How to Spot Deepfakes & AI Content

Master the art of detecting artificial content. Learn the telltale signs that reveal deepfakes, AI-generated images, and ChatGPT-written text.

🖼️ Image Deepfakes 🎥 Video Deepfakes 📝 AI Text 📄 Fake Documents
🖼️

Spotting Image Deepfakes

AI-generated images have subtle flaws that give them away

👁️

Eye & Face Issues

The eyes are windows to AI's soul - they're hardest to fake perfectly.

  • Asymmetrical pupils or iris patterns
  • Mismatched eye reflections
  • Unnatural eye color or glow
  • Weird ear shapes (often different on each side)
  • Teeth that merge together
💇

Hair & Edges

Hair is incredibly complex - AI often struggles with natural flow.

  • Hair that fades into background
  • Unnaturally smooth hairlines
  • Missing or floating hair strands
  • Sudden texture changes
  • Blurry edges where hair meets face

Lighting & Shadows

Physics doesn't lie - inconsistent lighting is a dead giveaway.

  • Shadows going different directions
  • Missing shadows under chin/nose
  • Face lit differently than background
  • Reflections that don't match lighting
  • Overly smooth or plastic-like skin
👔

Accessories & Details

Small details are where AI often fails spectacularly.

  • Earrings that don't match
  • Melting or warped glasses frames
  • Text on clothing is gibberish
  • Jewelry that defies physics
  • Buttons or zippers that lead nowhere
🖐️

Hands & Body

Hands are AI's biggest weakness - always check them first!

  • Wrong number of fingers (3, 6, 7...)
  • Fingers merging together
  • Impossible hand positions
  • Hands too large or small for body
  • Missing or extra limbs
🏠

Background Weirdness

The background often contains impossible geometry or patterns.

  • Repeating patterns that shouldn't repeat
  • Windows/doors that lead nowhere
  • Text in background is nonsense
  • Objects merging unnaturally
  • Perspective that doesn't make sense
💡 Pro Tip: The Zoom Test
Always zoom in on images! Deepfakes often look convincing at normal size but fall apart when you look closely. Check edges, transitions, and fine details. If something looks "painted" or "smudged" when zoomed, it's likely AI-generated.
🎥

Detecting Video Deepfakes

Motion adds complexity - but also more ways to spot fakes

👄

Lip Sync Problems

The mouth movements often don't perfectly match the audio.

  • Lips slightly behind or ahead of audio
  • Unnatural mouth shapes for certain sounds
  • Teeth appearing/disappearing oddly
  • Tongue movements that don't match speech
😊

Expression Glitches

Facial expressions in deepfakes often feel "off" or robotic.

  • Delayed emotional reactions
  • Expressions that don't reach the eyes
  • Sudden expression changes
  • Face "resetting" to neutral oddly
👀

Blinking & Eye Movement

Natural blinking patterns are hard to replicate perfectly.

  • Too much or too little blinking
  • Both eyes not blinking in sync
  • Unnatural eye tracking
  • Dead or unfocused stare
🔄

Head & Neck Issues

The connection between head and body often reveals the fake.

  • Head movements feel disconnected from body
  • Neck looks pasted on or blurry
  • Unnatural head angles
  • Wobbling or flickering at edges
📝

Identifying AI-Generated Text

ChatGPT, Claude, and other AIs have distinctive writing patterns

🔄

Repetitive Patterns

AI loves certain phrases and structures - watch for overuse.

  • "It's important to note that..."
  • "In conclusion..." (multiple times)
  • Starting sentences with "Moreover" or "Additionally"
  • Overuse of transitional phrases
  • Same sentence structures repeating
🎯

Too Perfect

Human writing has natural imperfections - AI is suspiciously flawless.

  • Perfect grammar ALWAYS
  • No typos or informal language
  • Perfectly balanced paragraph lengths
  • Never uses slang or colloquialisms
  • Lacks personal voice or style
📚

Generic Knowledge

AI gives surface-level information without real insight.

  • Vague statements that sound smart but say nothing
  • No personal experiences or anecdotes
  • Avoids specific details or examples
  • Wikipedia-style explanations
  • Hedging language ("generally," "typically," "often")
📋

List Obsession

AI loves making numbered lists and bullet points.

  • Everything becomes a list
  • Unnecessary categorization
  • "Here are X ways to..."
  • Points that don't quite fit together
  • Forced structure where none needed

🔍 Quick Comparison: Human vs AI Writing

Human Writing
"So I tried this new coffee place downtown - honestly wasn't great. The latte was okay I guess, but way overpriced. Might go back tho, the vibe was pretty cool."
AI Writing
"The new coffee establishment downtown offers a unique experience. While the latte was adequately prepared, the pricing may be considered premium. The ambiance, however, was notably pleasant."

Notice: Human writing has personality, opinions, informal language. AI writing is formal, balanced, and lacks genuine emotion.

📄

Spotting Fake Documents

PDFs, contracts, and official documents can be manipulated

🔤

Font Inconsistencies

Edited documents often have mismatched fonts or sizes.

  • Different fonts in same paragraph
  • Slightly different text sizes
  • Misaligned text baselines
  • Inconsistent character spacing
📅

Metadata Red Flags

Document properties often reveal tampering.

  • Creation date after supposed signing date
  • Multiple editing programs listed
  • Author name doesn't match signer
  • Suspicious modification history
✍️

Signature Problems

Digital signatures are often poorly faked.

  • Signature looks "too perfect"
  • Different resolution than document
  • White box around signature
  • Signature can be selected/moved
96%
of people can't spot deepfakes
10x
increase in deepfakes since 2019
3 sec
average time people look before sharing
70%
of students use AI for homework

Ready to Test Your Skills?

Use ObscuroScan to instantly verify any image, video, or document. Our AI catches what human eyes miss.

Try ObscuroScan Free
🎯 Remember: Trust but Verify
In today's world, seeing isn't believing. Always verify important content, especially if it seems shocking, perfect, or too good/bad to be true. When in doubt, use multiple detection methods and trusted sources.