Geolocation is the craft of establishing where an image was taken from what is visible in it. Done well, it is reproducible: another analyst following the same notes reaches the same coordinates. Done badly, it is a confident guess. The difference lies in method, not in a gifted eye.
Check what you have before you look at the picture
Before anyone studies the content, we preserve the file and read its metadata. Major platforms strip location and camera data on upload, but files sent directly, attached to email or posted on personal sites and forums often keep it. Where metadata survives, we treat it as a claim to be tested, not a fact: device clocks are wrong, time zones are misread and metadata can be edited. We explain what metadata can and cannot prove in What files say about the people who made them.
Search for earlier copies
The next step is reverse image search across several engines, because each indexes a different part of the web. The aim is not to identify anyone in the image. Our standards do not permit face recognition to name people. The aim is to find earlier and larger copies of the same image. The earliest copy is often the source, and its caption, page or uploader may state the location outright. Cropped, mirrored and recompressed versions are common, so we search with several crops of the distinctive parts of the frame as well as the whole image.
The sweep: reading the image systematically
When search does not answer the question, the analyst works through the image in a fixed order and records every observation, including the ones that turn out to be irrelevant. A fixed order stops the eye from settling on the first interesting detail.
- Text. Signs, shopfronts, vehicles, posters and graffiti. The language and script narrow the region. Partial words can be searched. Business names lead to listings with addresses.
- Roads and vehicles. Which side of the road traffic drives on, the color and pattern of center and edge lines, the shape and color of road signs, bollards and guardrails, and the format of number plates, which identifies the country and sometimes the region.
- Architecture and materials. Roof shapes and coverings, window styles, balconies, building height, the pattern in which bricks are laid and the period those details belong to.
- Infrastructure. Utility poles and their fittings, street lights, manhole covers, curb design, tram lines and power lines.
- Terrain and vegetation. Mountain profiles on the skyline, coastline, soil color, crops and tree species, all of which constrain climate and region.
- Light and reflections. Shadows give direction and, with a date, time of day. Windows, vehicles and puddles often reflect what is behind the camera, including signs and buildings that are not otherwise in frame.
Write down every observation before forming a hypothesis. An analyst who decides early that a scene is in one city will start seeing that city everywhere.
Narrowing from region to vantage point
Observations are combined into hypotheses that shrink the search area in steps: country, region, town, district, street. Each step is tested against mapping, satellite imagery and street-level imagery rather than intuition.
Distinctive features do the heavy lifting. A tall building with an unusual crown, a bridge, a church tower or a stadium can often be identified on its own, and then the question becomes where the camera stood. For an image taken from a height, we identify two or more buildings in view, note their relative positions and angles, and work back to the building and floor from which those angles are possible. For a street-level image, the alignment of foreground and background features fixes the position to within a few meters.
Confirming a match
A location is not confirmed because it looks right. Our standard requires at least three independent features that match in both appearance and relative position, checked against imagery whose date is known. We also look for features in the image that should be visible at the proposed location and are not. One unexplained mismatch can be enough to reject a candidate.
Places change. Shops close, buildings are painted and trees grow. When the available imagery is older or newer than the photograph, we say so and explain how differences were reconciled.
Chronolocation: establishing when
Once the location is fixed, the image can often be dated. The direction and length of shadows, measured against calculated sun positions for that place, give a time window, sometimes within minutes. Historical weather records confirm or rule out cloud, rain or snow. Vegetation, decorations, construction progress and visible events narrow the date. Chronolocation matters most in event verification, where an old image recirculated as new is one of the most common forms of misinformation we see.
Documenting a geolocation
Every geolocation we report includes the coordinates, the imagery used and its date, annotated side-by-side comparisons of each matching feature, the features that could not be matched and why, and a confidence level. The goal is that a court, a client or a skeptical colleague can follow the reasoning without taking our word for it.
Common pitfalls
- Mirrored images. Text reading backward is the obvious sign. Traffic on the wrong side of the road is a subtler one.
- Lookalike places. Chain stores, standard housing designs and planned towns repeat across countries.
- Outdated imagery. A confident match against imagery that is years old can be wrong today, and the reverse.
- Staged or edited images. Composites and generated images contain features that never coexisted. Inconsistent shadows are often the first clue.
- Confirmation bias. The first plausible candidate is not the answer until the alternatives have been ruled out.
Where we stop
The same techniques that verify a conflict video can reveal where a private person lives. We geolocate images to answer screened questions, such as where a threatening subject is, where an event happened or whether an image is what it claims to be. We do not locate private individuals for personal or domestic reasons, and requests of that kind are declined at screening. In protective work, we use the same craft in reverse, to show principals and their families what their own photographs reveal before someone else notices.