Detecting AI-generated content in 2026 is a survival skill. Whether you are a journalist verifying a source, a brand checking user-generated content, an HR manager reviewing candidates, or an ordinary user trying to spot a scam - the ability to spot AI content matters. This is the complete practical guide, covering free + paid tools, manual detection tells, Content Credentials verification + a step-by-step workflow.

1. Why Detection Matters More Than Ever

  • Deepfake fraud + extortion rising sharply
  • Political disinformation during elections
  • Fake celebrity endorsement scams
  • Job interview deepfakes (candidate impersonation)
  • Insurance + KYC fraud
  • Academic + journalistic integrity
  • Personal safety in dating + relationships

2. Detection vs Provenance

  • Detection - analyze content after the fact
  • Provenance - check embedded Content Credentials that prove origin
  • Use BOTH - provenance where available, detection as fallback

3. Detecting AI-Generated Images

Free Tools

  • Illuminarty - free web tool, decent accuracy
  • AI or Not - simple upload check
  • Hive Moderation (free trial tier)
  • Content Credentials Verify - contentcredentials.org/verify

Manual Tell-Tales

  • Hands - wrong number of fingers, distorted, unnatural pose
  • Ears - asymmetric, distorted, missing details
  • Teeth - blurred, inconsistent, wrong count
  • Eyes - reflections don't match, pupils asymmetric, iris texture off
  • Glasses - frames warp between shots, lenses distort inconsistently
  • Text in images - garbled or nonsense letters (older models; newer are better)
  • Backgrounds - objects merge nonsensically, geometry wrong
  • Shadows - inconsistent with light source
  • Reflective surfaces - don't reflect what should be there
  • Hair - unnaturally smooth or strands don't connect to scalp naturally
  • Symmetry - unnatural perfection or unnatural asymmetry

Metadata Check

  • Check EXIF via ExifTool or online EXIF viewers
  • Missing camera info + no editing history = suspicious
  • Check for C2PA Content Credentials chunks
  • Reverse image search (Google Images, TinEye, Yandex)

4. Detecting AI-Generated Video

Free Tools

  • Deepware Scanner - free, web + mobile
  • InVID + WeVerify Toolkit - journalists' free toolkit
  • Reality Defender
  • Sensity AI
  • Intel FakeCatcher (blood-flow detection)
  • Hive Moderation

Manual Tell-Tales

  • Temporal inconsistency - features change subtly frame-to-frame
  • Blink patterns - too rare or too frequent
  • Lip-sync mismatches - subtle misalignment on plosives (p, b, m)
  • Head-body edge - unnatural boundary or warping
  • Neck + shoulder area - motion doesn't match head
  • Skin texture - waxy, over-smooth, or shifts between shots
  • Lighting inconsistencies - face vs environment
  • Compression artifacts - deepfakes often have distinctive artifacts around face region

Workflow

  1. Reverse video search - Google Images, InVID key frame search
  2. Check original source + upload date
  3. Run through 2+ detectors
  4. Manually scrutinize face + hands + edges
  5. Check audio separately

5. Detecting AI-Generated Audio

Tools

  • Sensity + Reality Defender have audio modules
  • Pindrop for phone-call analysis (financial + KYC)
  • Deepgram + AI voice detection APIs

Manual Tell-Tales

  • Emotional flatness - AI voices lack subtle emotional inflection
  • Consistent cadence - too regular, no natural pauses
  • Breathing patterns - AI often omits or over-regularizes breath sounds
  • Ambient consistency - background noise too clean or inconsistent
  • Consonant articulation - subtle mismatches on complex consonants
  • Pitch stability - unnaturally stable pitch
  • Room acoustics - reverb / echo doesn't match implied environment

6. Detecting AI-Generated Text

Tools

  • GPTZero, Winston AI, Copyleaks AI Detector, Turnitin AI
  • Originality.AI for content marketing
  • Note: text detectors have higher false-positive rates than image/video

Manual Tell-Tales

  • Generic transitional phrases ("It is important to note that...")
  • Balanced pro/con structure on every topic
  • Lack of first-person specifics + anecdotes
  • Perfect grammar + no colloquialisms
  • Repetitive sentence structure
  • Hedge words + qualifiers throughout
  • No factual errors (paradoxically - real humans make small mistakes)

7. Content Credentials (C2PA) Verification Workflow

  1. Save the file locally (or use URL)
  2. Go to contentcredentials.org/verify
  3. Upload or paste URL
  4. Read the credential chain - who created, what tool, what edits
  5. Verify signatures + issuer trust
  6. Detect tampering (missing signatures, broken chain)

Also see our complete C2PA guide.

  • Google Images - drag + drop or URL
  • Yandex Images - often best for people + faces
  • TinEye - specialized reverse search
  • Google Lens - mobile-friendly
  • InVID Key Frame - extract video keyframes + reverse-search each

9. Detection Workflows by Use Case

Journalists

  1. InVID + WeVerify Toolkit for videos
  2. Reverse image search
  3. Content Credentials verify
  4. Sensity or Reality Defender if available
  5. Cross-reference source + upload date
  6. Manual visual scrutiny

Brand Safety Teams

  • Reality Defender or Hive Moderation for scale
  • Content Credentials reader integration
  • Escalation workflow for suspected AI misuse

HR + Recruiting

  • Live interview verification via Pindrop-style audio + real-time video detection
  • Cross-check LinkedIn photos with reverse search
  • Require live camera on for final interviews

Financial + KYC

  • Truepic + Sensity for identity document verification
  • Liveness detection at capture
  • Continuous authentication for high-risk sessions

Everyday User

  1. Deepware Scanner for suspicious videos
  2. Reverse image search for suspicious photos
  3. Manual tell-tale check
  4. Ask direct questions (voices + faces struggle with real-time complex Q&A)

10. Limitations You Must Understand

  • No detector is 100% accurate
  • New models regularly evade older detectors
  • Compression + re-encoding strip detection signals
  • False positives happen - real content flagged as AI
  • Text detection is particularly unreliable
  • Detection is one signal, not proof

11. Where Detection Is Heading

  • Provenance-first architectures (Content Credentials primary)
  • Real-time detection in video calls
  • Platform + government + academic shared benchmarks
  • Regulatory certification of detection tools
  • Multi-modal ensemble detection standard

Disclaimer: Snapshot as of September 23, 2026. Tools + policies + laws change rapidly. Verify with official sources before making career, legal, or commercial decisions.