Frequently asked questions

On clear-cut cases (unedited AI output, authentic human writing of 300+ words) our multi-engine blend performs strongly — but like every detector, accuracy drops on short, edited or paraphrased text. We publish limitations openly on our methodology page and attach a confidence level to every score instead of quoting a single marketing number.

Because accuracy claims without a fixed false-positive rate are meaningless. Tools claiming 98-99% have flagged the US Constitution as AI-written. Independent benchmarks like RAID (ACL 2024) show real-world performance is messier than marketing pages admit — ours included.

Yes, in both directions. False positives (human work flagged as AI) hit non-native English writers hardest — a Stanford study measured a 61% average false-positive rate on TOEFL essays. False negatives happen with edited or paraphrased AI text. That is why scores are evidence, never proof.

Confidence reflects how much weight the score can bear: it rises with text length, agreement between engines and language fit. Treat low-confidence results as a suggestion to gather more text or context, not a finding.

Text (paste up to 30,000 words), documents (PDF, DOCX, TXT, MD up to 20 MB), images (JPG, PNG, WebP, GIF, HEIC up to 15 MB) and videos (MP4, MOV, WebM up to 100 MB / ~3 minutes).

Text and images complete in a few seconds. Videos take 15–60 seconds because frames are extracted and scored individually — results wait in your history if you leave the page.

Overall score with verdict and confidence, per-engine breakdown, evidence signals, sentence-level analysis (text) or frame timeline (video), provenance findings, and our methodology summary — suitable for documentation and disputes.

Often yes — generator attribution reports the closest match among Midjourney, DALL·E, Stable Diffusion, Flux, Firefly and others, with its own confidence score.

The homepage scanner is free without an account (5,000 characters, 5 scans/day). A free account adds 300 credits monthly, the full multi-engine analysis, image/video scanning and PDF reports — forever free.

1 credit = 100 words of text or document checks. Images cost 5 credits, videos 25. Plan credits refresh monthly; one-time pack credits never expire.

One click in the customer portal (run by Polar, our merchant of record). You keep access until the end of the paid period. No retention tricks.

Anonymous free scans: no — only the score is logged. Account scans are stored privately so you can access history and reports; delete them anytime. We do not sell scan content or use it to train third-party models.

No. Scans are private to your account. We do not share results with institutions.