Face search for journalists and researchers

What you are up against

• Recycled facesA profile photo attached to an eyewitness, a whistleblower or a viral account that turns out to be a model, a stock image or a real person who has nothing to do with the story. Reverse image search misses it when the picture has been re-cropped or re-shot from a screen.
• Photos without provenanceAn image arrives by message with no date, no location and no author. You need to know where it has appeared before and in what context before it can carry any weight.
• Coordinated fake accountsNetworks of accounts that share a small pool of faces across platforms. Spotting the shared faces is often the quickest way to see that ten voices are one operator.

Reverse image search engines such as Google Lens, TinEye, Yandex and Bing remain the first tool for provenance, because they find copies of the exact picture and often the earliest one. A reverse face search complements them: it matches the face across different photos, so it finds the same person in a different picture under a different name. The comparison of the two explains when each is the right tool, and Bellingcat's reverse image search guide remains a solid grounding in the image-search side.

Three workflows

1. Verify a source photo

Before a profile photo or a supplied portrait goes anywhere near a story, crop to the face and run it. Three outcomes matter: the face appears under a consistent name on pages that predate your contact (supports the identity), the face appears under a different name or as a stock or agency image (undermines it), or nothing appears (tells you little; many real people have no public photos). Treat the first as supporting evidence to be corroborated by other means, never as confirmation. The step-by-step version is on verify the person you are dealing with.

2. Trace where an image came from

Run the reverse image engines first to find exact copies and the earliest publication. Then run a face search on each clearly visible face to find other photos of the same people, which often leads to the event, the venue or the original photographer. The how to reverse image search guide walks through Google Lens, TinEye, Yandex and Bing alongside FaceSearch. Record every URL and the date you saw it; source pages disappear.

3. Vet a tip-off profile

When a tip arrives from an account you do not know, search the account's photos before you invest reporting time. The same face across several accounts with different names and biographies is the classic signature of a sock-puppet network, and a face that belongs to an unrelated private person is a sign the account is not what it claims. What you do with that finding is editorial: it is a reason for caution, not a publishable fact about anyone.

OSINT hygiene

• Work from copies. Keep the original file untouched and search a cropped copy, so your evidence chain is intact.
• Log what you searched, when, and what came back, with screenshots of source pages. Results change as pages are published and removed.
• Use a dedicated account for investigative work so search history is separated from personal use, and delete scans you no longer need.
• Only upload material you have a right to use. A photo supplied by a source for verification is fine; a photo lifted from a private group is not.
• Remember that a face search is itself a privacy-sensitive act. Search the fewest faces needed to answer the question, and do not search bystanders.

Which plan fits

Rules that apply to you

• A match is a lead. It tells you that a similar face appears on a public page. It does not establish identity. Corroborate with documents, direct contact or independent witnesses before anything is published.
• Never identify a private individual from a face match alone. Look-alikes are common enough that doing so is a serious error waiting to happen; see the doppelgänger effect.
• Do not search minors, and do not search victims or witnesses without a clear editorial need and your editor's agreement. Our acceptable use rules prohibit searches of anyone under 18 and any use for harassment or exposure.
• Know the law where you and your subjects are. Biometric and data-protection rules differ between the US, the EU, the UK and India, and journalistic exemptions vary. The legal guide is a starting point, not legal advice.
• Read results as a tool output, not a database record. Ranking reflects visual similarity, nothing else; read how to read your results before you rely on one.