Nigerian English is a recognised variety of English with its own settled usage, and that is precisely why detectors struggle with it. An AI humanizer in Nigeria therefore gets reached for in two situations: polishing an AI-assisted draft, and trying to clear a flag on work somebody genuinely wrote. The second one is the common case and the rewrite is the wrong tool for it.
There is a free path here that needs no card, which matters in a market where most of this category prices in dollars and asks for payment details before showing you any output.
It affects university submissions and the Lagos professional-writing economy in the same way, for the same reason.
An average false positive rate of 61.3% across seven detectors on TOEFL essays by non-native writers, their measurement rather than ours between them (Liang et al., 2023); native-written controls in the same study were barely flagged.
Detectors measure how predictable your word choices are and how much that varies sentence to sentence. They are calibrated against a largely American and British native-speaker baseline. Nigerian English has its own settled register and idiom, and prose written confidently in it can read as statistically unusual in one direction and unsurprising in another, neither of which has anything to do with whether a model wrote it.
The exposure isn't only academic. Nigeria has a large professional and content-writing economy where a client-side flag can end an engagement with no appeal. In that setting the useful artifact is the same as in a university: a record of how the draft was built, not a smoother version of the finished text.
Same advice for a lecturer and for a client, for different reasons.
Rewriting is expensive here in a specific way: it destroys the only asset that settles an authorship dispute. Your sentences, and the dated history behind them, are the evidence. Replace them and you are asking someone to trust a draft you didn't write either.
For a university, assemble the history and ask what the flag was based on. For a client in the Lagos content economy, where there's no appeal and no disclosure obligation, send the document history unprompted alongside the published research on variety bias. The list of what counts is here.
This is the part worth getting right, because an aggressive rewrite doesn't just risk your meaning. It pulls your register toward the American baseline the model was trained on, which means confident Nigerian English comes back sounding like nobody in particular.
So keep the setting low and the scope small. Light works on sentence rhythm and connective filler and leaves your vocabulary alone, which is exactly what you want when the vocabulary is the thing you would rather keep. Balanced edits phrasing more broadly. Maximum replaces most of the wording and is a paid mode.
Work a paragraph at a time: 3 humanizations a day at up to 260 words with no account, or 3 actions a day (scans or humanizations, any mix) at the same 260 words on a free account. Read the pair and check the three things that break: a citation's claim, a technical term, a hedge. The check is here.
No vendor can see another system's verdict, so nobody can honestly tell you a rewrite clears a particular university or client tool. Those systems also retrain against exactly these techniques. A guarantee here's a sentence, not a measurement.
We score before and after with our own detector, labelled as ours, and show which words moved. That is a signal about the statistical properties this family of tools reads.
On language: the detector is measured for English and reads Nigerian English as English. It isn't measured on Nigerian Pidgin, Yoruba, Hausa or Igbo, and we publish no accuracy claims for them.
Ours runs 3 humanizations a day at up to 260 words each with no account and no card, and a free account gives 3 actions a day, scan or humanization in any mix, at up to 260 words per humanization in Light and Balanced modes. Most of this category asks for payment details before you see output on your own text, so the no-card path is the main practical difference from Nigeria rather than a marketing line.
Because they measure statistical predictability against a largely American and British native-speaker baseline, and Nigerian English is a distinct variety with its own settled usage and idiom. Confident prose in that variety can read as unusual on the axes these systems measure, which has nothing to do with whether a model wrote it. Liang et al. (2023) measured an average false positive rate of 61.3% across seven detectors on TOEFL essays by non-native writers.
Do not rewrite it. A rewrite replaces the sentences you would use to prove authorship and destroys the draft history that supports you. Gather version history, earlier drafts, notes and outline, and be ready to explain your argument. Then cite the published research on detector bias, because it reframes the discussion from your character to a documented property of the tool.
It helps less than showing your process. A client-side detector inherits the same bias, and rewriting removes the evidence that would have settled it. Send the version history and the outline instead. Use a rewriter on the next piece while you are drafting it, where it genuinely improves prose rather than erasing your defence.
Our detector is built and measured for English and reads Nigerian English as English, so the before and after score is meaningful there. It is not measured on Nigerian Pidgin, Yoruba, Hausa or Igbo, and we publish no accuracy claims for them. If you draft in another language and translate, score the English version, because that is the version that gets read.
Light in almost every case. It works on sentence rhythm and connective filler and leaves your vocabulary alone. Balanced edits phrasing more broadly and suits general prose where exact wording is not load-bearing. Maximum replaces most of the wording and is a paid mode; the free tier covers Light and Balanced per our published plans.
No. Nobody selling a humanizer can see another system’s verdict, and institutional detectors retrain against rewriting techniques. A guarantee on this is a claim about a system the vendor cannot measure. What we show is a before and after reading from our own detector, labelled as ours, which tells you whether the properties this family of tools reads have moved.
Score a draft before a university or a client does.
See the detector →The research on detector bias, written to be forwarded.
Read the evidence →What to gather instead of rewriting.
See the evidence list →What drifts first when you rewrite.
Read the method →The full tool, scored before and after.
See the tool →What our own score cannot tell you.
See the limits →Score a paragraph, rewrite it on the lightest setting, and read what actually moved. 3 scans a day at up to 1,000 words, no account, no card.