Say the limit first, because it decides whether this page is useful to you: this is an English rewriter, not a Bahasa Indonesia one. The model that rewrites your draft and the detector that scores it are both English-optimised. If your text is in Bahasa Indonesia, Javanese or Sundanese, nothing here will help and we would rather you knew that now than after signing up.
For the large amount of work Indonesians produce in English, though, skripsi chapters, journal submissions, Upwork deliverables, Shopee and Tokopedia listings, SEO articles for international clients, it does something most rewriters do not: the same detector we sell standalone scores your text before and after each rewrite, so you can see what changed instead of trusting that something did.
The honest boundary, because most tools on this keyword leave it vague and let you find out by paying.
This is the case the tool is built for, and it is a large one. Indonesian higher education runs a great deal of English-language work: skripsi and thesis chapters written for international examiners, journal manuscripts, conference abstracts, and coursework at UI, ITB, UGM, IPB and UNAIR where the submission language is English. Outside campus, Indonesian freelancers write in English for Upwork and Fiverr clients, and marketplace sellers write English product copy for cross-border listings.
All of that is in scope. The rewriter changes cadence, sentence-length variation and phrasing; it does not change your argument, your data or your sources.
Bahasa Indonesia is out of scope. Our detector reports non-English text as unsupported rather than quietly scoring it, and a rewriter working from an English-tuned model would produce unreliable output anyway. We also do not translate: pasting Bahasa in and expecting English out is not what this does.
If your workflow is "write in Bahasa, translate to English, then submit," be aware the translation step is its own risk. In the largest independent detector study, machine-translated documents scored around 75% accuracy and roughly 11% false positives, against about 2% on ordinary human text, so translated work is flagged more often regardless of who wrote it.
Source: Weber-Wulff, Anohina-Naumeca, et al., International Journal for Educational Integrity, 2023 (arXiv:2306.15666), 756 tests across 14 tools.
Second-language English is the single most mis-scored category in AI detection, which cuts in an unexpected direction.
Writing in English as a second language produces lower perplexity and lower burstiness than native prose. Those are the exact two properties detectors read as machine-written. The consequence is measured and uncomfortable: a Stanford study found an average 61.3% false positive rate across seven detectors on TOEFL essays by non-native writers, against a near-zero rate on comparable native-English writing.
So an Indonesian writer using a humanizer is often in a different situation from the one the ads assume. The draft may be entirely your own work and still read as machine-written to a classifier. What you need in that case is not a stronger rewriter. It is to see which sentences trigger it, decide whether to vary them, and keep your drafts either way.
That is why the before-and-after score is the feature rather than the rewrite. A number that barely moves tells you the detector was reacting to your register, not to provenance, and more rewriting will not fix that.
Source: Liang et al. (2023), Patterns (Cell Press). Our detector was not among the seven tested.
Three groups, three different reasons, one workflow.
The highest-stakes case. The lines that usually score are the abstract, the restated research question and the closing summary, which are also the most formulaic paragraphs in any thesis. Seeing them highlighted is useful editorially, not just defensively. Keep your revision history throughout; on a contested flag it is stronger evidence than any rewrite.
Upwork and Fiverr clients increasingly run submissions through a detector before paying, and a false flag costs a contract. Checking your own work first, and being able to show a before-and-after reading, is a practical answer to a client dispute that "I wrote it myself" alone does not settle.
Shopee and Tokopedia cross-border listings, plus SEO articles aimed at international search, tend to be produced fast and in bulk. Here the value is less about detection and more about the rewrite catching the template rhythm that volume production creates. A 10,000-character per-request ceiling applies, so long pieces go through in passes.
Prices are in USD and billed in USD; there is no separate IDR price list.
Without an account: 3 rewrites a day, up to 300 words each. On a new free account: 260 words per rewrite, plus 3 scans a day of detection.
Paid begins at $9.99/month for 40,000 rewritten words a month, or $7.49/month billed annually. Pro is $19.99/month for 80,000 words. The pricing page carries the full ladder with both columns.
One thing worth checking before you buy: the free tier is usually enough to find out whether the score moves on your kind of writing. If it barely moves on a sample of your own prose, a paid plan will not change that, and you have learned something more useful than a subscription.
Short list, and the first one rules out the most common reason people arrive here.
It will not make text undetectable, and we will not use that word. The published research shows rewriting reduces detection substantially and never removes it: field-wide accuracy fell to about 26% on paraphrased AI text, and the strongest detectors dropped only to between 52% and 71% on humanised output. Detectors also retrain, so any guarantee is about a moving target.
It will not help you pass off machine-written coursework as your own. Your campus policy is the authority, and we are not a way around it. The tool exists to show you what a detector reacts to in your own writing.
It will not work in Bahasa Indonesia, and it does not translate.
The score is ours. A better Humanization Score is a reading from our detector, not a clearance from whichever tool your university or client runs.
It will not fix a thin draft. Rewriting changes how prose reads, not whether the argument holds or the sources are real.
No. Both the rewriter and the detector that scores it are English-optimised, so Bahasa Indonesia, Javanese and Sundanese are out of scope. Our detector reports non-English text as unsupported rather than quietly returning a number for it. The tool is for the English work Indonesians produce: skripsi chapters and journal manuscripts in English, freelance deliverables, and cross-border marketplace and SEO copy.
Yes, with real limits rather than a trial. Without any account you get 3 rewrites a day at up to 300 words each. On a new free account that becomes 260 words per rewrite, plus 3 scans a day of detection. Paid starts at $9.99/month for 40,000 words, or $7.49/month billed annually. Pricing is in USD; there is no separate IDR list.
Unfortunately yes, and it is a documented pattern rather than anything you did wrong. Writing in English as a second language produces lower perplexity and lower burstiness than native prose, which is exactly what detectors read as machine-written. A Stanford study measured an average 61.3% false positive rate across seven detectors on TOEFL essays by non-native writers, against near-zero on native writing. Use the sentence-level highlights to see what triggered it, and keep your revision history, which is stronger evidence than any rewrite.
We will not tell you it will. Rewriting measurably lowers detection scores and does not eliminate them: field-wide accuracy fell to roughly 26% on paraphrased AI text in a 14-tool study, and the strongest detectors still held between 52% and 71% on humanised output in a 2026 study. Detectors retrain too. What we give you is a reading from our own detector before and after, not a promise about someone else's.
You can, but know the risk before you rely on it. In the 14-tool Weber-Wulff study, machine-translated documents scored around 75% accuracy with roughly 11% false positives, against about 2% on ordinary human text. Translated work gets flagged more often regardless of who wrote it, so a translation step can create exactly the problem you were trying to avoid.
Meaning preservation is tuned for ahead of how far the score moves, because a rewriter that scrambles your argument has handed you a worse document. Still read the output before you submit, especially on cited, numerical or technical material, and confirm that terminology, figures and references survived the pass intact.
Neither is shipped, and we will not announce a date we cannot keep. The detector is English-only today, and billing is in USD. We do publish an Indonesian-language detector page at /id/detektor-ai/ describing what the tool does, but the scoring itself remains English-first, which is the same thing our Indonesia detector page says.
The detection side of the same workflow, built for Indonesian English.
Check a draft →The same tool explained in Bahasa Indonesia, with the English-only scope stated.
Baca dalam Bahasa →The full rewriter, with the Humanization Score shown on both passes.
Open the humanizer →Why the before-and-after comparison only works when one model does both.
See how it works →The second-language problem in detail, and what to do when it hits your work.
Read the guide →Monthly and annual, with the rewritten-word allowance on every tier.
See pricing →Paste an English draft and see the Humanization Score before and after, with the sentences that carry the signal marked. 3 rewrites a day free, no account and no card.