Reverse image search for AI images

What a reverse image search does

A reverse image search starts from a picture instead of words. The search engine looks for copies of that picture, cropped or resized versions of it, and images that look similar, then shows you the pages where they appear.

When you want to know whether an image is AI, that history often answers the question before you look closely at a single pixel. A search can turn up:

• the original post, where the creator may say the picture was made with AI, or even share the prompt;
• a photographer's page, an agency or a news report that credits a real person with the shot;
• other photos of the same moment, taken by other people from other angles;
• a fact-check that someone has already written about the picture;
• an earlier date showing that the picture predates the event it supposedly shows.

Fact-checkers use it as their first move for a reason: it is free, it takes seconds, and it settles many questions outright.

Which search engine to use

Each engine indexes different parts of the web, so the same picture can turn up different results. When the answer matters, try more than one.

• Google Lens (and Google Images): the most widely used. It finds exact matches, similar images and the pages that use them. Its "About this image" panel can show when Google first saw an image and where else it appears.
• Bing Visual Search: Microsoft's equivalent, in Bing Images. It can surface pages Google doesn't, and lists the pages that include the image.
• TinEye: looks only for copies of the same picture, including resized and cropped ones. You can sort its results by oldest, which makes it handy for finding a first appearance.
• Yandex Images: covers some sites and regions the others cover less, and is worth a try when the others find nothing.

None of them sees every picture online, and none sees inside private chats or closed groups. Treat each search as a sample of the web, not the whole of it.

Step by step

1. 1

   Start with the best copy

   Use the largest version you can find, not a thumbnail or a screenshot of a screenshot. If someone has added a border, a caption or a logo on top of the picture, crop it off first.

2. 2

   Search the whole picture

   Upload the image, paste it, or paste the image's own address (not the address of the post it sits in). Look through the exact matches first, then the similar images.

3. 3

   Search the details

   Crop to one distinctive element and search again: a face, a sign, a landmark, a vehicle, a logo. A crop often finds the original when the whole picture has been reframed, combined with another or altered.

4. 4

   Sort by date

   Look for the oldest copy. TinEye can sort by oldest, and on Google, "About this image" can show when an image was first seen. The earliest copy is usually the closest to the source.

5. 5

   Open the results

   Read the pages themselves, not just the thumbnails. The caption, the date, the account and the replies often matter more than the picture.

6. 6

   Write down what you found

   Note the earliest link, its date and who posted it. If you are checking for work, keep screenshots, because posts get deleted.

How to read what you find

Results need reading, not just counting. Some findings settle the question; others only feel as if they do.

Read the first copy carefully. AI art accounts often say what they used in a caption or a hashtag, and that text is lost as the picture is shared onwards. Once you find it, you usually have your answer.

What a reverse image search can't do

A reverse search tells you about a picture's history, not about how it was made. Its limits matter most with AI images:

• A new AI image has no history. If someone generated it an hour ago, there is nothing to find, and an empty page of results is all you get.
• It can't see private spaces. Pictures that start in messaging apps, closed groups or disappearing stories may never be indexed.
• It doesn't read labels. A search engine looks at what the picture shows, and a re-shared copy has usually lost its Content Credentials anyway.
• It can't tell an edit from an original unless it finds both versions, and then you have to compare them yourself.

That is why a reverse search and a detector work well together. One looks outward, at where a picture has been; the other looks inward, at the picture itself.

On a phone

You don't need a computer. Both big phone platforms can run a reverse image search in a few taps:

• Android: with Circle to Search, long-press the home button or the navigation bar and circle the image. On other phones, open the picture in the Google app or Google Lens.
• iPhone: in the Google app, tap the Lens icon and choose the picture from your photos. In Chrome, press and hold an image on a web page and choose to search it with Google.
• From a chat: save the picture to your phone first, then search it from your photos.

Our guide to checking a photo on your phone covers the other built-in tools, and how to check any picture with Perspicax from your phone's browser.

Then check the pixels

When the search comes back empty, or the trail goes cold, check the picture itself. Perspicax combines proven computer vision techniques with TrueEye 2.3, our state-of-the-art detector. It examines every region of the image and answers in one of three bands:

It needs no history and no label, so it works on the brand-new pictures a reverse search can't find, and on the re-shared copies that have lost their details. In testing, it caught 98.6% of images from the newest OpenAI, Google and ByteDance models on average, and wrongly flagged fewer than 1 in 2,000 original camera photos.

Used together, the two checks cover each other's gaps. For the routine to run on a forwarded picture before you pass it on, see check a photo before you share it; for pictures in the news, see how to verify a news photo.