rel=nofollow citations.Does StealthWriter bypass GPTZero? Its website no longer says. Its blog still does, in a tutorial that accidentally shows half the job can be done with typos.
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
- StealthWriter’s current website makes no detection claim at all. The hero reads “Humanize your AI writing”, and GPTZero is named nowhere on it. In this market that restraint is rare.
- Its blog still hosts a GPTZero bypass tutorial, claiming StealthWriter output was “successfully able to bypass GPTZero and not a single sentence was detected as AI-generated.”
- That tutorial carries no date, no sample size and no method.
- It also undercuts the product it sells. The same post reports that “almost half of the text was able to bypass GPTZero using typos and adding punctuation” — before any tool was involved.
- Its Terms, updated 5 September 2026, prohibit “misrepresenting content to plagiarism detection systems” — which is what the tutorial teaches.
- The detection disclaimer is the flattest in this market: “StealthWriter does not guarantee that content generated by the tool will be scored as human-written by AI detectors.”
- GPTZero’s own February 2026 paper rates StealthWriter “Low” — a column that scores tools against naive detectors, not against GPTZero.
Does StealthWriter bypass GPTZero?
There are two StealthWriters, and they disagree.
The one at stealthwriter.ai has been cleaned up. Its hero reads “Humanize your AI writing… StealthWriter rewrites AI-generated content so it reads naturally human, polished, and ready to use.” The word bypass does not appear. GPTZero is named nowhere. Its own FAQ answer to whether the tool works is about “natural phrasing, rhythm, and flow” — readability, not detection.
The one at blog.stealthwriter.ai still publishes a step-by-step tutorial on beating GPTZero.
This page reports no first-party test. Everything below was fetched on 21 September 2026 and quoted verbatim.
What does the tutorial claim?
A total pass, in plain terms:
“The text generated by Stealthwriter was successfully able to bypass GPTZero and not a single sentence was detected as AI-generated.”blog.stealthwriter.ai, GPTZero vs Stealthwriter, read 21 September 2026
It carries no publication date, no sample size, no corpus, no generating model and no stated method. It describes screenshots rather than publishing a reproducible test, and because GPTZero stamps every live result with a model version, an undated detector image cannot be tied to any particular build of the classifier.
| Surface | What it says about detection | Dated? |
|---|---|---|
| Homepage (stealthwriter.ai) | Nothing. “Humanize your AI writing”; the word bypass does not appear | — |
| FAQ | “natural phrasing, rhythm, and flow” — a readability claim | — |
| Blog tutorial | “not a single sentence was detected as AI-generated” | no date |
| Terms, section 7 | Prohibits “misrepresenting content to plagiarism detection systems” | 5 Sep 2026 |
| Terms, disclaimer | “does not guarantee that content…will be scored as human-written” | 5 Sep 2026 |
What does the same tutorial accidentally show?
This is the part worth the click, and it is not the part the post is selling.
Before applying StealthWriter at all, the author modifies a ChatGPT sample by hand and reports that “almost half of the text was able to bypass GPTZero using typos and adding punctuation.”
It is also a reminder of what these tools actually do to writing. Passing a detector by inserting errors is not the same as writing something worth reading.
Can you check GPTZero yourself?
Turnitin sells no individual licences and shows students no AI report, so nobody outside an institution can produce a genuine Turnitin screenshot. GPTZero removes that excuse entirely. Anyone can paste up to 10,000 characters into gptzero.me right now with no account, no card and no signup; a free account raises the per-scan limit to 150,000 characters with a 10,000-word monthly quota, and paid tiers start at $9.99 a month.
So when a humanizer claims to beat GPTZero and shows you nothing, that is a choice, not a constraint.
It also means you can check your own text. A genuine current GPTZero result shows a three-class classification — human, AI or mixed — a confidence category rather than a bare percentage, a probability breakdown across all three classes, and sentence-by-sentence highlighting. An image showing only “X% AI”, or one displaying perplexity and burstiness, is either fabricated or dates to roughly 2023.
Do its own Terms permit the tutorial?
Not on a plain reading, and the Terms are recent. They were last updated 5 September 2026, sixteen days before we read them.
They open with a definition rather than a disclaimer: “StealthWriter is not a cheating tool and must not be used for academic dishonesty or any form of academic misconduct.” Section 7 prohibits “submitting AI-generated or rewritten content as your own work in academic settings, misrepresenting content to plagiarism detection systems, or any other form of academic misconduct.”
A tutorial explaining how to make AI-generated text pass GPTZero is a description of misrepresenting content to a detection system. The prohibition and the instruction are published by the same company, on the same brand, in the same month.
And the detection disclaimer is the flattest we have recorded anywhere in this series:
“StealthWriter does not guarantee that content generated by the tool will be scored as human-written by AI detectors or plagiarism-checking software.”StealthWriter Terms of Service, updated 5 September 2026
Credit where it is due: that sentence is more honest than the guarantees printed by most of this cluster. It is also the opposite of what the blog post says.
What has GPTZero said about it?
GPTZero’s February 2026 paper names nine bypass services in an appendix table, and StealthWriter is one of them, rated “Low” alongside Quillbot, StealthGPT, HIX and GPTinf. Grubby AI, TwainGPT and WriteHuman are “Medium”; only Undetectable is “High”.
What does the independent record show?
Thin, and none of it about GPTZero.
Pangram’s DAMAGE paper studied StealthWriter and rated it L2, the middle tier — but that tier measures “faithfulness and fluency, not… effectiveness at bypassing AI detectors”, and the paper then excluded it from every detection experiment it ran. It is absent from Pangram’s August 2025 benchmark of twenty named tools, with no explanation for the removal. Pangram’s July 2026 paper anonymised all thirteen tools it tested, making the question permanently unanswerable from outside.
On Turnitin, the one independent bypasser test put a StealthWriter document in the unspecified 1–19% band rather than a clean pass. We cover both in does StealthWriter bypass Pangram and does StealthWriter bypass Turnitin.
How reliable are the numbers on either side?
GPTZero’s own FAQ concedes: “Our classifier is not trained to identify AI-generated text after it has been heavily modified after generation.” Its developers page simultaneously claims a “Paraphraser Shield” means “even if AI content has been altered to look more human-like, GPTZero can detect it.” Both statements are currently published.
The two leading detector vendors publish opposite numbers on the same question. GPTZero’s February 2026 paper reports 93.5% recall on 1,000 texts run through nine bypasser services, and puts Pangram at 49.7%. Pangram’s DAMAGE paper reports GPTZero at 60.04% on humanized text, and 34.53% at GPTZero’s own default threshold. Each benchmark shows its publisher winning, and neither is peer-reviewed.
On false positives, GPTZero was one of seven detectors in Liang et al. (2023), which reported a 61.22% average false positive rate across those seven on 91 TOEFL essays. No per-detector figure was ever published, so that number is not GPTZero’s. GPTZero now claims it has cut its own TOEFL false positive rate to 1.1% — a figure no third party has verified.
What should you take from this?
StealthWriter’s current website is one of the more honest in this market. It promises readability rather than invisibility, its Terms forbid academic misconduct in specific language, and its detection disclaimer refuses to guarantee an outcome. Several of its competitors guarantee 100% and then disclaim it three clicks away.
What has not been cleaned up is the blog, which still tells you the tool beat GPTZero completely, with no date and no method, on the same brand whose Terms prohibit misrepresenting content to detection systems. And the most checkable number in that post says half the effect came from typing badly on purpose.
Believe the Terms, not the tutorial. Then scan your own text at gptzero.me, free, no account, up to 10,000 characters, and read the verdict with today’s model version stamped on it — and keep your drafting history either way.
If your real worry is being wrongly flagged on your own writing, keep your drafting history — it costs nothing and it 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 StealthWriter review, and the same question against Turnitin and Pangram. Also in this cluster: Rephrasy, GPT Scrambler, HueWrite.
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 StealthWriter bypass GPTZero?
Its own blog says yes and its own contract declines to promise it. A tutorial on blog.stealthwriter.ai states that StealthWriter output was successfully able to bypass GPTZero with not a single sentence detected as AI-generated. No date, sample size or method accompanies that. The Terms, updated 5 September 2026, state that StealthWriter does not guarantee content will be scored as human-written by AI detectors. No independent party has published a result, and GPTZero has never tested it directly.
Why is the main site different from the blog?
Because the site has been sanitised and the blog has not. The current homepage hero is Humanize your AI writing, the word bypass does not appear on it, and GPTZero is named nowhere. Its FAQ answer to whether it works is a readability claim rather than a detection claim. The blog, on a separate subdomain, still carries detector-specific bypass tutorials for GPTZero, Copyleaks, Originality and ZeroGPT.
What is the typos finding?
The most useful number in the tutorial, and it argues against buying anything. The post reports that almost half of the text was able to bypass GPTZero using typos and adding punctuation, before StealthWriter was applied at all. If roughly half the effect comes from manually degrading your own text, that reframes what the remaining half is worth paying for. We have not reproduced it, and neither has anyone else, because no method was published.
Do its Terms allow what the tutorial teaches?
No, on a plain reading. The Terms open by defining StealthWriter as not a cheating tool that must not be used for academic dishonesty. Section 7 prohibits submitting AI-generated or rewritten content as your own work in academic settings, and specifically prohibits misrepresenting content to plagiarism detection systems. A tutorial explaining how to make AI text pass GPTZero is a description of misrepresenting content to a detection system. Both documents are published by the same company.
Does GPTZero name StealthWriter?
Yes. GPTZero’s February 2026 paper includes an appendix table naming nine bypass services, and StealthWriter is rated Low alongside Quillbot, StealthGPT, HIX and GPTinf. Read that carefully: the column is headed bypassing ability on naive AI detection methods, so it scores those tools against weak detectors, not against GPTZero. The caption then asserts that all nine are ineffective against GPTZero, which is the publisher’s own claim about its own product.
Is there any independent evidence?
Very little, and none of it is about GPTZero. StealthWriter appears in Pangram’s DAMAGE paper rated L2, but that tier measures faithfulness and fluency rather than evasion, and the paper excluded it from every detection experiment. It is absent from Pangram’s August 2025 benchmark of twenty tools. In the one independent Turnitin bypasser test, a StealthWriter document landed in the unspecified 1 to 19 percent band rather than a clean pass.
Does it publish any GPTZero screenshot?
Not on the main site. The blog tutorial describes screenshots of GPTZero results, including a claim that a longer ChatGPT sample was detected as AI with 95% probability, but the post publishes no reproducible method and no dates, so the images cannot be tied to any particular model version. GPTZero currently stamps every result with a model version, which is what makes an undated detector screenshot hard to weigh.
What does it cost, and what is the refund position?
There is a free tier at $0. Paid plans run Starter at $20 a month, Plus at $50, Pro at $100 and Scale at $400. The refund window is 24 hours and voids the moment you use the product, with no performance guarantee of any kind attached to detector outcomes.
