How to Spot AI-Generated Videos, Fake Ads, and Deepfakes: A Visual Guide

Quick Reference: AI Detection Checklist

What to Check	AI Tell	Real Content
Video length	Under 15 seconds	Varies naturally
Subject framing	Perfectly centered every frame	Camera work varies, subjects move in/out of frame
Continuity between cuts	Backgrounds change, props shift, patterns alter	Consistent world across all angles
Text in image/video	Misspelled, garbled, or gibberish	Readable and correct
Contrast	Extreme — deep shadows offset by bright highlights	Can be flat, bright, dark, or anything
Body language	Gestures don't sync with speech	Hands, head, and breathing align with words
Smoke/fire/water	Appears from nothing, disappears into nothing	Follows physics — rises, spreads, lingers
Background details	Walls, trim, and signs fall apart on inspection	Architecture and objects hold up to scrutiny

---

Why This Matters

AI-generated content is being used right now to sell fake products, promote scam websites, forge celebrity endorsements, and flood platforms with misleading ads. The people creating this content are betting that you won't look closely enough to notice.

The good news: AI video and imagery still have consistent, detectable flaws. Visual effects professionals — people who build realistic-looking fakes for a living — are particularly good at spotting them. Much of the practical advice in this guide is informed by the analysis work of Corridor Crew, a team of VFX artists who have produced several detailed breakdowns of AI scams and deepfakes.

Here's what to look for.

---

1. Check the Video Length

Current AI video generators produce clips of about 5 to 15 seconds. If a suspicious video is shorter than 15 seconds and appears to be a single unbroken shot, that's a strong initial signal.

Longer clips can be stitched together from multiple AI-generated segments, but the cuts between them often feel abrupt and disconnected — more like a stock footage montage than a directed sequence.

The test: Is this a single short clip, or does it tell a story with connected actions? AI clips tend to show movement without progress — things sway, shimmer, and shift, but nothing actually happens.

---

2. Look at the Framing

AI video generators place the main subject dead center in the frame, perfectly exposed, every single time. Real camera work involves composition choices — a cinematographer might frame a subject off-center, use leading lines, or let the subject move in and out of frame.

If every shot in a video looks like the subject was placed precisely in the middle with flawless exposure, be suspicious.

---

3. Test Continuity Between Cuts

This is one of the strongest tests available right now. AI models don't have a persistent understanding of the world they're generating. Each shot is essentially a new generation.

What to check across cuts:

• Do background objects stay in the same position?
• Does the person's clothing, jewelry, or accessories stay consistent?
• Do patterns on walls, floors, or furniture match from one angle to another?
• Is there a cat door in one shot that vanishes in the next? An exit sign that changes spelling?

If the world is consistent across multiple camera angles, the content is almost certainly real. AI cannot maintain this kind of environmental consistency.

---

4. Watch for Physics Violations

AI doesn't understand physics. It approximates the appearance of physical behavior without actually simulating it.

Common giveaways:

• Wind: Something should be windy (an umbrella flying, flames moving) but trees and leaves are perfectly still
• Water: Splashes that don't leave surfaces wet, or volumes of water that appear from nowhere
• Fire and smoke: Flames that flicker at the wrong speed for their apparent size, smoke that appears from nothing and vanishes into nothing rather than rising and dissipating naturally
• Object interaction: Things pass through each other instead of colliding — tails through bodies, hands through objects

A campfire moves differently from a building fire because of their different scales, but AI doesn't know how big its fire is supposed to be. If a massive blaze moves like a candle flame, it's generated.

---

5. Examine the Contrast

Here's a technical tell that VFX professionals use: AI-generated images almost always have extreme contrast — very bright highlights paired with very deep shadows in roughly equal measure.

This happens because of how these images are generated. AI image models start from noise — an even distribution of random values. Since noise averages to a middle gray, AI images tend to inherit this property. For every bright area, there's a dark area to balance it out.

What real photos can do that AI can't (yet):

• A flat, overcast scene with no deep shadows
• A mostly dark image with just a sliver of light
• A sun-washed photo where everything is bright with minimal shadow

If you're looking at a product photo or an ad and everything has that hyper-contrasty, tone-mapped look, it's worth a closer inspection.

---

6. Spot Deepfake Faces

Deepfakes — where a real person's face or voice is manipulated to say things they never said — are being used for celebrity endorsement scams, crypto schemes, and fake product promotions.

Key deepfake tells:

• Mouth interior: The black levels inside the mouth are often darker than physically possible. If the darkest spot in the frame is inside someone's mouth (darker than their black clothing, darker than shadows), the mouth region is likely synthesized
• Edge artifacts: When a deepfaked face turns, the edges where the manipulation meets the original footage can blur, warp, or glitch — especially around the jawline and hairline
• Body language mismatch: Real people gesture in sync with their speech. Their hands emphasize syllables, their head tilts with inflection, their chest moves when they breathe. Deepfakes typically only manipulate the face, so the body's natural rhythm doesn't match the synthesized words
• The Adam's apple: This is a tell that deepfake tools consistently miss. When a real person speaks, their throat moves with their words. Deepfakes almost never animate the Adam's apple correctly — look for a static throat while the mouth is moving
• Breathing: If someone is speaking continuously without any visible breathing — no chest expansion, no pauses for breath — the audio is likely synthesized and laid over existing footage

---

7. Spot AI-Generated Product Photos

Fake online stores are using AI-generated product images to sell items that either don't exist or look nothing like what arrives. Some real manufacturers are also using AI to generate model photos instead of photographing their actual products — which means what you're buying is based on an AI's interpretation of the product, not the product itself.

What to look for:

• Inconsistent details across photos: A shirt with 5 vertical seams in one photo and 3 in the next. A necklace that changes shape between shots. A pattern that shifts or disappears
• Background architecture: Walls, trim, and columns tend to warp or garble in AI product photos. Look at straight lines in the background — they'll often curve, merge, or dissolve
• The person wearing the product: Check for extra fingers, impossible anatomy, or faces that appear on both the front and back of a head
• That AI contrast: Deep shadows and bright highlights in a studio setting where the lighting should be flat and even

---

8. Recognize AI-Generated Books and Written Content

AI-generated books have flooded Amazon and other platforms. They use AI art for covers and AI text for content, then promote themselves through paid sponsored placements and fake reviews.

Red flags for AI books:

• Sponsored placement: On Amazon, AI books rely heavily on paid promotion because they can't generate organic word of mouth. If a book you've never heard of appears as "Sponsored," investigate further
• Cover art style: Hyper-centered composition, a style that's not quite painted and not quite CG, with an unnaturally polished look
• Text quality: Excessive use of em dashes in unnatural places, emoji-heavy descriptions, generic motivational language that reads like it was generated by a chatbot
• Review patterns: Identical phrasing across multiple reviews, all posted within a short time window. Check the one-star reviews — real humans often call out AI slop directly
• Formatting: Open the preview. If the interior has zero effort in layout, typography, or design, and reads like a raw copy-paste from a text generator, that's your answer

---

9. Check the Account

Before trusting any content, check the source:

• New account with polished content? Real creators build a history over time. An account with 150,000 followers but only AI-looking content across the board is likely manufactured
• Celebrity photos that seem too good to be true? AI influencers pose with celebrities in photos where backgrounds change between cuts, details shift, and the "influencer" appears to exist only in these generated scenes
• Upload date: If a video was posted before 2023, it's almost certainly real. AI video generation didn't exist in any usable form before then

---

10. When in Doubt, Follow the Money

The most important question isn't always "is this AI?" — it's "what does this content want from me?"

If a video, ad, or social media post is asking you to:

• Send cryptocurrency to an address
• Buy a product from an unfamiliar website
• Click a link to claim a prize or discount
• Enter personal information

...then whether it's AI-generated is almost secondary. The intent is what matters. AI has simply made it cheaper and faster to produce convincing-looking scam content at scale.

---

Protect Yourself

AI detection is a moving target. The tools get better, and so do the fakes. The best defense is a combination of visual literacy (knowing what to look for) and healthy skepticism (questioning the intent behind content that asks for your money or attention).

If you encounter a suspicious website being promoted through AI-generated ads, check it on RiskScope. We analyze domains against 14+ threat intelligence sources and give you an instant risk score — no signup required.

Check a website now

---

This guide was informed by the AI scam analysis work of Corridor Crew, a Los Angeles-based visual effects team. Their breakdowns of AI-generated content are among the most detailed and accessible available — we recommend their channel for anyone wanting to develop a sharper eye for digital fakes.