Frequently Asked Questions

First, save your draft history immediately — Google Docs version history, Word track changes, screenshots with timestamps. Then ask your professor in writing which detector was used and at what threshold. Cite the published false-positive rates: GPTZero around 0.24% (their own figure), Originality.ai 12 to 18 percent on independent benchmarks. Calmly request a one-on-one review of your drafting trail. Most schools have a formal appeal path and the evidence almost always favors the student who kept receipts.

No catch. The home tool runs entirely in your browser — no server bills, no detector-API charges, no signup, no email collection. The Guide page carries one display ad, which is the only revenue stream and never on the actual detector. We deliberately do not offer a paid "humanizer" tier because we believe automated AI-text obfuscation crosses an ethical line we are not willing to cross. The cost to keep this online is roughly USD ten a year in domain renewal.

Treat it as a directional estimate, not a substitute. The real detectors run proprietary language models we do not have access to. Our score correlates with theirs because the underlying signals (burstiness, perplexity proxy, repetition) are well-documented industry-standard inputs, but the exact numerical output will differ. Use our score to decide whether your text deserves a paid detector check, not as a final verdict you can rely on for high-stakes submission.

No. Every calculation — bigram entropy, word-rarity scoring, sentence segmentation, detector simulation — happens inside your browser tab using static JavaScript. There is no backend API receiving your text. We do not log, store, or transmit your input to any server. You can verify by opening the browser developer tools network tab — you will see zero outbound requests once the page loads. Production prompts containing private essays or PII are safe to paste.

All four commercial detectors were trained primarily on English. Independent measurements report Korean false-positive rates anywhere from 8% to 30%, far more volatile than English. Our tool intentionally dampens Korean scores by 15% because the underlying signals (bigram entropy, word rarity) are less reliable when applied to a language the detector models barely saw during training. Read Korean scores as an early-warning indicator, not a verdict — and never rely on them for high-stakes Korean academic submissions.

Strongly no. First, most school academic-integrity policies treat automated AI-obfuscation as the same offense as AI generation, so you would be entering the exact category you are trying to escape. Second, Originality and Winston now train on humanizer output and recognize the patterns, so humanizing often raises your score rather than lowers it. Third, the responsible response is your draft history and version control as evidence, not a rewrite. We will never add a humanizer to this site.

Use detector scores as a conversation starter, never as a verdict. The published false-positive rates guarantee that a 200-student class produces multiple wrong accusations per semester. When a score flags, ask the student to walk you through their draft in person, show their drafting tool history, and explain a few key decisions. A student who actually wrote the essay can answer those questions in two minutes. A student who did not, cannot. That conversation is the actual evidence — the score is just the trigger to have it.