How to Spot AI-Generated Images and Deepfakes -

Quick answer: There’s no single test that catches every AI-generated image or deepfake. The most reliable approach is to combine three things: check for a hidden watermark or Content Credentials with free tools, look closely for visual and audio glitches, and verify the claim through another source before you share it or send money.

AI images and videos have become very convincing. The old giveaways, like six-fingered hands and garbled text, are much less common than they were a couple of years ago. That doesn’t mean you’re helpless; it means you need to rely more on checking where something came from than on spotting flaws by eye.

This guide, updated for October 2026, walks through the free tools that work today, the warning signs still worth looking for, and what to do when a video call or voice note might be fake.

Can you really tell if an image is AI-generated?

Sometimes, but not always. Detection tools are getting better, and big AI companies now add invisible watermarks or labels to what their tools create. But those only help when the creator used a tool that adds them, and when the file hasn’t been altered too much since.

So the honest answer is: you can often confirm that something is AI-generated, but you usually can’t prove that something isn’t. Keep that in mind whenever a tool says it found nothing.

Step 1: Check for a hidden AI watermark with Gemini

Google adds an invisible watermark called SynthID to images, video and audio made with its AI tools. According to Google’s announcement from November 2025, more than 20 billion pieces of AI-generated content had been watermarked with SynthID since 2023. You can check for it for free in the Gemini app.

Google’s help page lists these steps:

Go to gemini.google.com (or open the Gemini app) and sign in.
Click Add files and upload the image, video or audio clip.
Ask something like: “Was this image created or edited by Google AI?”
Read the answer. If Gemini finds SynthID, it means all or part of the content was created or edited with Google’s AI.

The limits matter. Gemini “can currently only recognize content created by Google AI tools,” so an image from another company’s generator won’t be flagged. Files must be 100MB or less, videos under 90 seconds, and there are daily caps of roughly 10 checks each for images, videos and audio.

Step 2: Look for Content Credentials

Content Credentials are a kind of digital label, based on a standard called C2PA, that records how a file was made and edited. Some cameras, editing apps and AI tools add them. Google said in November 2025 that images from its Gemini image model would start including C2PA metadata, with more products to follow.

To check a file:

Open the free Content Credentials Verify tool at contentcredentials.org/verify.
Click to select a file from your device, or drag and drop it in.
Review what it shows: the app or AI tool used, and any edits recorded.

The tool supports common image formats like JPEG, PNG, HEIC and WebP, plus MP4 and MOV video and some audio. The big catch is that labels can be lost. Screenshots, and many apps that compress images when you share them, can remove this information. A missing label is not proof that an image is real.

Step 3: Look closely for visual and audio clues

Watermarks won’t catch everything, so your own eyes still matter. The New York City Department of Consumer and Worker Protection lists these deepfake red flags:

Jerky movements or visual glitches
Lighting that doesn’t match across the face and background
Unusual blinking or odd shadows around the eyes
“Strange word choices, stilted language, and choppy sentences”

For still images, it also helps to zoom in on the details AI still sometimes gets wrong:

Text: signs, labels and number plates that look almost right but don’t make sense.
Hands, teeth and ears: fewer errors than before, but still worth a look.
Backgrounds: objects that melt into each other, or repeated patterns.
Too-perfect skin and lighting: a glossy, airbrushed look across the whole image.

Treat these as hints, not proof. Real photos can look odd, and good AI images can look flawless.

Step 4: Verify the story, not just the image

The most useful question is often “Who’s sharing this, and why?” rather than “Is this pixel-perfect?”

Find the original. Use a reverse image search, such as Google Lens, to see where else the image appears and when it first showed up.
Check trusted outlets. If a photo shows a major event, reputable news sites will usually be covering it.
Look at the account. New accounts, no history, or pages that only post viral content are warning signs.
Read the comments and community notes. Others may already have spotted the fake.

How to spot voice-clone and deepfake video scams

Scammers now use cloned voices and fake video to pose as family members, bosses or officials. The NYC consumer agency lists these warning signs:

You’re contacted out of the blue.
You’re pressured to act immediately, with no time to think.
You’re asked to pay by wire transfer, gift cards, payment apps or cryptocurrency.
You’re asked for personal or private information.
You’re told to keep the request secret.

What to do:

Ask a personal question that only the real person would know.
Hang up and call them back on a number you already have.
Take your time. Real emergencies can survive a five-minute check.
Agree on a family “safe word” in advance for emergency calls.
Report fraud. In the US, use reportfraud.ftc.gov. In India, use the government’s National Cyber Crime Reporting Portal.

Quick comparison: which check should you use?

Method	Best for	Main limitation
Gemini SynthID check	Images, video and audio made with Google AI	Doesn’t detect other companies’ AI
Content Credentials Verify	Files from apps and AI tools that add C2PA labels	Labels are often stripped when shared
Visual inspection	Quick first look at any image or video	Good fakes may show no flaws
Reverse image search and source check	Viral photos and news claims	Brand-new fakes may have no history yet

Staying safe with AI also means being careful about what you give it. See our guide on how to stop AI chatbots training on your data, and find more practical guides in our AI section.

Frequently asked questions

Is there a free AI image detector that works on everything?

No. Free tools like Gemini’s SynthID check and the Content Credentials Verify tool only work when the content carries their watermark or label. Third-party “AI detectors” can be wrong in both directions, so don’t rely on a single result.

Can Gemini tell if a ChatGPT image is AI-generated?

Not through SynthID. Google says Gemini can currently only recognise content made with Google AI tools. For images from other generators, try the Content Credentials Verify tool, which reads C2PA labels if they’re present.

Does a screenshot remove AI labels?

A screenshot creates a new file, so Content Credentials attached to the original usually won’t come with it. Invisible watermarks like SynthID are designed to be harder to remove, but no method is perfect.

What should I do if I get a call that sounds like a relative asking for money?

Don’t send anything during the call. Hang up, call the person back on a number you already have, or ask a question only they could answer. If it’s a scam, report it to your national cybercrime authority.

How to Spot AI-Generated Images and Deepfakes