Geolocating a Photo in Six Passes

Geolocation looks like intuition and is actually bookkeeping. The researchers who are good at it are the ones who work the image in a fixed order, exhausting each class of evidence before moving on, instead of staring at a picture waiting for recognition. Six passes cover almost everything.

Pass 1 — What the file already knows

Check for embedded geotags first: an original upload from a phone often carries coordinates, a timestamp, a device model, and sometimes a compass heading. This costs seconds and occasionally ends the investigation immediately. Expect it to fail most of the time, because messaging apps and social platforms strip metadata on upload — a stripped file is itself a signal that the image passed through a platform rather than arriving from the camera.

Pass 2 — Reverse image search

Run the whole image through a reverse image search, then crop and repeat. A wide shot with a dominant sky often matches nothing, while a tight crop of a shopfront, statue, or distinctive roofline returns an exact hit. Crop generously around anything manufactured — human-made objects are far better search anchors than natural terrain. If you get a match, open the earliest version you can find; later copies are usually the same photo with a wrong caption attached.

Pass 3 — Terrain and light

Natural features narrow a continent to a region. Hilliness, vegetation type, soil colour, and the character of the sky all carry regional signatures. Shadows carry time and latitude: the angle and length of a shadow, combined with the timestamp on the post, can rule out entire time zones. Sun bearing is a coarse tool, but a coarse tool that removes half your candidates is worth using early.

Pass 4 — Infrastructure

The built world is a catalogue. Road markings, kerb heights, street-lamp design, bollard spacing, utility-pole hardware, guardrail profiles, and manhole covers are specified by local authorities and published in design manuals — the same manual that tells you which country a kerb belongs to is online and searchable. Railway catenary style, traffic-light housing shape, and the colour of a curbstone have all resolved locations that reverse search could not.

• Read text on the ground: address plaques, milestone stones, utility stencils, and painted road markings are usually legible at full resolution even when they look unreadable in a thumbnail.
• Check the phone network name in any visible status bar, and the currency or unit on any sign.
• Note the driving side, and whether parked cars face the correct direction for it.

Pass 5 — Signage and language

Language narrows fast, but the trap is that most of the world's signage is bilingual or in Latin script. What separates regions is typography, colour convention, and the specific phrasing of official terms — not the alphabet. A regional language on a road sign, a distinctive script variant, or an unusually worded public notice is a much sharper marker than 'this is Spanish'.

Pass 6 — Match and prove

Once you have a candidate, prove it rather than admire it. Pull street-level imagery at that spot and line up the geometry: the same fence, the same tree line, the same pole in the same relative position. Then verify the time. A location match with the wrong date is a different event, and the difference often shows up as a building that had not been constructed yet, a shop that had not opened, or a tree that was removed years earlier.

A geolocation is not finished when the view looks right. It is finished when the view looks right and something in the frame could not have been anywhere else.

Write the result as coordinates, a confidence level, and the two or three features that carry the whole conclusion. If someone removes those features from your reasoning and the answer still looks correct, your answer was never about the features — it was about what you expected to find.