AI-generated images and videos are becoming increasingly realistic. A picture that looks like a normal photograph—or a video that appears to show a real person—may have been created or altered using artificial intelligence.

That makes one old rule especially important:

Don't trust your eyes alone.

There are practical ways to investigate suspicious content, from checking the original source to looking for AI-generated artifacts and using provenance tools such as Content Credentials and SynthID. Google currently allows users to check images and videos for its AI watermark through Gemini, while Content Credentials can provide information about how a file was created or edited.

1. Look at the Small Details

AI-generated media can still contain inconsistencies.

When examining an image, zoom in and check:

  • Hands and fingers
  • Teeth
  • Eyes and pupils
  • Earrings and jewellery
  • Hair
  • Glasses
  • Clothing patterns
  • Reflections
  • Shadows
  • Background objects
  • Signs and written words

AI-generated images can sometimes contain distorted hands, strange-looking text, repetitive patterns or lighting that doesn't make physical sense. Google also recommends checking these types of visual inconsistencies when evaluating potentially AI-generated media.

But don't treat one strange finger or blurry detail as proof.

Real photographs can contain imperfections too.

2. Check the Text in the Image

This is one of the quickest things to examine.

Look closely at:

  • Shop signs
  • Street names
  • Newspaper headlines
  • T-shirt writing
  • Logos
  • Number plates
  • Posters
  • Screens
  • Product labels

If letters appear randomly arranged, words don't make sense or the same text changes between frames of a video, the content deserves further investigation.

However, modern AI systems have become much better at generating readable text, so perfect-looking text doesn't prove an image is real either.

3. Look at Reflections and Shadows

Ask:

Does the lighting make sense?

For example, if a person is standing in sunlight, their shadow should correspond to the direction and position of the light.

Look at:

  • Shadows
  • Mirrors
  • Windows
  • Sunglasses
  • Water
  • Shiny surfaces

AI-generated content can sometimes produce reflections or shadows that don't match the objects creating them.

Again, this is a clue, not proof.

4. Watch the Video More Than Once

Deepfakes can be harder to spot when you're watching a video normally.

Slow it down or pause it.

Look for:

  • Facial features changing shape
  • Hair or earrings suddenly changing
  • Teeth appearing differently
  • Strange blinking
  • Lip movements that don't match speech
  • Skin texture changing between frames
  • Objects appearing or disappearing
  • Backgrounds warping
  • Unnatural movements around the face

Pay particular attention to transitions between frames.

A video may look convincing when played normally but reveal inconsistencies when paused.

5. Listen to the Audio

Deepfakes aren't limited to someone's face.

AI can also generate or manipulate voices.

Listen for:

  • Unnatural pauses
  • Strange pronunciation
  • Robotic rhythm
  • Sudden changes in voice quality
  • Background noise that doesn't match the environment
  • Audio that doesn't quite match the person's mouth movements

If a video claims that a public figure, celebrity, bank official or government representative said something shocking, don't rely on the video alone.

Look for confirmation from credible sources.

6. Check Where the Image or Video Came From

This is one of the most useful checks.

Ask:

Who posted it first?

When was it posted?

Is there an original source?

Are reputable news organisations reporting the same event?

Does the account have a history of posting reliable information?

A viral video without a clear original source deserves more scrutiny than footage that can be traced back to a credible organisation or identifiable eyewitness.

7. Try a Reverse Image Search

If you're checking an image, use a reverse image search such as Google Lens.

You can upload the image or search using it to find:

  • Older versions of the image
  • Similar images
  • Websites where it appeared previously
  • Different captions attached to the same photograph

Google recommends reverse image search as one method for investigating potentially AI-generated media.

This can reveal something important:

The image may be real—but the story attached to it may be fake.

For example, an old photograph from 2020 could be reposted in 2026 with a completely different claim.

8. Check for Content Credentials

One of the more technical—but increasingly useful—methods is checking Content Credentials.

Think of Content Credentials as a digital history attached to a piece of media.

They can provide information about:

  • Where the content came from
  • What tools were used
  • Whether it was edited
  • The history of modifications

The system is based on the C2PA standard, which uses cryptographically signed provenance information. It can be used for both AI-generated and non-AI content.

If credentials are present, they can give you information that simply looking at the picture cannot.

But there's an important limitation.

No Content Credentials does not automatically mean the content is real.

Metadata can be removed or lost when a file is edited, converted, uploaded or shared.

So treat the absence of credentials as “no provenance information found,” not “this is definitely real.”

9. Use Google's AI Verification Tools

Google currently provides AI verification through Gemini.

You can upload an image or video and ask something like:

“Is this AI-generated?”

or:

“Was this created or edited using Google AI?”

Gemini can check for Google's SynthID watermark and can also provide information from compatible Content Credentials.

What is SynthID?

SynthID is Google's invisible watermarking technology.

It embeds a digital signal into AI-generated content that can be detected by Google's verification technology. Google says SynthID is used across supported AI-generated images, videos and audio.

There's an important limitation:

SynthID is primarily useful for identifying content made or edited with Google's AI systems.

If it doesn't detect a Google watermark, that does not prove that another AI system wasn't used.

10. Don't Trust Random “AI Detector” Websites Blindly

You may find websites claiming:

“Upload your image and we'll tell you whether it's AI.”

These tools can sometimes be useful, but don't treat a detector's result as absolute proof.

AI detection is difficult, particularly as image and video generators improve.

A better approach is to combine several pieces of evidence:

Source + visual inspection + reverse search + provenance information + verification tools.

A Simple 60-Second AI Detection Test

When you see a suspicious image or video online, try this:

STOP

Don't immediately share it.

LOOK

Zoom in and inspect faces, hands, text, reflections, shadows and background details.

SEARCH

Use reverse image search to find where the content originally appeared.

VERIFY

Check Content Credentials or use an appropriate AI verification tool.

CONFIRM

Look for independent reporting or evidence from a reliable source.

If the content involves something serious—such as a political statement, financial announcement, celebrity death, emergency or breaking news—wait for independent confirmation before sharing it.

What AI Detection Can—and Can't—Tell You

This is important because AI detection isn't perfect.

A verification tool may tell you that an image contains a watermark associated with a particular AI system. That can be strong evidence of its origin.

But provenance doesn't automatically tell you whether the claim surrounding the image is true.

For example, a genuine photograph can be given a completely false caption.

Likewise, an AI-generated image can be used to illustrate a fictional story without being presented as real.

Authenticity and truth are not exactly the same thing.

OpenAI similarly notes that provenance signals can indicate origin but don't establish that content is accurate, unedited, legally owned or presented in the correct context.

The Biggest Mistake to Avoid

Don't say:

“It looks real, so it must be real.”

And don't say:

“It looks weird, so it must be AI.”

Both assumptions can be wrong.

Instead, ask:

Where did this come from?

Can I find the original?

Can I verify the claim independently?

Does the file contain provenance information?

Do trusted sources confirm it?

Quick AI Detection Checklist

Before sharing a suspicious image or video:

☑️ Check the source

☑️ Zoom in on small details

☑️ Check hands, faces and text

☑️ Look at shadows and reflections

☑️ Watch the video frame-by-frame

☑️ Listen for unnatural audio

☑️ Run a reverse image search

☑️ Check Content Credentials when available

☑️ Use AI verification tools where appropriate

☑️ Look for independent confirmation

The Bottom Line

AI-generated media is getting harder to identify by appearance alone.

The safest approach isn't to become an expert at spotting every strange finger or unnatural face.

It's to become better at verification.

Pause. Investigate the source. Check the details. Search for the original. Use provenance tools. And verify important claims before you share them.

Because sometimes the most convincing part of a fake image isn't the image itself—it's the false story attached to it.