Behind every pin on our map there is a chain of data decisions. Movie databases, fan wikis and production notes often describe locations in natural language: “village near the canyon”, “street in central London”, or simply a country name. Turning that into a coordinate that a map can understand is a careful process.
First, we normalise the raw text. That means stripping formatting, separating scene descriptions from possible addresses, and detecting languages or obvious place names. In some cases we split a single note into a “scene description” and a “geocoding hint” so that we can preserve both.
Next, we send the geocoding hint to an open-source geocoder backed by OpenStreetMap data. We resolve ambiguous results by comparing them with other locations from the same movie: if most scenes for a film cluster in Tunisia, for example, a random result in Canada is likely wrong.
We also keep track of confidence levels. High-confidence results can be displayed directly, while low-confidence ones may be hidden or flagged for manual review. Over time, community feedback and better data help us promote more points from “uncertain” to “verified”.
This approach lets Where Was It Filmed scale to thousands of titles while still striving for accuracy. Each location on the map is more than a dot: it is a tiny piece of detective work connecting cinema to geography.