Does UnAIMyText bypass Turnitin? It is the one tool here that publishes a method — and the method is precisely what makes the claim impossible to accept.
The only tool that bypassed Pangram and Turnitin in 2026.
Most humanizers clear one detector and get caught by the other. StealthWriter is the one that gets past both — run Ghost 5.2 Pro at level 7–8, section by section, and re-check before you submit.
Try StealthWriter →Key Takeaways
- UnAIMyText is the only tool in this market that publishes a Turnitin methodology — 50 samples a round, named corpus models, exact settings, a month-by-month history.
- That is the problem. A disclosed method invites the obvious question: how did it run Turnitin 50 times? Turnitin sells no individual licences and shows students no AI report.
- It claims 94% of raw AI text caught by Turnitin, falling to 2% after processing — a 98% pass rate. There is not one screenshot anywhere on the site.
- The test corpus is built from 2024-era models — GPT-4o, Claude 3.5 Haiku, Gemini 1.5 Flash — run in July 2026 against a detector updated for GPT-5 and Claude 4.5.
- Only two months of the “monthly” history exist. The August row still read “pending” on 31 August.
- To its credit, its homepage warns readers off exactly this kind of claim: “no humanizer can promise a specific score on every detector, every time.”
- Turnitin has published no accuracy figure at all for the bypasser detection it shipped in August 2025. Neither side of this has a number you can audit.
The only tool that bypassed Pangram and Turnitin in 2026.
Most humanizers clear one detector and get caught by the other. StealthWriter is the one that gets past both — run Ghost 5.2 Pro at level 7–8, section by section, and re-check before you submit.
Try StealthWriter →Does UnAIMyText bypass Turnitin?
It says so, with a number and a stated method. That already puts it ahead of every competitor in this cluster, most of which assert “all major detectors” and publish nothing.
Its detector-tests page reports Turnitin catching 94% of raw AI text, falling to 2% after Ultra processing — a 98% pass rate. Alongside it: GPTZero 97%, Copyleaks 96%. Last round July 2026.
And it discloses how. Fifty samples per round. The models that generated the corpus, broken out by count. The content-type split across academic, business, blog and creative. The exact humanizer configuration. A 250 to 600 word range. All detectors run inside a four-hour window “to minimize model drift”. Error samples discarded and replaced. A five-point limitations section, and a month-by-month history rather than one flattering snapshot.
This page reports no first-party test. Everything below was fetched on 31 August 2026.
Why does the disclosure create the problem?
Because a stated method can be interrogated, and a vague boast cannot.
“Fifty samples per round, checked against Turnitin” is a specific operational claim. It says somebody submitted fifty documents to Turnitin, read fifty AI writing reports, and did it again the following month. So the fair question is: through what account?
Turnitin’s help centre states it “does not sell individual licenses or single-use subscriptions”. Inside a licensed institution, “AI writing detection reports are not visible to students” — an instructor must download and share them. Draft Coach, the one student-facing product, has no AI detection at all.
Every other figure on that page is reproducible by a reader. GPTZero, Copyleaks and ZeroGPT all sell public self-serve access. Turnitin is the single row nobody can check, and it carries the highest number.
What else weakens the number?
Three things, none of which requires doubting anyone’s honesty.
| What the page discloses | Why it matters |
|---|---|
| Corpus generated by GPT-4o, GPT-4o mini, Claude 3.5 Haiku, Gemini 1.5 Flash | All 2024-generation models, run in July 2026 — against a detector Turnitin says now covers GPT-5, Claude Opus 4.5, LLaMA-4 and Grok 4.1 |
| Monthly history: June and July 2026 only | Two rows is not a trend, and the August row still read pending on 31 August |
| Prose describes Standard mode; figures are Ultra | The narrative and the numbers describe different configurations |
| No raw samples, no per-sample data, no screenshots | Every figure is a self-reported HTML table — nothing is auditable |
Set against detectors calibrated on 2026 output, a corpus of 2024 model text is an easier test than the one a student would actually face. That inflates a pass rate without anyone intending it to.
The vendor’s own caution is the best thing on the site
Directly beneath its results table, the homepage says: “no humanizer can promise a specific score on every detector, every time, and you should be sceptical of any tool that claims otherwise. Detectors change constantly and also flag fully human writing as AI.”
Its Terms add: “We do not guarantee that humanized content will be undetectable by all AI detection tools at all times”, and a separate clause states it “does not condone or encourage the submission of humanized AI text as original, unaided work” where an institution prohibits it.
Unlike most of this cluster, there is no gap between the marketing and the contract — the caution appears in both. That is genuinely better practice, and it is worth saying so plainly. It also means the strongest argument against the 98% figure is the vendor’s own sentence about being sceptical of specific scores.
What can be checked on the other side?
Almost nothing, which is the deeper problem with this entire question.
Turnitin shipped AI paraphrasing detection in July 2024 and dedicated bypasser detection in August 2025, and has published no accuracy figure for either in the year since. It has never published an overall accuracy percentage at all — only false-positive rates, under 1% at document level and around 4% at sentence level, the latter a June 2023 number never revised.
On 4 August 2026 it merged the AI-paraphrased category into AI-generated, so instructors now see a single blue highlight and can no longer tell how much of a flag came from bypasser detection. The last externally observable signal is gone.
The independent picture, meanwhile, is not flattering to the detector. Hadra, Cambridge and Mesbah (February 2026, 192 texts) measured Turnitin at 0.61 overall accuracy and 0.51 recall, and found it performed poorly on hybrid human-and-AI text — exactly what a humanizer produces. Weber-Wulff et al. (2023) put field-wide accuracy on machine-paraphrased text at 26%, with Turnitin the best of 14 tools. No independent study has tested the bypasser feature since launch, because researchers cannot buy access either.
What should you take from this?
UnAIMyText has professionalised the presentation of evidence further than anyone else here, and it deserves the credit. But presentation is not verification. Its Turnitin row rests on an unnamed access route, a 2024-model corpus, two months of history and zero artefacts — and its own homepage tells you not to trust specific scores.
Betting a submission on either side of this is a bad trade in both directions. Turnitin’s own guidance says its model “may misidentify human-written, AI-generated, and AI-paraphrased text, so it should not be used as the sole basis for adverse actions against a student”, and four universities — Vanderbilt, Waterloo, Curtin and Washington State — have disabled it with documented reasons.
If your real worry is being wrongly flagged on your own writing, keep your drafting history — it costs nothing and it is the one defence that survives a detector being wrong. It matters most for the writers detectors treat worst: ESL writers and AI detection and false positives for neurodivergent students. Check before you submit with the best pre-submission check and the Turnitin self-check, and strip the obvious tells by hand first via what to remove before using an AI humanizer.
For the product itself see our UnAIMyText review, and for the detector it never names, does UnAIMyText bypass Pangram. Other Turnitin cells: Clarity Bubble, WriteHuman, Ryne AI, Undetectable AI, StealthGPT, GPTHuman and SuperHumanizer.
Get the free prompt pack instead
The manual rewriting prompts that lower AI signals without a subscription — and without betting a submission on a number nobody has reproduced. Free, no card.
Frequently asked questions
Does UnAIMyText bypass Turnitin?
It publishes a figure saying it does — a 98% pass rate — and it is the only tool in this category that publishes a method alongside it. But the method cannot be reconciled with how Turnitin is sold. Turnitin does not sell individual licences and does not show students the AI report, so there is no disclosed route by which 50 monthly Turnitin checks could have been run.
What exactly does it claim, and what does it disclose?
Its detector-tests page reports Turnitin catching 94% of raw AI text, falling to 18% after Standard mode and 2% after Ultra — a 98% pass rate, alongside GPTZero at 97% and Copyleaks at 96%. It discloses 50 samples per round, the models that generated the corpus with counts, a content-type split, exact humanizer settings, a 250 to 600 word range, a four-hour testing window and a limitations section. That is more disclosure than most detector vendors publish about themselves.
So why not take the number at face value?
Four reasons, and only one of them is the access problem. The corpus is generated by 2024-era models — GPT-4o, GPT-4o mini, Claude 3.5 Haiku, Gemini 1.5 Flash — and run in July 2026 against a detector Turnitin says now covers GPT-5, Claude Opus 4.5 and Grok 4.1. Only two months of the monthly history exist. The prose describes Standard mode while every headline figure is Ultra. And there is no raw data, no per-sample result and no screenshot of any kind.
Why can’t anyone verify a Turnitin number?
Turnitin states it “does not sell individual licenses or single-use subscriptions”, its AI writing report is instructor-facing and “not visible to students”, and Draft Coach — the only student-facing Turnitin product — has no AI detection. So no vendor, and no reviewer including us, can lawfully obtain a genuine Turnitin AI score.
Does UnAIMyText itself warn about bypass claims?
Yes, and it is the reason this page is not a hit piece. Directly under its own results table the homepage says: no humanizer can promise a specific score on every detector, every time, and you should be sceptical of any tool that claims otherwise. Its Terms add that it does not guarantee humanized content will be undetectable by all AI detection tools at all times.
Has Turnitin published anything to check this against?
No. It shipped AI paraphrasing detection in July 2024 and dedicated bypasser detection on 27 August 2025, and has published no accuracy figure, detection rate, benchmark or sample size for either. On 4 August 2026 it merged the AI-paraphrased highlight category into AI-generated, removing the last signal instructors had for how that capability performs.
What does it cost, and can I get a refund?
A visitor gets 450 words per input free with no account; a free account gets 500. Basic is $9.99 a month for 1,000 words per input and 30,000 a month; Pro is $14.99, anchored against a struck-through $24.99, for 2,000 per input and 100,000 a month. Refunds run 7 days and only if fewer than 2,000 words have been used — 6.7% of one month on Basic.
Who operates it?
“CoreAI LLC”, named consistently across the legal pages, with no state of incorporation, no registration number and no postal address published. The contact page lists “Location: Global Support” and “Call us at: Coming Soon.” None of the four legal pages carries a last-updated date.
