How to Spot an AI-Generated Video: A 2026 Field Guide

In this guide

1. The habit that beats every trick: check the source
2. Visual tells that still work (sometimes)
3. Audio tells
4. Tools that help
5. Protecting your family
6. The bottom line

A year ago, spotting AI video was easy: melting hands, flickering backgrounds, silent footage. In 2026, the top models on the video leaderboard produce photorealistic clips with synchronised sound, in 4K.

You can't rely on "it looks weird" anymore. But you're not helpless. Here's a practical field guide.

The habit that beats every trick: check the source

Before you look at a single pixel, ask:

• Who posted this first? Is it a verified news outlet, or an account created last month?
• Is anyone else reporting it? Real major events are filmed by lots of people from lots of angles.
• Does it make you feel a strong emotion, and want to share right now? That's exactly what fakes are designed to do. Pause first.

This one habit catches more fakes than any visual check.

Visual tells that still work (sometimes)

Modern models have fixed the obvious flaws, but these are still worth checking:

1. Text in the scene. Signs, labels and screens can have garbled or shifting letters.
2. Physics over time. Watch objects that should stay put. Do they subtly drift, merge or change shape?
3. Continuity between shots. Clothing, jewellery and background details can change from one shot to the next.
4. Hands and teeth under motion. They're much better than before, but fast movement can still produce odd shapes.
5. Too-perfect lighting. Everything looks like a film set: glowing skin, cinematic haze, even on a "phone video".
6. Lip-sync on unusual sounds. Watch closely during laughter, whispering or shouting.

Audio tells

• Room tone that doesn't match. Voices that sound studio-clean in a noisy street.
• Breathing and pauses. Real speech has messy breaths, "ums" and overlapping talk.
• Emotion that's slightly off. The words and the tone don't quite fit.

Tools that help

• Content credentials (C2PA). More cameras, apps and AI tools now attach tamper-evident "content credentials" that record where media came from. Look for a small "CR" or info icon on platforms that support it.
• Detection services. Companies focused on media authentication and forensics are growing fast as demand rises. They're useful, but no detector is 100% accurate, so treat their results as one clue among several.
• Reverse image search. Screenshot a frame and search it. You'll often find the original clip, or the AI tool's showcase it came from.

Protecting your family

• Agree on a family "safe word" for phone or video calls asking for money urgently. Voice cloning scams are real.
• Tell older relatives: no legitimate bank, government office or grandchild will ask for gift cards or crypto over a video call.
• Report suspected deepfake scams to the platform, and to your local fraud reporting service.

The bottom line

Pixels can lie, but context rarely does. Check the source, slow it down and don't rush to share. That combination beats nearly every fake out there.

Free · every Friday

Get the standings in your inbox

Who moved up, who fell, and what's worth trying this week, in a five-minute read. No spam.

Email address Subscribe