Does Text Polish bypass Pangram? Unknown — and a tool marketed as bypassing “every major detector” has never once named the detector that benchmarks humanizers.
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 sells itself on “Bypasses every major detector”, and its FAQ answers “Does it bypass Turnitin AI detection?” with a flat “Yes.”
- Across a crawl of all 192 pages in its sitemap, Pangram appears zero times. Turnitin appears 663 times.
- The domain was registered 4 October 2025. Its Terms are dated 1 January 2025 and 21 blog posts carry schema dates back to 28 November 2024.
- It publishes two contradictory sets of bypass rates for the same detectors — GPTZero at 96% and 97%, Copyleaks at 97% and 93% — with no method behind either.
- The homepage’s own stat block ships in the HTML as “0+ Happy Users, 0% Success Rate, 0 Detectors Supported.”
- Refunds: “all purchases are non-refundable”, with an explicit waiver of the 14-day withdrawal right — while its own blog warns readers off “services without money-back guarantees”.
- No legal entity, no registration number, no address. Governing law is “the jurisdiction in which TextPolish operates”, which is never named. Contact is a Gmail address.
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 Pangram?
No evidence says yes, and the vendor has never claimed it. That claim is worth pinning precisely, because Text Polish is not a quiet tool. Its homepage sells “Bypasses every major detector”, and its FAQ asks and answers the question directly: “Does it bypass Turnitin AI detection? Yes.”
We crawled every URL in its sitemap — 192 pages — and string-matched each one. Turnitin appears 663 times across 192 pages. GPTZero 620. Copyleaks 532. Pangram: zero, on zero pages.
“Every major detector” is doing a lot of work in a sentence written by a site that has never acknowledged this one exists. This page reports no first-party test; everything below was fetched on 28 August 2026.
Which bypass numbers does it publish?
Two sets, and they disagree with each other about the same detectors.
| Detector | Figure on one blog page | Figure on another | Method published |
|---|---|---|---|
| GPTZero | 96% | 97% | None |
| Turnitin | 94% | 95% | None |
| Copyleaks | 97% | 93% | None |
| ZeroGPT | 95% | 96% | None |
| Pangram | — | — | never mentioned |
Neither page carries a method. And the homepage’s own headline statistics are stranger still: the counters ship in the delivered HTML as “0+ Happy Users”, “0% Success Rate” and “0 Detectors Supported”, animated client-side with no server-rendered value behind them. There is no published success rate in the page source at all. The “AI DETECTED (94%)” panel next to them is hardcoded HTML in a mock interface, not a captured detector output.
Why do the dates not add up?
The domain text-polish.com was registered on 4 October 2025. Set that against what the site says about itself.
The Terms of Service and Privacy Policy are both stamped 1 January 2025 — nine months before the domain existed. Twenty-one blog posts carry datePublished schema predating registration, the earliest 28 November 2024, more than ten months early. Those are not incidental pages: they include the bypass-detectors guide that publishes the accuracy figures, and the post headed “Why TextPolish Leads the Industry in 2025” that carries the second set of numbers.
What do the Terms actually promise?
Nothing about detection. The Terms of Service disclaimer is generic boilerplate: the services are provided “AS IS”, use is “at your sole risk”, and the vendor does not warrant that content will be accurate or complete. It never mentions AI detectors. That is the finding — the marketing makes a specific product promise while the contract retreats to a general shield that does not cover it.
On refunds the Terms are unusually blunt: “All purchases made through our Services are non-refundable… Refund of credits or payments will not be provided under any circumstances”, plus an explicit statement that you “waive your right to the standard 14-day withdrawal period.” Cancel and “you will lose any remaining usage credits or allotted usage immediately” — while the FAQ promises you “keep access until the end of your billing period.” Both cannot be true.
Has Pangram ever named it?
No. Text Polish is absent from the DAMAGE paper‘s 19 named tools and from the August 2025 benchmark‘s 20 scored tools. The Pangram 4 technical report from July 2026 tests 13 commercial humanizers and anonymises every one as “Commercial A” through “Commercial M”, chosen by web traffic in early 2026 — so presence there can be neither confirmed nor excluded.
datePublished from all 156 blog pages and cross-referenced them against the RDAP registration record for the domain. We requested each image asset directly and recorded its HTTP status. We quote the homepage, pricing, FAQ, terms and privacy pages verbatim, and searched Reddit for any mention of the tool. Where a source has a commercial stake, we say so.Where is the proof, and where are the users?
There are no detector screenshots on the site. The only images are six greyscale detector logos, rendered at 40% opacity — Turnitin, Copyleaks, ZeroGPT, QuillBot, Grammarly, GPTZero. All six files load; none is evidence. Originality.ai, Winston AI and Sapling are named in the copy with neither a logo nor a result. Pangram has neither, because it is never named.
The community footprint is empty in a way that is hard to fake in either direction. Searching Reddit posts and comments for Text Polish and text-polish.com returns zero mentions, ever. This is a niche where competing tools run visibly seeded threads; here even the seeding is absent. The vendor’s own blog calls its Chrome extension the market leader and gives install instructions; we could not find that extension on the Chrome Web Store. The affiliate page returns a 404.
What should you take from this?
The honest answer to the title question is that nobody has tested it, and Text Polish has arranged its marketing so that the question never comes up. It claims to beat every major detector while never naming the detector that publishes the hardest numbers, and it dates its evidence to before its own domain existed.
What the record does support is Pangram’s published ceiling: across 13 unnamed commercial humanizers it reports catching humanized text 97.69% of the time, with the best evader still caught 92.78% of the time. Nothing on text-polish.com gives a reason to expect an exception.
If the real worry is being wrongly flagged on your own writing, keep your drafting history — it costs nothing and beats any rewriting tool as a defence, and 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. For the Pangram cells with real evidence, see Undetectable AI, StealthGPT, QuillBot, StealthWriter, Phrasly, Humbot, Rephrasy, Ryne AI, Clarity Bubble, SuperHumanizer and UnAIMyText, plus the Pangram review.
The one tool that got past both
Pangram and Turnitin catch most humanizers — one or the other, and usually the one you did not think to test. StealthWriter is the only tool that bypassed both in 2026. Run Ghost 5.2 Pro at level 7–8, section by section, and re-check before you submit.
Free plan to start, no card. Or get the free prompt pack first →
Affiliate link — I earn a commission if you upgrade.
Frequently asked questions
Does Text Polish bypass Pangram?
There is no evidence either way, and Text Polish has never claimed it. We crawled all 192 URLs in its sitemap and the word “Pangram” does not appear once. Text Polish is likewise absent from every Pangram publication. No third party has tested the pair.
But doesn’t it say it bypasses every major detector?
It does, on the homepage. The same site’s FAQ then says “no tool can guarantee 100%” — on the same page as “Does it bypass Turnitin AI detection? Yes.” And the Terms retreat to a generic AS IS disclaimer that never mentions detection at all. The specific promise lives in the marketing; the contract only offers a general one.
What is wrong with the dates on the site?
The domain text-polish.com was registered on 4 October 2025. The Terms of Service and Privacy Policy are both dated 1 January 2025, nine months earlier. And 21 blog posts carry datePublished schema predating the domain, the earliest 28 November 2024 — including the very pages that publish the bypass percentages. The evidence base is dated to before the company existed.
Which bypass numbers are real?
Neither set can be checked, and they disagree with each other. One vendor blog page gives GPTZero 96%, Turnitin 94%, Copyleaks 97%, ZeroGPT 95%. Another gives GPTZero 97%, Turnitin 95%, Copyleaks 93%, ZeroGPT 96%. No sample size, corpus, model or test date accompanies either. The homepage’s own animated counters ship as literal zeros in the delivered HTML.
Is Text Polish in Pangram’s published research?
No. It is absent from the DAMAGE paper’s 19 named tools and from Pangram’s August 2025 benchmark of 20. The July 2026 Pangram 4 report tests 13 humanizers anonymised as “Commercial A” to “Commercial M”, so presence there cannot be confirmed or excluded. The relationship is mutual silence.
What does it cost and can I get a refund?
Free gives 250 words total. Basic is $2.99 a month on annual billing, Pro $6.99, Ultra $9.99, with 1,000, 2,000 and 5,000 words per request. The pricing FAQ contradicts the pricing card, quoting Basic at $4.99. And there are no refunds: the Terms say all purchases are non-refundable “under any circumstances” and that you waive the standard 14-day withdrawal period.
Does it have any users?
None that we could find talking about it. Reddit returns zero mentions of Text Polish or text-polish.com, ever — in a niche where competitors’ threads are heavily seeded. The site claims 100,000+ users, and its blog calls its own Chrome extension market-leading; we could not locate that extension on the Chrome Web Store.
Who operates it?
Unstated. There is no legal entity name, no registration number, no address and no country of incorporation anywhere on 192 pages. The only contact is a Gmail address. The governing-law clause reads “the laws of the jurisdiction in which TextPolish operates” — a clause that identifies nothing. The domain registration is redacted behind Namecheap.
