When a dramatic photo or video starts circulating, the first questions are always the same. Is it real? Where was it taken? When? Answering them is called verification, and the most powerful part of it is geolocation: finding the exact place where an image was captured.
This walkthrough describes the workflow used by open-source investigators. It works for news events, due diligence and research alike.
Step 1: Save everything first
Before you analyse anything, preserve it. Posts get deleted, and accounts disappear.
- Download the highest-quality version available.
- Record the URL, the account that posted it, the date and time you found it, and any caption.
- Archive the page with a web archiving service and keep a screenshot.
- Calculate a hash (for example SHA-256) of the file, so you can later prove it has not changed.
The Berkeley Protocol treats this documentation as essential. Without it, others cannot check your work, and evidence may not be usable later.
Step 2: Find the earliest version
Many “new” images are old ones recycled from a different event.
- Run a reverse image search on several engines, because they index different sites.
- For video, extract key frames and search those. Verification toolkits such as the InVID-WeVerify plugin do this automatically.
- Sort results by date and look for the earliest copy. Who posted it, and what did they say it showed?
If you find the same image years earlier, you are done: it is miscaptioned.
Step 3: Check metadata, but do not trust it
Original files can contain EXIF metadata: camera model, time and sometimes GPS coordinates. Most social platforms strip it, and it can be edited. Treat metadata as a lead, never as proof.
Step 4: Read the image like a map
Now look for clues to location. Work from broad to narrow:
| Clue | What it tells you |
|---|---|
| Language and script on signs, shop names, phone numbers | Country or region |
| Vehicles, license plates, which side of the road traffic drives on | Country |
| Architecture, roof styles, power poles, road markings | Region |
| Vegetation, terrain, weather | Climate zone, season |
| Unique landmarks: towers, bridges, mosques, churches, mountains | Specific places |
| Street layout, junctions, building footprints | Exact location |
Write down every clue, even small ones. Geolocation usually comes from combining several weak clues, not one obvious landmark.
Step 5: Match against maps and satellite imagery
When you have a candidate area:
- Use satellite imagery and 3D views to compare building shapes, road layouts and terrain.
- Use street-level imagery where available to match the ground view.
- Look for features that align exactly: the angle between two roads, the position of a tree relative to a wall, the skyline.
- Mark the camera position and its viewing direction on a map.
A good geolocation can be checked by anyone. It shows the original image side by side with the matching satellite or street view, with the same features marked on both.
Step 6: Establish the time
The place is confirmed. Now check when:
- Shadows. The length and direction of shadows depend on the sun’s position. Tools such as SunCalc show the sun’s angle at a given place, date and time, so you can estimate the time of day or test a claimed time.
- Weather. Compare with historical weather records for that place.
- Changes in the scene. Satellite images from different dates show when a building was built or destroyed, which narrows the possible dates.
- Other footage. Several videos of the same event from different angles can corroborate one another.
Step 7: Assess and report
Summarise your findings with a confidence level:
- What you verified (location, date range, original source).
- What you could not verify.
- How confident you are, and why.
- The evidence trail, so others can reproduce it.
Be explicit about what verification does not show. Proving a video was filmed in a particular town on a particular day does not prove who did what in it.
A note on AI-generated images
Generative AI adds a new question: did this scene ever exist? Look for inconsistent details, such as text that does not make sense, impossible reflections or mismatched shadows. Check for content credentials (provenance metadata) where platforms support them. The geolocation workflow is also a strong test. A real place can be matched on satellite imagery. A generated “place” usually cannot.