Most people, asked to do a reverse image search, drop the photo into Google Lens and call it a day. Sometimes that works. More often, especially for verification work, it gets you the wrong answer or no answer at all. The reverse-image-search engines have very different strengths, and learning which one to reach for in which situation is the difference between an investigation that produces a confident finding and one that doesn’t.
Here’s the working knowledge a verification analyst would have. Tool-by-tool, then a quick decision tree.
The engines that matter
Google Lens / Google Images
The default. Best in class for: products (it’ll identify a specific shoe model from a partial photograph), text in images (excellent OCR with translation), well-known landmarks, and images that have circulated widely on the open Western web. Use Lens through the Google app on mobile or images.google.com on desktop.
Where it fails: photographs that haven’t been indexed widely (small forums, regional press, anything outside the English-speaking web), faces (Google deliberately suppresses face matching), and most non-Latin-alphabet contexts.
Yandex Images
The single most powerful reverse-image-search engine for OSINT work, by a wide margin. Yandex’s index is much deeper into the Russian-language and Eastern European web than Google’s, and its image-similarity algorithm is unusually good at matching photos taken in the same place from slightly different angles. If you’ve geolocated something in Russia, Eastern Europe, Central Asia, or the Caucasus, Yandex is the first tool you reach for. Available at yandex.com/images.
Yandex also returns a "places" panel that suggests likely geographical contexts based on the visual content of the image. This panel is famously useful for street-view and architectural matching, and it’s the one feature Google has never replicated.
TinEye
The original. TinEye specializes in finding the first time an image appeared online and tracking its propagation. It’s not great for "where is this?" but excellent for "is this image as old as the source claims?" If you suspect an image is being recycled from an older event, TinEye is the right tool. The free tier is enough for most use; paid plans are available for volume work. tineye.com
TinEye’s "sort by oldest" and "sort by most changed" results are unique to the platform and are why journalists keep coming back to it.
Bing Visual Search
Underrated. Microsoft’s image index is different from Google’s in non-trivial ways, and Bing sometimes surfaces matches the others miss, particularly for stock photography and corporate imagery. Worth running as a third pass after Yandex and Google. bing.com/images
Baidu, Naver, SOGOU
Region-specific. If you’re working a Chinese-language case use Baidu (qihoo’s image search is also worth a try). For Korean cases, Naver. These tools have indexed parts of the regional web that are simply not in Google or Yandex.
PimEyes, Clearview, FaceCheck.ID
Face-search engines. Drop a face, get a list of other places that face appears on the public internet. This category is operationally powerful and ethically loaded. PimEyes is the consumer-facing example; FaceCheck.ID is the cheaper alternative; Clearview is law-enforcement-only.
I’m going to be direct: the legitimate uses of face-search tools are narrow. Reverse face search of yourself (to know what’s out there about you) is fine. Investigative journalism into public figures acting in public roles, with editorial review, can be defensible. Identifying a person from a public crime scene image is law enforcement’s job, not yours. Almost everything else, including the marketing pitch most of these tools make about "finding people you’ve lost touch with," shades into territory where the subject of the search has not consented and would object if asked.
PimEyes has been sued, fined, and ordered to delete data in multiple jurisdictions. The European Data Protection Board, the UK ICO, the French CNIL, and the German DPAs have all issued findings against face-search services as a category. If you use these tools, document your purpose, document your subject’s status as a public figure if you’re going to claim that defense, and consult counsel before publishing.
I list these tools here because pretending they don’t exist doesn’t help analysts make good decisions about them. Most readers should not use them outside narrow circumstances.
The decision tree
When I get a photograph and a question, the question I ask first is: what kind of answer do I want?
- "What is this object?" → Google Lens, then Bing.
- "What does this text say?" → Google Lens (OCR + translate), then Yandex.
- "Where was this taken?" → Yandex first, then Google, then a regional engine if the photo’s geographical hints point that way.
- "How old is this image / has it been published before?" → TinEye first, then Yandex with sort-by-date, then a manual archive search at the Wayback Machine.
- "Has this photograph been edited?" → TinEye for older versions, plus a side-by-side using a tool like InVID’s image verification toolkit, which is a free browser extension that runs the standard forensic checks (error level analysis, metadata, magnifier).
- "Who is the person in this photograph?" → most of the time the right answer is to leave that question alone. If you have a documented legitimate purpose and the subject is public-acting, the technique is documented in this Bellingcat piece, but the bar is high and the consequences of getting it wrong fall on a real person.
The trap
The trap is getting one hit on the first engine you tried and stopping. A single match is a candidate, not a finding. Run the image through three engines and look at the convergence. If Yandex, Google, and TinEye all surface the same source for an image, you can be reasonably confident in that provenance. If only one does, you have a lead, not a fact.
A short workflow you can use today
- Open the image in InVID’s browser plugin. Note the EXIF, the size, the format, anything obviously edited.
- Run reverse search on Google Lens. Save the top three matches.
- Run on Yandex Images. Save the top three matches and the "places" panel suggestions.
- Run on TinEye, sort by oldest. Save the earliest match found.
- Compare. Note where the engines agree and where they don’t. Note dates. The finding is the synthesis of all four passes, not the output of any one.
Twenty minutes per image, faster as you build the muscle. It is the single workflow that I’d most like every social-media editor in the world to internalize, because most of the misinformation that goes viral does so on the back of an unverified image with a misleading caption, and the technique to debunk it has been free and public for fifteen years.