Each question has two audio clips. One is from a real speech. One is AI-generated using voice cloning software. For each question, select which clip you think is real — then see how well you can detect synthetic speech.
Your time will be recorded — the highest score in the shortest time wins a NewsGuard cap.
Please enter your name and a valid email to continue.
✓ Your result has been recorded. We'll contact you if you win.
How to spot a deepfake voice
1
Cadence
Real speakers stumble, breathe, and pause unpredictably. AI voices tend to flow too smoothly — an even rhythm with no natural hesitation.
2
Background noise
Authentic recordings carry ambient sound — a crowd, a room, a mic. Deepfakes are often eerily clean, as if recorded in a studio that doesn't exist.
3
Emotion
AI can mimic emotion, but often misses the intensity or places it in the wrong moment.
4
Consonants and breath
Listen closely to hard consonants (p, t, k) and the breaths between sentences. Deepfakes often get these slightly wrong — too sharp, too soft, or missing entirely.
How we made the deepfakes
NewsGuard selected five authentic audio clips from prominent political figures and used the free voice cloning software Parrot AI to generate convincing deepfakes — in under 30 seconds each.
Share this quiz
Please forward this fake audio competition to friends or family members to show them how hard it is to tell authentic audio from fakes — and why they might want to take the steps we suggest to protect themselves.
