Start with the thing most pages on this keyword will not tell you: no one outside a licensed institution can test a humanizer against Turnitin. Turnitin is sold to universities and schools, not to individuals, so there is no account any tool vendor can buy to benchmark against it. Any product claiming a measured Turnitin bypass rate is claiming a test it could not have run.
What is measurable is what rewriting does to detectors in general, and that research is public. It shows rewriting cuts detection substantially and never eliminates it. Below: what the studies actually found, what Turnitin publishes about itself, the honest use for a humanizer before you submit, and the things we will not claim.
This is the structural fact that makes most marketing on this keyword impossible to support.
Turnitin licenses to institutions and has no consumer tier. Submissions flow through a university or school account, tied to a course, an assignment and a grader. There is no way for an individual, or for a humanizer company, to create an account and run a batch of rewritten samples through it to measure a pass rate.
So when a tool advertises "bypasses Turnitin" or quotes a percentage against it, ask where the number came from. The plausible answers are a handful of samples run through a student's own institutional account, which is a tiny and non-repeatable sample, or nothing at all.
We are in the same position. We have no Turnitin access and we have run no test against it, so we do not publish a figure, and this page will not give you one. What we can do is point at the peer-reviewed research on rewritten text, which is public and was run by people who documented their method.
Not Turnitin specifically, but the category it sits in. The direction is consistent across every study.
This nine-author study in the International Journal for Educational Integrity ran 756 tests across 14 detection tools, including Turnitin. Average accuracy on AI text fell to about 42% after manual editing and about 26% after a pass through QuillBot. Stated the other way, roughly 52% of manually revised AI text and 71% of paraphrased AI text went undetected.
The authors put it plainly: content obfuscation techniques "significantly worsen the performance of tools." Turnitin scored highest of the 14, and the same paper still concluded no tool was reliable enough for standalone misconduct decisions.
Source: Weber-Wulff, Anohina-Naumeca, et al., "Testing of detection tools for AI-generated text," International Journal for Educational Integrity, 2023 (arXiv:2306.15666).
A University of Missouri study tested six detectors on 294 samples. On reasoning-model output run through a humanising rewrite, the strongest detectors dropped to 71% for Copyleaks, 58% for QuillBot and 52% for GPTZero. The authors called humanisation "the most effective adversarial attack" in the study.
Read the last number carefully. 52% is not a bypass. It is a coin flip that still lands on "AI" half the time, and that is against the weakest of the three, in a study none of those tools had a chance to tune against.
Source: Alshammari and Rao (arXiv:2507.17944), University of Missouri.
Rewriting moves the needle a long way and stops well short of a guarantee, against tools that are weaker and more testable than Turnitin. Turnitin ranked first in the one study that measured it beside its competitors, retrains on its own schedule, and publishes nothing you could tune against.
The rational conclusion is not "humanizers do not work." It is that a rewrite is a probabilistic change to a probabilistic reading, and betting an academic record on it is a bad trade at any price.
Worth knowing before you act on any score, and it cuts both ways: it is also the basis for challenging a wrong flag.
Two numbers from Turnitin's own 2024 accuracy documentation are worth carrying into any conversation about a flag: under 1% false positives at document level above its 20% reporting threshold, and about 4% across the full distribution. Independently, Weber-Wulff measured 4% to 12% for Turnitin, the higher end driven by short submissions and translated text. Our full audit of those figures has the eval-set caveats and the ESL evidence.
The threshold matters as much as the rate. Turnitin tells instructors not to act below 20%, and to treat 20% to 50% as a conversation rather than proof. Vanderbilt disabled the feature outright in August 2023 over fairness concerns. None of that is an argument for rewriting AI work; it is the argument for keeping drafts if you wrote your own.
If you are here because you were wrongly flagged, those figures matter more than any humanizer. A 4% rate across a 200-student cohort is roughly eight human-written papers flagged, and version history is stronger evidence than any rewrite.
Sources: Turnitin AI Writing Detection Accuracy Documentation (2024); Weber-Wulff et al. (2023); Vanderbilt University Center for Teaching, August 2023.
There is a legitimate reason to run your own draft through a humanizer before submitting, and it is not the one the ads sell.
Our humanizer scores your draft with the same detector we sell standalone, before and after the rewrite, and highlights the specific sentences carrying the machine signal. On most student drafts those are the opening restatement of the question, the transitional scaffolding, and the summary paragraph that repeats the body. They are also, usually, the weakest writing in the document.
Rewriting those, or better, rewriting them yourself once you can see which they are, improves the essay and lowers the signal at the same time. That is a real benefit and it does not depend on any claim about Turnitin.
This is common and it is not your fault. Formal academic register, technical writing and English as an additional language all produce low perplexity and low burstiness, which is the exact statistical signature detectors read as machine-written. Seeing which sentences trigger it tells you what a grader's tool will react to, and lets you decide whether to vary the rhythm or simply keep your drafts.
Google Docs revision history, Word autosaves, commit logs, notes and outlines are the only evidence in this area that no tool can manufacture and no rewrite can substitute for. A document that visibly grew over eleven sittings is a stronger answer to a flag than any score. If a dispute is already underway, our appeal-letter template cites the published literature rather than arguing about your percentage.
If you came for a guaranteed bypass, this is the part that saves you the money.
We will not say "undetectable". Not about Turnitin, not about anything. Detectors retrain, and a guarantee about a moving target is marketing rather than a measurement.
We have not tested against Turnitin and will not imply that we have. No individual can buy access, so no such test exists on our side. Any number we quoted would be invented.
We will not help anyone launder AI work into a graded submission. The tool exists to show you what a detector reacts to in your own writing so you can improve it. If your plan is to submit machine-written work as your own, we are not the right product and we would rather say so here than after you have paid.
The score is ours, not a universal verdict. A better Humanization Score is evidence about our detector. Turnitin runs a different model against a different threshold and may read the same paragraph differently.
It is English-only, and a 10,000-character per-request cap applies before any monthly word allowance.
3 rewrites a day at up to 300 words with no account. On a new free account, 2,000 words a month with 1,500 words per request. Paid starts at $9.99/month for 20,000 words, or $7.49/month billed annually, and Pro is $19.99/month for 50,000 words. Full ladder on the pricing page.
Nobody can honestly tell you, including us. Turnitin is licensed only to institutions, so no individual and no tool vendor can create an account and measure a pass rate against it. Any product quoting a Turnitin bypass percentage is quoting a test it had no way to run. What the public research shows is that rewriting reduces detection substantially across the detectors that can be tested, and never eliminates it.
Two studies matter. Weber-Wulff et al. tested 14 tools across 756 tests and found average accuracy falling to about 42% on manually edited AI text and about 26% after a QuillBot pass, with roughly 71% of paraphrased AI text going undetected. A University of Missouri study measured the strongest detectors dropping to between 52% and 71% on humanised reasoning-model output, calling humanisation the most effective adversarial attack it tested. Substantial reduction, not elimination.
That depends entirely on what you are rewriting. Running your own writing through a tool to see which sentences read as machine-generated, then improving them, is editing. Generating an essay with a chatbot and rewriting it to avoid detection is submitting work that is not yours, whatever the tool did in between. We build for the first case and we will not help with the second. Your institution's policy is the authority, not us.
It can show you what triggered it, which is genuinely useful, but it is not the strongest response. Formal academic register, technical writing and English as an additional language all produce the low perplexity and low burstiness detectors read as AI. Turnitin's own documentation reports about 4% false positives across the full distribution, and Weber-Wulff measured 4% to 12%. Your drafts and version history are better evidence than a rewrite, and our appeal-letter template cites the literature rather than your score.
Turnitin's documentation, last updated 2024, states under 1% at document level when the AI percentage is above its 20% reporting threshold, and approximately 4% across the full distribution, on an internal eval set of undisclosed size. Weber-Wulff independently reported 4% to 12%, the higher end driven by short submissions and translated text. Turnitin recommends instructors do not act below 20% and treat 20% to 50% as a conversation, not evidence.
No, and we will not imply otherwise. There is no consumer access to Turnitin, so the test is not available to us. What we do publish is our detector's own false-positive measurement with the dataset attached, 1,180 academic papers at 5.85%, and the before-and-after Humanization Score on your actual text so you can read a real measurement instead of a claim about a tool we cannot reach.
Run your own draft, read the sentence-level highlights, and rewrite the flagged lines in your own words rather than accepting every suggestion. The lines that score highest are usually the opening restatement, the transitional scaffolding and the closing summary, which tend to be the weakest writing anyway. Keep your version history throughout. That sequence improves the essay and leaves you with evidence, which is more than any bypass claim offers.
The published false positive rate, the 20% threshold, and the ESL evidence.
Read the audit →Why Turnitin is closed to individuals, and what you can actually use instead.
See the options →The version that does not involve laundering anything: write, check, keep your drafts.
Read the guide →The tool itself, with the Humanization Score shown before and after every rewrite.
Open the humanizer →If you were flagged for work you wrote, this cites the research instead of your score.
Use the template →1,180 academic papers, 5.85% measured rate, dataset downloadable so you can check it.
Check our numbers →See which sentences a detector reacts to in work you wrote, before a grader does. Free, and with no number attached about a tool nobody outside an institution can test.