How OmniDetect Works
We don't guess — we verify with multiple sources.
Three Steps to Certainty
You paste your text
Paste up to 50,000 characters. Your text goes to the engine that reads it, and a copy stays in your history until you delete it.
Multiple engines analyze independently
Your text goes to the engine that measured most accurate for its language on our own frozen corpus — Winston for English, Pangram for Japanese. One call returns both the verdict and the specific passages behind it.
You get a consensus report
Results are aggregated into an OmniScore (0-100). When engines agree, confidence is high. When they disagree, you know to investigate further. Our AI Writing Coach explains flagged patterns and suggests improvements.
Our Detection Engines
Sapling AI
Pioneer in AI detection. Uses perplexity and burstiness analysis to identify AI-generated text. Widely trusted by educators.
Winston AI
Advanced AI content detector with multi-language support. Uses proprietary models for accurate detection across diverse writing styles.
ZeroGPT
Advanced AI and plagiarism detection built for content marketers and publishers. Combines AI detection with content verification.
Why we route instead of averaging
No single detector is both accurate and authoritative, so we hold several and test them ourselves. Averaging them, it turns out, is not the answer — a weak engine drags a strong one down.
- Measured on our frozen corpus, one engine misjudged 66.7% of English academic writing as AI — averaging it with a good engine makes the good one worse
- Engines differ by language, not just by quality — one of them cannot read Japanese at all, and says so with an error rather than a wrong answer
- So the useful question is not how many engines answered, but which one was measured best for the language in front of it — and what its error rate on that kind of writing actually is
- When engines disagree — that's valuable information too
OmniDetect acts as your auditor: we don't produce a diagnosis, we verify it across multiple independent sources.
Privacy by Design
Delete it whenever you want
Your original text is never saved to our database. It's sent directly to detection engines and discarded.
SHA-256 hashing
Only a cryptographic hash of your text is stored — for caching purposes. The hash cannot be reversed to recover your text.
No training data
Your text is never used to train AI models. Detection engines receive your text for analysis only.
Honest Limitations
We believe in transparency. Here's what you should know:
- AI detection technology is probabilistic, not definitive. Results represent likelihood, not certainty.
- Detection accuracy varies by text length. Longer texts (500+ words) produce more reliable results.
- Heavily paraphrased AI text may evade detection. No detector catches everything.
- ESL (English as a Second Language) writers may receive higher AI scores due to simpler sentence structures.
- Results should complement — not replace — human judgment.