Comparison

AI content detectors compared

An objective feature comparison of warning-only AI detectors, based on each product's own public documentation. Our measured figures come from a held-out set the model never trained on.

Feature comparison

Verified from public product pages, August 2026.

Featureisthisaigenerated.appGPTZeroOriginality.ai
Image detectionYes — free, no accountNo (text-focused)No (text-focused)
Sampled-video detectionYes — free, no accountNoNo
Text detectionYes — warning-only, 60+ wordsYes — free tierYes — 3 free scans/day, up to 2,000 words
Document detection (PDF/DOCX)Yes — free, per-sectionLimitedLimited
APIYes — public, free tier 100 tokens/weekYes — paidYes — paid
Measured accuracy publishedYes — recall + false-positive rate with sample sizesClaims, no public held-out numbersClaims 97.8% multilingual, no public held-out set
Warning-only framingYes — "signals, not proof"PartialPartial — "probability, not guilt"
Marketplace checkersYes — Vinted, Marktplaats, 2dehands, Tinder, CatawikiNoNo
Free tierUnlimited image/video checks, 100 API tokens/weekLimited free scans3 free scans/day

How we measure

Our image detector reports 78.1% strong warnings on 653 AI images and 4.0% false warnings on 795 real images, from a held-out set. Video: 82.6% caught on 23 videos. Competitor figures are their own public claims, not independently verified here.

No detector output is standalone proof. A low score does not prove human origin; a high score does not prove AI origin.