Reverse image search

Why it matters

The most common fake online is not a doctored image. It is a real image with the wrong caption: a flood from 2013 presented as yesterday, a crowd from a concert presented as a protest, a video-game clip presented as combat footage. Making one takes thirty seconds and no skill. The photo is genuine, so it passes the “does it look real?” test every time (see out-of-context media).

Reverse image search flips the question. Instead of “does this look real?”, you ask “where has this image been before?”. Search engines have indexed billions of pictures; if the photo has been online since 2013, they usually know. The same trick exposes stolen profile pictures on sockpuppet accounts and recycled “eyewitness” videos.

How to do it

1. Get the image. On a phone: long-press the picture and choose “Search image with Google” (Chrome, Android), or share it to the Google app’s Lens. On a desktop: right-click → “Search image with Google Lens”, or drag the file into images.google.com (camera icon) or TinEye. If the post is a screenshot of a photo, crop out the frame first.
2. For video, grab frames. Take screenshots of two or three distinctive moments (a sign, a building, a face) and search those. The InVID-WeVerify browser extension automates this: paste a video link, it extracts keyframes and sends them to several search engines at once.
3. Use more than one engine. Google Lens is good at objects and text, TinEye at finding exact copies and sorting them by date, Bing at faces and at content from Eastern Europe. No single engine sees everything. Two searches take a minute.
4. Sort by oldest. TinEye lets you sort results by “Oldest”. The earliest indexed copy is your lead: open it and read the date, the place and the caption. The original is usually larger and shows more; the version you saw was probably cropped to remove a logo, a date stamp or something inconvenient.
5. Compare. Does the earliest copy’s date and place match the post’s claim? Same event, or just a similar-looking one? Also try the image flipped horizontally: mirroring is a cheap way to defeat lazy searches, and some engines handle it badly.

Example

@civil_streets_now, 1h HUNDREDS OF THOUSANDS on the streets tonight against the government. The media will show you nothing. Share so they can’t hide it. [photo: a huge night-time crowd, stage lights in the distance]

Step one: long-press the photo and search with Google Lens. Lens suggests “concert” and a festival’s name. Step three: the same image in TinEye: 212 matches. Step four: sort by oldest, and the earliest copy is from 2017, in a music festival’s own gallery, with the headliner’s name in the caption. The crop in the post removed the bottom edge, where the stage was visible.

So: the crowd is real, the protest may well be real too, but this picture is not from it. That is the finding, no more, no less. The right comment is “this photo is from a 2017 festival, here is the link”, not “the protest is fake”. For images that have no earlier copy because they were generated last night, see deepfakes.

Common mistakes

• Stopping at “no results”. A fresh photo, a heavy crop or a filter can defeat one engine. Try another, try a tighter crop on the most distinctive object, try the mirror image.
• Trusting the most popular copy instead of the oldest. Popularity tells you who shared it; the date tells you where it came from.
• Reading the engine’s label as the answer. Lens saying “protest” is a guess about what is in the picture, not a verdict about when it was taken. Open the actual earliest page.
• Relying on metadata. Platforms strip EXIF data on upload, and it can be edited anyway. The search trail is the evidence; the file properties are not.
• Overreaching. “The photo is old” means the photo is old. It does not prove the event never happened or that the poster lied on purpose. Say what you found.