Skip to content

Tech explained for non-geeks

Is that photo fake?

Written by Daniel Gélinas, checked against primary sources. Last checked .

The short answer

Often you can't tell by looking. New AI pictures usually get the hands right, and AI detector tools get fooled too. So don't ask “does it look real?”, ask “where did it come from?” Search with the photo itself (Google Lens, Bing or TinEye) to see where else it's been, and follow it back to the oldest post you can find.

The video

Watch it on YouTube

It's all over the family chat: a shark, swimming down a flooded highway, past the cars. Aunt Jo shared it, and now everyone's sharing it.

“Is that photo fake?” asks Dan.

“Honestly?” says Liz. “You can't trust your eyes on this anymore.”

Why looking doesn't work anymore

A few years ago there were easy giveaways: six fingers, melted hands, garbled writing. New AI pictures usually get those right. Hands can still go wrong now and then, and odd text or backgrounds can still be a clue, but a picture that passes every check can still be fake.

And people are not as good at spotting fakes as they think:

  • In a Microsoft experiment with about 12,500 people and nearly 300,000 pictures, people were right 62% of the time, which the researchers called “only slightly higher than flipping a coin”. Landscapes and street scenes, like a flooded highway, fooled them most.
  • In a 2022 study of faces, people were right only 48% of the time, about chance. Training got them only to 59%.

Can't a detector website tell me?

Not reliably. “AI detector” sites look at a picture and give you a score, but they guess too. A 2026 study tested 16 detection methods on 2.6 million images: on pictures from today's popular image generators, they were right only 18–30% of the time on average, and none was best on everything.

So a detector saying “Real” doesn't make a picture real, and one saying “AI” doesn't make it fake. Treat it as one more guess, not an answer.

Ask a different question

So don't ask “does it look real?”. Ask “where did it come from?”

A fake photo almost never stays where it started. It gets shared, then shared again, and somewhere along the way it loses its caption, its credit, or the label that said what it was. Follow it back, and the answer is usually waiting at the start.

In the video, the shark went from Aunt Jo, who got it from a friend, who got it from a big page with millions of followers. Asking around only gets you so far. The quick way is to search with the photo itself.

Search with the photo itself

This is called a reverse image search: instead of typing words, you give the search engine the picture, and it finds other pages with the same one. First save the photo, or take a screenshot of it (crop it tight around the picture).

Google Lens

PhoneHow
AndroidOpen the Google app or Chrome, tap the Google Lens icon in the search bar, then pick your photo or screenshot. For a picture on a website, touch and hold it, then tap Search image with Google Lens
iPhoneOpen the Google app (or Chrome or Safari), tap Google Lens in the search bar, then tap the photo picker and choose the photo

Scroll down past the first answers to the pages that show the same picture. Where Google offers it, About this image also shows when Google first saw the picture, and older pages that used it. It isn't available in every country.

Bing and TinEye

  • Bing: in the Bing app, tap the visual search icon, then take or pick the photo. On a computer, look for the visual search icon in the Bing search box.
  • TinEye (tineye.com): upload the photo, then sort the results by Oldest. That puts the earliest copies TinEye found at the top.

Try more than one. Each one knows about different pages.

Follow it back to the oldest post

Look through the results for the oldest post you can find, and open it. Ask:

  • Who posted it first? An artist's page, a meme account or a satire site is a very different source from a news photographer.
  • What did they say about it? Many viral fakes start as honest AI art, clearly labelled, and lose the label as they're reshared. That's what happened to Dan's shark: the oldest post was a digital artist's page, tagged “Made with AI”.
  • Is it older than the story? If a photo of “today's flood” was already online three years ago, it isn't today's flood.

One honest caveat: the oldest post you can find isn't always the first one ever. TinEye's dates are when its crawler first saw a picture, not when it was posted. If the trail runs out, that isn't proof either way.

Is it news? Check the news

If a photo claims something happened, like a flood, a fire or a famous person somewhere, a real event leaves a trail beyond one picture. Look it up on a news outlet you trust. If nobody reliable is reporting it, wait before you share it.

What about the “cr” label?

Some cameras and apps now attach Content Credentials: a kind of label inside the file saying how the picture was made and edited. Where a site or app shows them, you'll see a small “cr” icon you can tap. Google's Pixel 10 camera adds them to every photo it takes, for example, and Google Photos shows them in a photo's details.

They're useful when they're there, but:

  • No label doesn't mean fake. Most cameras and phones don't add them yet, and screenshots and many apps strip them out.
  • A label isn't a verdict. It says where a file came from and what was done to it, and it's used for real photos and AI pictures alike.

Google's Gemini app can also check a picture for SynthID, Google's invisible watermark, but it only recognizes pictures made with Google's own AI tools. “No watermark found” doesn't mean a picture is real.

The same goes for voices

AI can copy voices as well as pictures, and the advice is the same: don't trust your senses, check the source. See Could that really be my grandson's voice on the phone?

Setting the family chat straight

“So the shark's fake!” says Dan. “I'm telling the family chat.”

No need to make anyone feel silly: these fool everyone. Share what you found (“it's AI art, here's the artist”), and the next person who sees it won't have to wonder.

What you can do

  • Don't trust your eyes, or a detector website, to tell real from fake.
  • Ask where it came from, not whether it looks real.
  • Search with the photo itself in Google Lens, Bing or TinEye, and look at the pages that have it.
  • Find the oldest post you can, and check who posted it and what they said about it.
  • For news, check a news outlet you trust before sharing.
  • No “cr” label doesn't mean fake, and no watermark doesn't mean real.

Sources

  1. Roca et al. (Microsoft AI for Good Lab): How good are humans at detecting AI-generated images? Learnings from an experiment (arxiv.org)

    About 12,500 people judged about 287,000 real and AI-generated images in the “Real or Not” quiz and got 62% right: “only slightly higher than flipping a coin”. They did worst on natural and urban landscapes.

  2. Lancaster University: AI generated faces are more trustworthy than real faces (on Nightingale and Farid, PNAS, 2022) (lancaster.ac.uk)

    People shown real and AI-made faces were right 48% of the time, close to chance. After training with feedback, they reached only 59%.

  3. Ren et al.: How well are open sourced AI-generated image detection models out-of-the-box (2026) (arxiv.org)

    16 detection methods tested on 2.6 million images from 291 generators. Modern commercial generators “defeat most detectors”, with only 18–30% average accuracy on Flux Dev, Firefly v4 and Midjourney v7, and no single detector wins everywhere.

  4. Google Search Help: Search with an image on Google (support.google.com)

    Open the Google app or Chrome (or Safari on iPhone), tap Google Lens in the search bar, and take a photo or pick one. On Android you can also touch and hold an image on a website and tap “Search image with Google Lens”.

  5. Google Search Help: Learn more about an image (support.google.com)

    About this image shows when Google may have first seen the image or similar ones, other pages using it, and pages that appeared much earlier. It may show whether a photo was made by a camera or with AI, but metadata can be modified. Only available in some regions.

  6. TinEye Help: TinEye tutorial, and What does “first found on” mean? (help.tineye.com)

    You can sort results by “Oldest”. The date is when TinEye's crawler first saw the image, not when it was posted, but if an image presented as new was first seen years ago, that's a red flag.

  7. Microsoft Bing: Visual Search (explore.microsoft.com)

    Upload, photograph or paste an image to find similar images and pages; in the Bing app, tap the visual search icon.

  8. Google: How Pixel and Android are bringing a new level of trust to your images with C2PA Content Credentials (2025) (blog.google)

    Pixel Camera on Pixel 10 attaches Content Credentials to every JPEG photo it takes, and Google Photos adds them to images edited with AI tools and shows them in its About panel. Content Credentials are not the sole solution, and people can wrongly believe unlabeled content is “not AI”.

  9. Gemini Apps Help: Verify AI-generated images, videos, and audio (support.google.com)

    Gemini can check for SynthID, Google's invisible watermark, but can currently only recognize content made by Google AI tools; content without it could have been made by other AI systems. Content Credentials are used for both AI and non-AI content and can't tell you a file is definitely AI-generated.

  • What's an AI agent?

    You ask ChatGPT, and it tells you what to buy and where. An AI agent goes and does it, with your browser and your accounts. Here's what that means, and the one habit that keeps it safe.

  • What does AI actually know about me?

    Told a chatbot something personal? What it learned in training is one thing. What you tell it, and what you let it open, is another, and that's the part your settings control.

  • Do I need another AI app?

    That AI app ad keeps following you around. Before you pay, start with what you want it for: lots of phones and email apps can already do it.

Everything in AI

More on YouTube

Dan and Liz answer one everyday tech question at a time, most of them in under a minute.