Geolocation for Journalists: How to Verify Where an Image Was Taken

When the video surfaced in 2018, the Cameroonian government dismissed it as fake. The footage showed soldiers executing two women and two young children, and officials insisted it had not even been filmed in Cameroon. A team working with BBC Africa Eye set out to test that claim using nothing more than publicly available tools. By matching the ridgeline of the mountains in the background to Google Earth, and then lining up a bend in the dirt track and nearby buildings with satellite imagery, they pinned the killings to a specific location in the country’s Far North region. The resulting investigation, Anatomy of a Killing, went on to win a Peabody Award, and four soldiers were later sentenced to ten years in prison.

That is the power of geolocation. For journalists working under governments that deny, deflect, and label inconvenient evidence “fake news”, the ability to independently establish where and when an image was captured is no longer a niche skill. It is a core defence of the public record.

This guide sets out the essential techniques, the free tools that open-source investigators rely on, and the ethical safeguards every reporter should apply before publishing.

What geolocation is (and what chronolocation adds)

Bellingcat founder Eliot Higgins describes geolocation as using the clues inside a photo or video to find the precise spot where it was captured, either confirming the location claimed by whoever shared it, or establishing a location when none is given.

Its companion technique is chronolocation: working out when an image was taken. The two are usually done together. Location tells you whether a piece of footage shows what it claims to show; timing tells you whether it is genuinely new or recycled from an earlier event.

Before you start: preserve the evidence

Online content disappears, posts are deleted, accounts are suspended, and platforms remove material. Before analysing anything, save the original file, the post URL, the account details, and a timestamped screenshot, and archive the page.

The Berkeley Protocol on Digital Open Source Investigations, published by the UN Human Rights Office and the Human Rights Center at UC Berkeley, sets international standards for gathering, analysing and preserving digital information in a professional, legal and ethical way. It deliberately avoids prescribing particular tools, focusing instead on principles such as clear record-keeping and basic operational security. Even if your story never ends up in a courtroom, following its logic- collect, preserve, verify, then analyse- makes your work far easier to defend.

A note on metadata: EXIF data embedded in a photo can contain GPS coordinates and a capture time, so it is always worth checking. But don’t count on it. As Bellingcat notes, most images open source researchers handle carry no EXIF data, because many social media platforms strip it automatically on upload. Metadata can also be edited, so treat it as a lead, never as proof.

A step-by-step geolocation method

1. Read the image before you search

Start by listing everything visible: the language and script on signs, shopfronts, phone numbers, licence plates, road markings, the side of the road traffic drives on, architectural styles, vegetation, terrain and soil colour. Small infrastructure details often narrow a search dramatically, GeoHints, listed in Bellingcat’s toolkit, catalogues regional variations in items such as traffic lights, utility poles and bollards.

Distinctive landmarks are your best anchors. In his early guide to geolocating videos from Syria, Higgins showed how a mosque’s minaret could be matched to satellite maps, noting that minarets are frequently a useful reference point in footage from the Middle East.

2. Run a reverse image search

Someone may already have published the same image, or a near-identical view, with a location attached. Run the image, or key frames from a video, through several engines, such as Google Lens and TinEye, since each indexes different material. Crop to individual features (a building, a sign) if a search of the full image returns nothing useful.

3. Narrow the search area with map data

When you know the general region but not the exact spot, map databases can generate candidate locations. Bellingcat’s OpenStreetMap Search tool lets you look for combinations of features that appear near each other, say, a church, a petrol station, and a railway line within a few hundred metres. Overpass Turbo, a web-based tool for querying OpenStreetMap, was highlighted by data journalists at NICAR 2026 as a quick route to geolocation leads. Remember that OpenStreetMap coverage is uneven: a feature missing from the map is not proof that it doesn’t exist.

4. Match the scene on satellite and street-level imagery

Once you have candidates, compare them with satellite and street-level views. Google Earth remains the workhorse, and its historical imagery lets you see how a site looked in previous years. Look for the relationships between features, the angle of a road, the spacing of buildings, the shape of a skyline, rather than any single object.

Be exhaustive. Describing the Cameroon investigation, Ben Strick of the Centre for Information Resilience explained that the team tried to match every tree they could find in the satellite image, so the conclusion would be as hard as possible to dispute. One matching feature is a hypothesis; several independent matches are a finding.

Using the sun and shadows

Shadows are one of the most powerful and most misunderstood sources of evidence. The direction and length of a shadow depend on where the sun is in the sky, which in turn depends on location, date, and time.

In the Cameroon case, analysing the angle of sunlight allowed investigators to date the footage to between 20 March and 5 April 2015, three years before it went viral.

Satellite imagery and change over time

Comparing satellite images from different dates can tell you when something happened. In the Cameroon investigation, a building visible on satellite imagery from one year but absent from another helped narrow the window in which the video could have been filmed.

Can AI do it for you?

Not reliably, at least not yet. Bellingcat has tested AI chatbots on geolocation several times. In 2025, it ran 500 tests on 20 models using the same 25 photos and benchmarked them against Google Lens. A follow-up covering 24 models found Google’s AI Mode the most capable overall, but Bellingcat’s conclusion was blunt: users should not rely solely on LLM answers.

AI tools can suggest leads and spot small details faster than a human, but they can also state wrong answers with total confidence. Treat any AI output exactly as you would an anonymous tip: a starting point that must be independently confirmed with the methods above.

Ethics and safety

Geolocation can expose people as well as facts. The Engine Room’s guide to responsible open source investigations warns that this research can reveal the identity and location of witnesses, victims or perpetrators in near real time, potentially putting them in danger. Before publishing, ask:

  • Does revealing the precise location put the person who filmed it, or anyone shown, at risk?
  • Would a less precise description (a neighbourhood rather than a street address) serve the public interest just as well?
  • Has the person who captured the footage consented to its use, and do they understand the risks?

Protect yourself too. Research by the Stanley Center found that analysts and journalists working with open-source material regularly worried about vicarious trauma from repeatedly viewing graphic content. Limit exposure where you can, work with audio muted, view footage at reduced size, take breaks, and talk to colleagues. Basic operational security matters as well: the Berkeley Protocol also covers how investigators can protect the digital, physical, and psychosocial safety of themselves and others.

Show your work

Open source investigation is persuasive precisely because it is reproducible. Strick argues that showing audiences how a finding was reached, and which public tools were used, lets them retrace the steps and check the conclusion themselves. Publish side-by-side comparisons, annotated screenshots, and coordinates where it is safe to do so. When officials claim footage is fabricated, transparent methodology is often the most effective reply.

Quick checklist

  1. Archive the original file, URL and account details before doing anything else.
  2. Check metadata, but treat it as a lead, not proof.
  3. List every visual clue: text, infrastructure, landmarks, terrain, vegetation.
  4. Run reverse image searches across several engines.
  5. Narrow the area with map data, then match features on satellite and street-level imagery.
  6. Confirm with multiple independent features, not just one.
  7. Use the sun and shadows to test the claimed time and, where useful, the location.
  8. Compare satellite images from different dates to establish a time window.
  9. Never rely on an AI answer without independent confirmation.
  10. Weigh the risks to sources and subjects before publishing exact locations.
  11. Document and publish your method.

Where to learn more

Multiply our Impact: