Frequently asked questions

It rewrites a draft, often one produced with an AI assistant, so it reads like careful human writing, then checks that the rewrite still says the same thing. Every rewrite comes with a human-like score, a style score and a meaning check.

We do not promise that, and nobody honestly can. Detectors disagree with each other and change often. What we do is remove the habits readers notice and keep your meaning intact. Use it for writing you are allowed to get help with.

That depends on your institution and your teacher. Many allow AI help for some tasks and not others, and some ask you to disclose it. Check the rules for your assignment first. Our terms ask you to follow them.

A 0 to 100 measure of writing habits, from our own rules: stock phrases, flat sentence rhythm, dashes, filler connectors and similar. Higher means fewer habits. It is not a prediction of what any detector will say.

Numbers and units, names, negations, qualifiers like "may" or "associated with", and large changes in length. It errs on the side of flagging, so read the flags before you copy.

The strength of the rewrite. The engine rewrites your text in passes, each starting from the last: Medium is 3 passes and stays closest to your draft, Extra is 5, Max is 10 and rebuilds the text (the setting of the research paper, and the one that drifts the most). Versions are independent runs at the same strength, up to three, ranked by the meaning check so the most faithful comes first. Free includes Medium, Basic adds Extra, Pro and Ultra add Max.

English for now. The engine is tuned on English prose, so drafts in other languages may come back in English.

With history on, your drafts and rewrites stay in your account for up to 90 days so you can come back to them. With history off, neither the draft nor the versions are stored at all. We never train models on it.

An open model on our own GPU server, never a third-party AI service: Qwen3 (8B, base model) with the HIP adapter released by its authors at Carnegie Mellon (Humanization by Iterative Paraphrasing, 2026), served with vLLM. Your text is not sent anywhere else.

Each word of the draft you paste, once. Rewrites that fail on our side are not counted.

An optional second pass. After the rewrite, an instruction-tuned model compares each passage with your draft and proposes the smallest edits that restore lost or changed figures, units, names, negations, qualifiers and claims; our rules apply only exact, short, unambiguous fragments and the meaning check runs again. Free accounts get one run to see it, Basic pays it at 1.5x words, Pro and Ultra have it included, and every credit run includes it. It cannot verify facts your draft got wrong.

Yes. Every account has a personal link (Affiliate, in the top bar). Anyone who signs up through it within 90 days gets 5 free welcome credits (short texts, up to Extra strength, one version, one Fidelity run) once their email is confirmed, and becomes your referral: you earn 20% of everything they pay for 12 months: plans, renewals and credits. Commissions are held 14 days for refunds, then you can withdraw from 15 USD by bank transfer. Or trade 5 paying referrals for a month of Pro.

Yes. Credits cost 1 USD a run, with 10% off for 10, 20% off for 50 and 30% off for 100. One credit runs one rewrite of up to 3,000 words at any strength with up to 3 versions, on any plan. A pack is valid 30 days from purchase. Your plan is always used first; a credit is only spent when the plan does not cover the run.

Yes, from your account page. You keep your plan until the end of the period you paid for.

Yes, on Pro and Ultra. One endpoint, the same engine and checks, billed against your monthly words. See the API page for an example.