimes none at all, where the address itself isn't in the geocoder's database.

Rate limits and operational considerations

The most common production bug is silently auto-selecting a low-confidence match. The fix is to always check the confidence and surface the candidate list to the user when confidence drops below "high." The /learn/forward-vs-reverse-geocoding support article covers this UX in depth.

Common misconceptions