How to Check Whether a Photo or Video Might Be AI-Generated
A video appears in a family group chat.
It shows a well-known television personality promoting an investment opportunity. The face looks familiar, the voice sounds convincing, and the video includes a company logo and a website address.
Someone asks:
“Is this real?”
That question is getting harder to answer from appearance alone.
AI-generated and AI-altered photos, videos, and audio can now look convincing enough that a quick glance may reveal very little. Spotting one strange finger or one awkward blink is no longer a reliable strategy.
What matters more is learning how much evidence a claim deserves before you believe it, act on it, or pass it along.
A Convincing Video Can Still Have Weak Evidence
A polished video may look authentic even when there is little evidence behind it.
Natural facial movement, consistent lighting, and a familiar voice can make something feel trustworthy, but none of those details tells you where the video came from or whether the claim is true.
A better place to start is outside the video.
Is there an original source? Does the person shown acknowledge the announcement? Does the company mentioned in the video confirm it? Are reliable independent sources reporting the same event?
The FBI advises people to verify unusual images, video, and audio through trusted or official sources rather than relying only on what appears on the screen.
A well-produced video can be convincing. That alone does not make it authentic.
Visual Clues Can Raise Suspicion, but They Rarely Settle the Question
AI-generated content can still contain mistakes.
You may notice unusual hands, strange facial movement, mismatched lighting, changing details, speech that does not quite match the mouth, or background text that shifts shape.
Those details are worth noticing, but they should be treated cautiously.
A genuine image can look strange because of compression, motion blur, poor lighting, editing, or an unusual camera angle. An AI-generated image, meanwhile, may show none of the classic mistakes people have learned to watch for.
Visual oddities are useful when they prompt a closer look. They are much less useful as proof.
The Source Carries More Weight Than the Face
Suppose the investment video was forwarded by a relative.
Your relative may have received it from someone else, who received it from another account. You know where the video reached you, but not where it began.
When possible, trace the post back to the original account or website.
Is it really the public figure's known account? Is it the company's official site? Or is it an account using familiar names, logos, and images without a clear connection to the person shown?
Experts use the word provenance for information about where digital content came from and what happened to it along the way.
For everyday use, the idea is simple: can the content be traced back to a source that is in a position to know whether it is genuine?
If not, the evidence is weaker than the video may appear.
A Real Event Usually Leaves More Than One Trace
Suppose the video claims that the television personality has launched a new investment program.
That is not just an image. It is a claim about something that supposedly happened in the real world.
Real events often leave evidence in more than one place.
- an announcement on the person's official website,
- a company press release,
- reporting from established news organizations,
- another recording of the same event,
- or information from a regulator or institution connected to the claim.
The important question is whether the event can be confirmed independently.
One of the most useful habits is to stop asking a single video to prove itself and look for evidence elsewhere.
If a dramatic event supposedly happened yesterday but the only results you can find are accounts reposting the same clip, that does not prove the video is fake. It does, however, leave the claim with weak independent support.
What Stronger and Weaker Evidence Looks Like
Consider the same investment video.
| What you notice | Evidence strength | Why |
|---|---|---|
| The mouth movement looks slightly odd | Weak | Compression or editing could cause it |
| One hand looks unusual in a single frame | Weak | A bad frame does not prove AI generation |
| The video is missing from the person's official account | Moderate | It raises questions but does not settle the issue |
| The investment program is absent from the company's official website | Stronger | The claim and the official source do not match |
| Reliable independent sources cannot confirm the announcement | Stronger | The event has little outside support |
| The video pressures you to send money immediately | Risk is much higher | The consequence of being wrong is serious |
The lesson is not that one row proves the answer. It is that some evidence deserves more weight than other evidence.
A strange-looking hand should not outweigh several independent confirmations, and a polished face should not outweigh the absence of any credible source.
Try Looking for the Same Image or Claim Somewhere Else
For a photo, an image-search tool such as Google Lens can help you see whether the same image appeared earlier or in a different context.
For a video, you can capture a clear frame and search that image.
This does not directly prove whether the content was created by AI. What it can reveal is where the image has appeared before, whether it has been reused, or whether the surrounding claim has changed.
You can also search the claim itself without using links supplied by the suspicious post.
If the video says:
“John Smith launches ABC Investment Program today”
search those words separately.
Then compare what you find.
Does the company's official website mention it? Does the person shown mention it? Is the same image attached to an older, unrelated story? Has the clip been reposted with a different description?
The purpose of the search is to find out whether the claim has a history and context outside the post that brought it to you.
The Consequence Changes How Much Evidence You Need
Not every questionable image deserves the same level of investigation.
A funny AI image of a dog wearing a suit has little consequence. A video asking you to transfer money is different.
As the possible consequence becomes more serious, the evidence should become stronger.
Be particularly cautious when content asks you to:
- send money,
- invest,
- provide banking information,
- enter a password,
- reveal a verification code,
- install an application,
- click an unfamiliar link,
- or act immediately.
If that pressure arrives through an unexpected message, our guide How to Tell If a Text Message Is a Scam explains how urgency, unfamiliar links, and requests for sensitive information are commonly used together.
If a QR code is being used to send you to a payment or login page, see How to Know If a QR Code Is Safe Before You Scan It for a different way to check the destination before trusting it.
Suppose the investment video says:
“This offer closes tonight. Click below and invest now.”
At that point, determining exactly how the face was generated matters less than deciding whether the evidence is strong enough for the action being requested.
If the consequence of being wrong is large, weak evidence is not enough.
AI Detectors Can Help, but They Should Not Make the Decision for You
Some websites claim to detect AI-generated images or videos.
These tools can provide another piece of information, but they can also disagree.
Results may vary depending on how the image was generated, edited, resized, compressed, or reposted.
The National Institute of Standards and Technology has examined synthetic-content detection, watermarking, and provenance systems while also documenting technical limitations and open challenges.
A detector result is best treated as one piece of evidence rather than a final answer.
A high “AI likelihood” score should not automatically outweigh clear information about the original source. Likewise, a “likely real” result does not prove that the claim attached to the image is true.
A real photograph can be shared with a false caption, while an AI-generated image can be used in a harmless and clearly labeled way.
Authenticity and truth are connected, but they are not identical.
Sometimes the Correct Answer Is “I Don’t Know”
There will be situations where you cannot prove whether a photo or video is AI-generated.
The image may look normal. The original source may be difficult to find. News coverage may be incomplete. An AI detector may give an uncertain result.
In that situation, you do not have to force the evidence into a simple REAL or FAKE answer.
Sometimes the most accurate conclusion is:
I do not have enough evidence to know.
You can still decide what to do next. You can avoid sending money, wait before sharing, look for better confirmation, or simply leave the claim unproven.
The Most Useful Skill Is Not Better Eyesight
AI-generated photos and videos will continue to improve, and obvious visual mistakes may become less common.
That makes one habit more durable than any checklist of strange fingers, awkward blinking, or imperfect shadows: match your confidence to the evidence.
A visual oddity may justify a closer look, but a traceable source and independent confirmation deserve far more weight. When a photo or video asks you to take a serious action, the standard of evidence should rise with the consequence.
You may not always be able to answer:
“Was this made by AI?”
But you can often answer a more useful question:
“Do I have enough evidence to trust this?”
That is the question worth asking.
