rel=nofollow citations. The promoted call-to-action above is a StealthWriter affiliate placement with its own disclosure.Does Text Polish bypass GPTZero? It says 97%, on a page dated ten months before its own website existed.
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
- Text Polish publishes a GPTZero leaderboard putting itself first at 97%. That post is dated 28 November 2024.
- The domain was registered on 4 October 2025. We confirmed it today at Verisign’s registry — the post is dated 310 days before the website existed.
- We pulled the schema date from all 165 blog posts. 21 of them predate the domain. They include both pages carrying GPTZero numbers.
- Its new dedicated /bypass/gptzero page carries no percentage at all — no method, no date, no screenshot. The numbers live only on the backdated posts.
- That guide describes GPTZero’s method as “perplexity and burstiness analysis” at “~80% accuracy”. GPTZero’s current FAQ uses neither word, and publishes no such figure.
- The bylines include “Dr. Sarah Chen, Educational Technology Expert” — a bio naming no institution, no degree and no verifiable profile, with an initial where a photo would be.
- Refunds: “all purchases are non-refundable” with an explicit waiver of the 14-day withdrawal right — on a site whose own blog warns readers off “services without money-back guarantees”.
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 Text Polish bypass GPTZero?
There is one published number and it does not survive a date check.
Text Polish is not quiet about detectors. GPTZero is named across its site second only to Turnitin, it has a dedicated conversion page at /bypass/gptzero, and its homepage sells “Bypasses every major detector.” But the only place a GPTZero figure actually appears is on a blog post whose date is impossible.
This page reports no first-party test of the product. Every check below was run on 4 September 2026 and is reproducible from public records.
What does the date check show?
The domain text-polish.com was registered on 4 October 2025 at 17:45 UTC through Namecheap. That is not an estimate: it is the registration event in Verisign’s registry record, which we queried directly today.
We then pulled the datePublished value out of the structured data on all 165 blog posts in the sitemap, programmatically rather than by sampling.
| Check | Result |
|---|---|
| Domain registration (Verisign RDAP) | 4 October 2025 |
| Blog posts with a schema date | 165 |
| Posts dated before the domain existed | 21 |
| Earliest post date | 28 November 2024 — 310 days early |
| Post carrying the GPTZero leaderboard | dated 28 November 2024 |
| Post carrying the GPTZero method description | dated 10 December 2024 |
The two posts carrying every GPTZero claim on the site are both in the backdated set. The earliest one publishes this ranking:
“Against GPTZero: 1. TextPolish: 97% success rate. 2. Undetectable.ai: 90% success rate. 3. HIX Bypass: 85% success rate. 4. BypassGPT: 78% success rate. 5. StealthWriter…”text-polish.com/blog/best-ai-humanizer-tools, datePublished 28 November 2024, read 4 September 2026
A competitive leaderboard, with the publisher in first place, carrying no sample size, no corpus, no model and no test date — presented as having been published before the publisher had a domain.
Why does a date matter on a bypass claim?
Because it is doing quiet work that most readers never audit.
A 97% GPTZero figure stamped November 2024 reads as a number with almost two years of standing behind it. That is the impression the date creates, and it is the impression that makes the number persuasive. Here the number was published by a site that did not exist on the date shown, about a detector that has shipped multiple model revisions since — the current build stamps results as Model 4.4b.
What does the dedicated GPTZero page say?
Less than the backdated blog, which is its own kind of answer.
The page at /bypass/gptzero is headed “Bypass GPTZero AI Detection” and states that Text Polish “rewrites AI text so it reads as natural human writing, reliably bypassing GPTZero’s detection”, and that your output “is now undetectable by GPTZero and other AI tools.”
It carries no percentage, no results table, no sample size, no test date, no mode and no screenshot. It is one of seven /bypass/ pages added since our previous crawl, and none of them publish a number.
Its description of the detector is also stale. The backdated guide characterises GPTZero as using “perplexity and burstiness analysis” with “accuracy ~80% for ChatGPT content”. Perplexity and burstiness were GPTZero’s 2023 framing; its current FAQ uses neither word, describing instead classification into HUMAN_ONLY, MIXED or AI_ONLY with confidence scores, AI-vocabulary highlighting and sentence-by-sentence highlighting. The ~80% accuracy figure is attributed to no source at all.
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.
Which is why the absence is worth naming. There is no detector screenshot anywhere on the site, for GPTZero or anything else. The “AI DETECTED (94%)” panel on the homepage is hardcoded HTML inside a mock interface, not a captured result. The headline counters beside it ship in the delivered source as “0+ Happy Users”, “0% Success Rate” and “0 Detectors Supported”, animated client-side with no server-rendered value behind them.
Who is writing this?
Six bylines appear across the blog: Dr. Sarah Chen, Alex Rivera, Jennifer Liu, Marcus Thompson, Maria Rodriguez and TextPolish Team.
The Sarah Chen page is headed “Dr. Sarah Chen — Educational Technology Expert” and says she “specializes in the intersection of artificial intelligence and academic writing” with “a background in educational technology.” It names no institution, no degree, no field of doctorate, no publication and no external profile, and the avatar is the letter D rather than a photograph.
Has any detector company tested it?
None has. Text Polish is absent from the appendix table in GPTZero’s February 2026 paper, which names nine bypass services individually; absent from the nineteen tools in Pangram’s DAMAGE paper; and absent from the twenty in Pangram’s August 2025 benchmark. A full sitemap crawl also finds Pangram mentioned zero times on the site, which we cover in does Text Polish bypass Pangram.
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 happens if it does not work?
Nothing you can act on, and the site says so while advising the opposite.
Its terms state that all purchases are non-refundable, and include an explicit waiver of the 14-day right of withdrawal. There is no named legal entity, no company number and no address. Governing law is “the jurisdiction in which TextPolish operates” — a jurisdiction the site never names. Contact is a Gmail address.
Meanwhile its own blog advises readers to be wary of “services without money-back guarantees.”
datePublished value programmatically — a census, not a sample. We queried Verisign’s RDAP record for the domain to obtain the registration event date. We then read at source the two posts carrying GPTZero claims, the new /bypass/gptzero page and one author page, quoting each verbatim and recording the presence or absence of a percentage, sample size, method, date and screenshot. We compared the site’s technical description of GPTZero against GPTZero’s own current FAQ language. Refund and governing-law terms are quoted from the site’s own terms page. We have no commercial relationship with Text Polish.What should you take from this?
The honest position is that nobody knows whether Text Polish beats GPTZero, including Text Polish, because it has published nothing that would tell you.
What it has published is a 97% leaderboard dated before its domain existed, a detector description two architectures out of date, a conversion page with no number on it, a mock detector panel hardcoded into the homepage, counters that ship as zeroes, a doctorate with nothing behind it, and terms making every purchase non-refundable.
Check the date on any bypass figure before you weigh it. Then check your own text: gptzero.me takes 10,000 characters free with no account, and the result it gives you will be dated today.
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 Text Polish review, and the same question against Turnitin and Pangram. Also in this cluster: AuraWrite, UnAIMyText, Ryne AI.
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 Text Polish bypass GPTZero?
Nothing checkable says so. Its only published GPTZero figure is 97%, and that figure sits on a blog post dated 28 November 2024, roughly ten months before the domain was registered. Its current dedicated GPTZero page publishes no number, no method and no screenshot. GPTZero has never tested Text Polish, and no independent party has published a result on it.
What exactly is wrong with the dates?
The domain text-polish.com was registered on 4 October 2025 at 17:45 UTC, which we confirmed directly against Verisign’s registry record on 4 September 2026. We then pulled the datePublished value from all 165 blog posts in its sitemap. 21 of them carry dates earlier than the registration, the earliest being 28 November 2024. That post is the one containing the GPTZero leaderboard.
Why does a publication date matter for a bypass claim?
Because readers and search engines both use dates to judge whether a claim has stood the test of time. A 97% GPTZero figure presented as having been published in late 2024 reads as a number with nearly two years of track record behind it. In this case it is a number published by a website that did not exist on the date shown, about a detector that has been rebuilt repeatedly since.
What does the GPTZero page itself claim?
Assertion without measurement. The page at /bypass/gptzero is headed “Bypass GPTZero AI Detection” and says Text Polish “rewrites AI text so it reads as natural human writing, reliably bypassing GPTZero’s detection” and that the result “is now undetectable by GPTZero and other AI tools.” There is no percentage, no sample size, no test date, no methodology and no screenshot anywhere on it.
Is its description of GPTZero accurate?
It is out of date. The backdated guide describes GPTZero’s detection method as “perplexity and burstiness analysis” with “accuracy ~80% for ChatGPT content.” Perplexity and burstiness were GPTZero’s original 2023 framing; its current FAQ does not use either term and instead describes classification into HUMAN_ONLY, MIXED or AI_ONLY with confidence scores and sentence-level highlighting. The ~80% accuracy figure is attributed to nothing.
Who writes these posts?
Six bylines appear, including “Dr. Sarah Chen, Educational Technology Expert”, “Alex Rivera”, “Jennifer Liu”, “Marcus Thompson”, “Maria Rodriguez” and “TextPolish Team”. The Sarah Chen page asserts a doctorate and a research background but names no institution, no degree, no field, no publication and links to no external profile, and the avatar is a letter rather than a photograph. We are not asserting the person does not exist. We are recording that nothing on the page allows a reader to check.
Has any detector company tested it?
None has. Text Polish is absent from the appendix table in GPTZero’s February 2026 paper naming nine bypass services, and absent from Pangram’s DAMAGE paper and its August 2025 benchmark of twenty tools. We also crawled the full sitemap and found Pangram mentioned zero times, which we cover in the Pangram version of this question.
What happens if it does not work?
Nothing. The terms state that all purchases are non-refundable and include an explicit waiver of the 14-day right of withdrawal. There is no named legal entity, no company registration number and no postal address; governing law is given as “the jurisdiction in which TextPolish operates”, which is never specified, and contact is a Gmail address. Its own blog, meanwhile, advises readers to avoid “services without money-back guarantees.”
