Skip to content

Deepfake, decoded

What is a deepfake? A video, photo or voice faked to look like a real person. How they are made, why phone calls are the dangerous ones, and three checks.

By Nick

2 min read

AI Basics Beginner
On this page

A deepfake is a video, a photo or a voice recording made to look like a real person said or did something they did not. The same tools that will draw you a birthday card can, pointed the other way, put words in a politician’s mouth or your mother’s voice on the phone.

How they are made

From examples. A few minutes of someone’s video, or a few seconds of their voice, is enough for a model (an AI system trained on examples) to learn the pattern and produce new footage or speech in that style. It is the same trick as every other generative tool: learn the pattern, produce more of it.

The dangerous one is the phone call

Fake videos of famous people are a news problem. The one that reaches ordinary people is a voice: “Mum, it’s me, I’m in trouble, I need money now.” A few seconds of a relative’s voice from a social video is enough. The fix is not detection; it is a family password agreed in advance.

Three checks before you believe or share

  1. Who posted it? A news organisation, or an account that appeared last week?
  2. Is it anywhere else? A real event is reported by several places. A fake is one clip.
  3. Does the supposed source say so? If a famous person “said” it, is it on their own channels?

None of those involve squinting at the pixels. Ask the AI to do the legwork.

Here is a video, image or claim I have been sent: [describe it or paste the link]. Do not
tell me whether it looks real. Tell me: who originally published it, whether reputable
outlets have reported the same thing, and whether the person it shows has confirmed or
denied it. If you cannot find any of that, say so plainly.
Example reply (trimmed). Your app will word it differently.

A clip of a chef 'announcing' a free cookware giveaway, plus the prompt above

  1. Who published it? I could only trace it to an account created last week, with no earlier posts.
  2. Reported elsewhere? No reputable outlet has reported the giveaway.
  3. Confirmed or denied? The chef’s own channels say nothing about it.

I could not find this anywhere else.

“I could not find this anywhere else” is a complete answer, and it means do not share it. The two-minute check still applies to any numbers.

Common questions

Can I tell by looking?

Less and less. Hands, teeth and blinking used to give them away. The reliable checks are about where it came from, not what it looks like.

Are the detection tools any good?

Some, sometimes. None are reliable enough to trust alone, and the fakes improve faster than the detectors. Source checks beat pixel checks.

Could someone make one of me?

From a few seconds of video or audio, yes, which is why the family-password page exists. For most people the realistic risk is a voice, not a video.

What to do next

Related reading

Last reviewed 9 October 2026 by Nick.

Software changes fast. If something here is out of date, tell me and I'll fix it.

Get new guides by email

Occasional and practical. No spam, no hype, unsubscribe whenever.

Occasional practical guides from UncomplicateAI. You will get a confirmation email before anything else is sent. Unsubscribe any time. Privacy

You are not subscribed until you confirm the email. Unsubscribe any time.