How to work out where a photo was taken

There are two ways to work out where a photograph was taken. One is to read the metadata, which is fast, precise, and usually unavailable. The other is to read the photograph, which is slow, approximate, and almost always possible.

Most guides cover the first and stop. The first is the one that has largely stopped working.

Metadata: why it usually isn't there

Cameras and phones write technical details into the image file — the EXIF block — including, when location services are on, the GPS coordinates.

That data is real, and when it survives it is decisive. The problem is that it rarely survives, because the major platforms strip it on upload. A photo saved from Instagram, Facebook, X, or most messaging apps has had its metadata removed as a routine privacy measure. Screenshots have none to begin with, since the screenshot is a new file created by your own device.

Where metadata does tend to survive:

• files sent as email attachments, or as "document"/"file" rather than "photo"
• direct transfers — AirDrop, a cable, a memory card
• images downloaded from sites that don't process uploads, such as some forums and personal blogs
• the original on the photographer's own device

So the practical rule: if you received an image through a social platform, assume there is no location data and move on. Spending time on it is the most common wasted step in this whole exercise.

Reading the photograph

This is the durable skill, and it works on any image including screenshots. You are narrowing, not identifying — country, then region, then locality, then sometimes the exact spot.

Language and script first. Any text — signage, adverts, packaging, a menu, a van's livery — narrows the world fast. Not just the language but the specific conventions: accents, character sets, phone number formats, the layout of addresses.

Which side of the road. Instantly halves the map. Confirm with parked cars, not just moving ones.

Licence plates. Shape, proportions, colour, and the position of any coloured band identify a country at a glance, often without reading a single character. Many countries also encode the region.

Road furniture. Signage typefaces, sign shapes, road markings, kerb painting, bollards, traffic light placement (overhead versus roadside, near-side versus far-side) are strongly national and change at borders.

Infrastructure. Utility pole construction, pylon lattice design, plug sockets, fire hydrants, post boxes, and bin design are all regionally characteristic and rarely staged.

Buildings and materials. Roof pitch and material, window proportions, balcony style, brick versus render versus timber. This distinguishes regions within a country better than most people expect.

Vegetation and terrain. Native species, whether trees are in leaf, snow line, soil colour, whether the landscape is managed or wild.

Sun and shadow. Shadow direction gives you an approximate bearing; shadow length with a known date gives latitude. This is slow and imprecise, but it can falsify a claim outright — a shadow pointing the wrong way for the stated hemisphere is conclusive.

Named things. A shop name, a bus route number, a sports club crest, a phone number on a van. This is usually what converts a region into an address, because it is searchable. One legible business name often beats an hour of terrain analysis.

Working the funnel

The order matters, because each step shrinks the search space for the next:

1. Hemisphere and country — driving side, plates, script, vegetation.
2. Region — architecture, terrain, road signage conventions, plate region codes.
3. Locality — a searchable named thing: business, street name, bus route, landmark.
4. Exact position — confirm in satellite and street-level imagery. Match at least two independent fixed features: building outline plus a kerb line, a pylon plus a treeline. One match is a coincidence.

Stop at whatever level the evidence actually supports. "Somewhere in southern Spain" that is correct is worth more than a specific street that is wrong.

What cannot be determined

Being honest about the failure cases matters more here than in most subjects, because confident wrong answers do real damage:

• Interiors. A plain room, a hotel, a car interior, a bathroom. Usually nothing. Occasionally a plug socket or a light switch narrows it to a country and no further.
• Generic nature. Most beaches, forests, and fields are unresolvable. A coastline silhouette against known terrain is the exception.
• Close-ups and portraits. A face against a blurred background contains no location information at all.
• When. Photographs carry no reliable timestamp once metadata is gone. Seasonal and shadow cues suggest conditions, not dates, and an old photo can be posted today.
• Whether it was taken where it was posted. The two are unrelated, and assuming otherwise is the most common error in this whole area.

If the honest answer is "this image does not contain enough to say," that is the answer. A plausible guess presented with confidence is worse than nothing, because someone will act on it.

The part that matters most

Geolocation is a capability with an obvious misuse, and the line is not subtle.

Verifying an image is reasonable: checking that a rental listing shows the building it claims, that a photo used in a news report was taken where it says, that a marketplace seller's photo isn't lifted from another listing, that a picture sent to you is what it purports to be. The subject of the analysis is the image.

Working out where a specific person lives, works, or spends their time from photos they posted is not verification. It is locating a person who did not consent to being located, and the fact that they published the photo is not consent — nobody posting a picture from their street is agreeing to have that street identified. Where it leads to contact, surveillance, or turning up somewhere, it is stalking, and it is treated as such regardless of how public the source material was.

The technique does not distinguish between those two uses. You have to.