An OSINT Photo Geolocation Workflow
· 4 min read
Short answer
A sound OSINT photo geolocation workflow runs in one order: check metadata, search for existing copies, read the visual clues, form a hypothesis, then verify it against reference imagery. Recording what you ruled out matters as much as what you concluded.
The order below is not a preference. Each step is cheaper than the one after it and can end the search outright, so running them out of sequence mostly wastes time — and skipping the last one produces confident answers that are wrong.
Step one: what does the file itself say?
Check the metadata before anything else. If GPS coordinates survived, the question is answered to within a few metres and every later step is unnecessary. It takes seconds and is skipped surprisingly often.
Read it locally rather than uploading to a web viewer — the tools for that are already on your machine. Treat whatever you find as a claim: every EXIF field is writable, so coordinates are evidence of what the file says, not of where a camera was.
Step two: has this image been published before?
Run it through TinEye sorted by oldest and through at least one similarity engine. An earlier publication under a different caption settles the matter immediately, and it is the single most common outcome for images circulating online.
Crop before searching, and keep the crop above roughly 800 px on its long edge — below that, matching degrades sharply. A full scene gives the matcher a dozen competing signals; one facade or sign gives it one. Which engine to use for which subject is worth knowing, because they fail on different photographs.
Step three: what do the clues say?
Read in a fixed order — driving side, script and signage, road markings, infrastructure, then vegetation and light. Regulated things stop at borders and are worth far more than landscape, which changes gradually.
The full ordering, and why it goes cheapest-first, is in the country clues guide. The discipline that matters here is writing down each clue and what it eliminated, not just what it suggested.
| Step | Can settle | Cannot settle |
|---|---|---|
| Metadata | Exact coordinates, if present | Whether they are genuine |
| Reverse search | Prior publication and context | Anything, if the scene is unpublished |
| Clue reading | Country and often region | A specific street on its own |
| Hypothesis + verification | A precise location, when imagery exists | Anything in uncovered regions |
Step four: how do you verify a hypothesis?
Find the candidate location in reference imagery and check details you did not use to form the hypothesis. Matching the clues that led you there proves nothing — you already knew those agreed.
This is where most amateur geolocation goes wrong. Confirming with the same evidence that generated the guess is circular; the test is whether unused details also line up — a building behind the subject, a kerb pattern at the edge of frame, the position of a drain.
And verification depends on imagery existing at all. Coverage is very uneven, so in thinly mapped regions the honest end point is a region rather than a street.
What should you record?
Every clue, what it eliminated, and the confidence you ended with. A conclusion nobody can retrace is not a finding, and negative results — the countries you ruled out and why — are what let someone else check your work.
- The image as received, unmodified, plus a hash if provenance matters.
- Each clue and the specific possibilities it removed.
- Which engines you ran and what they returned, including nothing.
- The unused details you verified against, not the ones you started from.
- A final confidence, stated plainly — "probably this city" is a legitimate answer.
That last point is the discipline the whole workflow exists to protect. An honest "region, low confidence" is more useful than a precise answer nobody can reproduce.
Frequently asked questions
- What is the first step in geolocating a photo?
- Check the file’s metadata. If GPS coordinates survived, you have an answer accurate to a few metres and nothing else is needed. It takes seconds and is the step most often skipped.
- How do you avoid confirmation bias?
- Verify using details you did not use to form the hypothesis. Matching the same clues that produced the guess is circular reasoning, and it is the most common error in amateur geolocation.
- Is OSINT geolocation legal?
- Analysing published images is generally lawful. Building a picture of a specific person’s movements is not, in most jurisdictions — where the line sits is worth reading before you start.
- When should you stop?
- When you have a place, or when the evidence runs out. If the next step would be about a person rather than a location, that is the stopping point regardless of what is technically possible.
Sources
- Berkeley Protocol on Digital Open Source Investigations — UN OHCHR / UC Berkeley Human Rights CenterEstablished methodology for open-source investigation, including verification standards.
- GPS Accuracy — GPS.govThe 4.9 m figure that makes step one worth running first.
Keep reading
How to Identify a Country From Photo Clues
Regulated things beat natural things. The clues that change abruptly at borders, ordered by how much each one narrows the search and how fast you can read it.
· 4 min read
Is It Legal to Find Where a Photo Was Taken?
Reading an image is rarely the problem — what you do with the result is. Where the line sits between analysis, research and conduct that is actually unlawful.
· 5 min read