How to Spot AI-Generated Content and Deepfakes

Last month a cousin of mine forwarded me a video. A well-known politician, standing at what looked like a podium, was promising free money to anyone who registered through a link by the end of the week. The lighting was good. The voice sounded right. My cousin wanted to know if it was real, because a friend of hers had already clicked the link.

It wasn't real. The voice was cloned, the face was stitched onto borrowed footage, and the link led to a form that harvested phone numbers and mobile-money details. But here's the part I want you to sit with: she didn't ask me whether it was real before she shared it. She asked after. The video had already traveled through three group chats.

That gap — between sharing and checking — is where almost all of the damage happens. And in the next few years, it's going to get wider unless we change how we think.

Why this matters now

For most of my life, "seeing is believing" was a reasonable rule. A photo was hard to fake well. A voice recording was hard to forge. Video was nearly impossible. That world is gone, and it left quietly.

Today, text, images, voice, and video can all be generated convincingly by software that anyone can use. Not just intelligence agencies or Hollywood studios — ordinary people with a laptop and a few minutes. A scammer can clone a relative's voice from a short clip and call you in distress asking for money. A bad actor can produce a fake image of a flood, a riot, or a candidate doing something they never did. A fraudster can spin up fifty fake "customer reviews" before lunch.

The two real-world harms are scams and misinformation. Scams take your money. Misinformation takes something harder to get back: a shared sense of what actually happened. Both run on the same fuel — your willingness to believe and forward something before you've checked it.

I don't say this to frighten you. Most of what you see online is still ordinary and real. But the cheap, convincing fake is now part of the landscape, and pretending otherwise leaves you exposed.

The honest clues for each medium

Let me give you the tells people talk about. I'll be honest about each one, because some of this advice is already going stale.

Text. AI-generated writing tends to be generic, over-smooth, and oddly tidy. It rarely contradicts itself, rarely rambles, and rarely says "I'm not sure." It can also be confidently wrong — stating a false fact in the same calm tone it uses for true ones. Watch for writing that says a lot of words without a single specific detail: no names, no dates, no places, no lived experience. Real people are messier and more specific than that.

Images. The famous tells are hands (extra fingers, fused fingers), garbled text on signs and labels, too-perfect or too-many teeth, jewelry that melts into skin, and backgrounds that don't quite make sense when you look closely. Reflections that don't match. Ears and earrings that disagree with each other.

Video and voice. In video, look for lip movements that don't quite sync with the words, unnatural or absent blinking, a face that seems subtly "pasted" onto the head, stiff necks, and edges that shimmer when the person moves. In audio, listen for a flat emotional tone, strange pacing, breaths in the wrong places, a slight metallic or underwater quality, and background noise that cuts in and out unnaturally. Cloned voices often nail the timbre but miss the rhythm of how someone actually speaks.

All of these are real clues. None of them are reliable on their own, and all of them are getting harder to use as the technology improves. Which brings me to the part that actually matters.

The durable strategy: check the source, not the artifact

Here is the shift I want you to make. Stop trying to win a staring contest with the pixels. You will lose that contest more often every year. Instead, ask a different question: where did this come from, and can I trace it back to something trustworthy?

The artifact — the image, the clip, the message — can be faked. The source and the chain of distribution are much harder to fake convincingly. This is the same instinct that protects you from phishing: you don't judge the email by how official it looks, you judge it by where it really came from.

A few practical moves:

Reverse image search. Right-click an image and search for it. If a "breaking news" photo from today actually appeared in an article three years ago, you have your answer. This single habit debunks a huge share of misinformation.

Find the original publisher. Did a real news outlet report this? Does the official account of the person or organization say it? A shocking claim that only exists in forwarded screenshots and no credible outlet is a screenshot, not a fact.

Cross-reference. If something important truly happened, more than one independent source will cover it. One dramatic clip with no corroboration anywhere is a red flag, not a scoop.

Be suspicious of urgency and emotion. This is the most important one. Fakes are engineered to make you feel something fast — outrage, fear, greed, tribal loyalty — so you forward before you think. When a piece of content makes your pulse jump and demands you act right now, that feeling is the warning. Slow down precisely when you least want to.

It's an arms race, and that's the point

I want to be straight with you: deepfake detection is an arms race, and detection is usually a step behind generation. Every clever "tell" gets trained away. Every detector tool gets fooled by the next model. If your whole strategy is spotting glitches, you are signing up to lose slowly.

That's not a reason for despair. It's a reason to anchor on the things that don't expire. Source-checking, cross-referencing, and a healthy resistance to urgency are not in the arms race. They worked before AI, they work now, and they'll work when today's models look quaint. Skepticism plus a verified source beats pixel-spotting every time.

This matters especially around elections and scams here in Africa, where messages travel fast through WhatsApp and Telegram, trust in shared chats is high, and a single convincing fake can spread to thousands before any correction catches up. The fix isn't paranoia about everything. It's a small pause and a habit of checking before you become the person who passed it on.

How-to

How to verify if something is real before you share it

A quick, repeatable routine to check a suspicious image, video, voice note, or message before you forward it to anyone.

Estimated time: PT5M

1. 01

   Pause and notice the feeling

   If the content makes you instantly angry, scared, or excited, treat that as a warning sign. Strong emotion is the engine of misinformation. Take a breath before doing anything else.

2. 02

   Find the original source

   Ask where this came from. Look for the original publisher, the official account, or a credible news outlet. If it only exists as a forwarded screenshot or an unattributed clip, be skeptical.

3. 03

   Cross-reference with other sources

   Search the claim. If something important really happened, more than one independent and trustworthy source will report it. Silence everywhere else is a red flag.

4. 04

   Reverse image search any photo

   Right-click images and run a reverse search to see where else they appear. An old photo reused as breaking news is one of the most common fakes.

5. 05

   Verify people directly through a known channel

   If a message or voice note claims to be from someone you know asking for money or details, contact them on a number you already trust. Never act on the claim alone.

6. 06

   When in doubt, don't share

   If you can't confirm it, hold it. Not sharing an unverified thing costs you nothing. Spreading a fake costs other people something.

A note on staying calm

It's tempting to react to all of this by trusting nothing. I'd ask you not to. A world where you believe everything is dangerous, but a world where you believe nothing is also a victory for the people producing fakes — because then real evidence stops mattering too.

The healthy middle is simple to say and takes practice to live: extend normal trust to ordinary things, raise your guard when something is surprising, emotional, or asking you for money or a vote, and always be willing to check before you forward. That's not cynicism. That's care.

Frequently asked questions

Are AI detection tools reliable?

Not reliably enough to bet on alone. Detection tools can help, but they are locked in an arms race with generation tools and are often a step behind. They produce false positives and false negatives. Use them as one input, never as your final answer. Source-checking is more dependable.

Is it still useful to look for weird hands or glitches in images?

A little. If you spot extra fingers, garbled text, or melting jewelry, that's a useful clue. But modern image models have largely fixed these, so the absence of glitches tells you nothing. Treat visual tells as a bonus, not your main defense.

How can I protect myself from voice-cloning scams?

Be deeply suspicious of any urgent call or voice note asking for money or sensitive details, even if the voice sounds exactly like someone you know. Hang up and call the person back on a number you already trust. Consider agreeing on a private code word with close family for emergencies.

Why is this especially important during elections?

Fakes spread fastest in trusted group chats, and a convincing fake video or quote can reach thousands before any correction does. The goal of election misinformation is often just to confuse and inflame. Checking the source before you forward is one of the most useful civic habits you can build.

What's the single most important habit?

Pause when content makes you feel a strong emotion, and check the source before you share. Almost all misinformation depends on you forwarding it quickly and without thinking. Slowing down breaks the chain.

Further reading on this site

• How to Protect Yourself from Phishing — the same source-checking instinct, applied to scams that target you directly.
• AI Bias and Ethics, Explained Simply — how the systems behind generated content can carry hidden assumptions.
• Browse Technology — more calm, practical writing on the tools shaping our lives.

If this was useful, subscribe to the newsletter for more honest, practical guides delivered when they're ready.