How to detect fake AI videos: six checks before you believe what you see

The short answer: you detect a fake AI video less by staring at the picture and more by asking where it came from. Who posted it? Is anyone trustworthy reporting the same thing? And what does the video want you to do next? Those three questions catch far more fakes than looking for odd hands or strange blinking — because the pictures keep getting better, and the tricks behind them stay the same.

This article walks through six checks, starting with the ones that work best. You don’t need any technical skill, and you don’t need to become suspicious of everything. You just need a habit.

First, what a fake AI video actually is

An AI-made video — often called a deepfake — is a video that was generated or altered by software so that a person appears to say or do something they never did. The face may belong to a real person, the voice may be copied from real recordings, and the words are whatever the maker typed in.

A few years ago these were clumsy. Today, at the time of writing, the good ones can pass a quick look on a phone screen, which is exactly where most of us see them. That’s why the old advice — “look for blurry edges” — is no longer enough on its own. It’s still useful. It’s just not the first thing to do.

If you’d like the wider picture of how these tools produce convincing things that aren’t true, how to tell when AI is wrong is a good companion to this piece.

Check 1: Who posted it, and where?

This is the strongest check, and it takes ten seconds.

A genuine clip of a famous doctor, politician or investor will come from that person’s own account, or from a news organisation you already know. A fake almost always arrives sideways: a page you’ve never heard of, a friend-of-a-friend’s share, an advert in your feed, or a message from a stranger.

Ask yourself: if this were real, is this really where I’d first see it?

Tap the account name. Look at when it was created, what else it posts, and whether it has a long history or appeared last month with a hundred near-identical clips. New, thin, and busy is a warning sign.

Check 2: Is anyone else reporting it?

If a well-known person really said something remarkable, other people will be talking about it within hours — newspapers, broadcasters, the person themselves. Search the person’s name together with a few words from the video. If the only results are the same clip on the same sort of pages, be very cautious.

Searching the person’s name plus the word “fake” or “deepfake” is also worth doing. The most widely shared fakes are usually identified quickly, and you’ll often find the original interview that was altered.

Check 3: What does it want you to do?

This is the check that protects your money.

A fake video is rarely made for fun. It’s made to move you towards something: click a link, “register” for an investment, buy a miracle product, send a donation, or hate someone. So look past the face and ask what the clip is trying to get from you.

Be especially wary when a video of a trusted figure is paired with an offer — guaranteed returns, a limited-time scheme, a cure. Real experts don’t recruit strangers through social media adverts, and no honest investment promises a fixed high return. We cover this pattern, and what to do if you’ve already clicked, in AI scams aimed at people over 50.

Check 4: Does the mouth match the words?

Now to the picture itself. The mouth is where fakes most often slip, because matching lips to invented speech is hard.

Watch a few seconds with the sound on and your eyes on the lips. Do the lip shapes fit the sounds? Do the teeth look the same from one moment to the next, or do they seem to change or smear? Does the face stay oddly still while the voice is emotional? Does the voice sound slightly flat, evenly paced, or without the little breaths and stumbles of real speech?

None of these prove anything on their own. Poor video calls have the same problems. But if the mouth feels wrong and the clip failed checks 1 to 3, you have your answer.

Check 5: Look at the edges, the hands and the background

Fakes often fool the eye where you’re looking — the face — and fall apart where you’re not.

Watch the edge of the face, the hairline and the jaw when the person turns their head. Glasses and earrings sometimes flicker or drift. Hands can have the wrong number of fingers, or fingers that bend strangely. Text in the background — a sign, a book spine, a caption — may be jumbled or shift from frame to frame. Lighting on the face sometimes doesn’t match the light in the room.

If you can, watch on a larger screen and pause the video. Fakes are made to be seen small and fast. Slowing down works against them.

Check 6: Does it feel designed to make you react right now?

Fakes are built to trigger a feeling — outrage, fear, excitement, urgency — because feeling makes people share before they think. If a video makes your heart race and your finger hover over “share”, treat that reaction itself as a signal. The stronger the pull to act immediately, the more the clip deserves a pause.

That pause is your best defence. Make a cup of tea. Come back in ten minutes. Most fakes cannot survive ten minutes of calm attention.

What if you still can’t tell?

Sometimes you won’t be sure, and that’s fine. You don’t have to decide whether it’s fake. You only have to decide what to do — and the safe answer is the same either way:

Don’t share it. Don’t click the link. Don’t send money or details because of it. If it involves a family member — say, a video call or clip of a grandchild in trouble — hang up and ring them on the number you already have.

If you want a second opinion, show the clip to someone you trust. Saying “does this look right to you?” out loud is remarkably effective. And if the clip claims to be from a real organisation — a bank, a health service, a charity — contact them through their real website or the number on your paperwork, never through anything in the video.

A word about “AI detection” tools

You may see websites or apps that promise to tell you whether a video is AI-made. Some are helpful, but at the time of writing none are reliable enough to trust on their own. They can mark real videos as fake and miss well-made fakes. Use them, if at all, as one more voice — never as the final word. The checks above, especially the first three, will serve you better.

The short version

• Source first: real news comes from real accounts and real news outlets, not from a page you’ve never heard of.
• If nobody else is reporting it, be very cautious.
• Ask what the video wants from you. Money, clicks or outrage are the giveaway.
• Only then look at the picture: mouth, hands, edges, background.
• When in doubt, do nothing. Don’t share, don’t click, don’t pay — and ask someone you trust.