Does Phrasly Bypass GPTZero? Its Changelog Says Not Reliably (2026)

Does Phrasly bypass GPTZero? Its own engineering changelog records GPTZero catching it three separate times, and no pass rate at all.

By Vlad Ivanov — publisher of Detection Drama, tracking AI-detector and humanizer behaviour across 245 published tests and teardowns. Last updated: September 2026

Does Phrasly bypass GPTZero?

For once this can be answered from records rather than claims, because Phrasly keeps something almost nobody else in this market does: a real versioned engineering changelog, fifteen dated entries running back to August 2023.

That changelog mentions GPTZero three times. Every one of them is a repair.

This page reports no first-party test. Everything below was fetched on 21 September 2026 and quoted verbatim.

What does the changelog actually record?

Three dated entries, each logging GPTZero’s detection catching Phrasly’s output and Phrasly adapting afterwards.

Source: phrasly.ai/changelog, all fifteen entries read 21 September 2026. No sample size or numerical result accompanies any GPTZero entry.

None of these is a marketing document. A changelog entry saying you resolved a regression is written for engineering reasons, by people recording that something broke. That is what makes it worth more than a guarantee.

Why is the missing review the loudest part?

Because of which detector it is missing for.

Phrasly published a dated compatibility review for Turnitin on 27 August 2025, timed to Turnitin’s own product update and claiming 100 documents, all returning a 0% AI score. We examined that claim in does Phrasly bypass Turnitin. Its central problem is access: Turnitin sells no individual licences and its AI report is instructor-facing, so no vendor can lawfully produce a hundred such reports.

GPTZero is the exact opposite. Free, instant, no account, up to 10,000 characters a scan. It is the one major detector where a hundred-document run is genuinely achievable by anybody, including Phrasly.

So the company ran a large numbered study on the detector it could not possibly have tested at that scale, and published no study at all on the detector it could.

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.

What happened when GPTZero tested it?

On 23 December 2024, GPTZero published its own hands-on review of Phrasly. It generated a ChatGPT essay introduction about the Roman Empire, ran it through Phrasly’s aggressive humanizer setting, which produced five different rewrites, and scanned the results.

The original ChatGPT text scored 100% confidence as AI. After Phrasly, GPTZero flagged the rewritten paragraph as likely AI and specifically labelled it “AI paraphrased” — the Paraphraser Shield firing — with inconsistent results across the five drafts. Its conclusion was that “both the Phrasly AI Humanizer and Phrasly’s AI Detector are inconsistent and misleading when it comes to the outputs and promises of their product.”

Has Phrasly changed what it claims?

Yes, and the shift is worth recording accurately rather than mocking.

Its current homepage FAQ reads: “The result is better writing that also scores well on platforms like Turnitin and GPTZero, not because it’s designed to trick them, but because quality writing naturally resembles human writing.” It adds that users should treat Phrasly as “a refinement tool, not a substitute for their own thinking” and “follow your institution’s policies.”

Its Terms disclaim the outcome in capitals: “WE CANNOT GUARANTEE THAT THE TEXT GENERATED BY OUR SERVICES WILL ALWAYS PASS SUCH DETECTORS.” Section 31 states it “does not support using our technology to circumvent educational AI detection systems.” Refunds require “absolutely no usage…whatsoever” — zero documents processed across the account’s entire history.

A quality-tool framing is a narrower and more defensible claim than a bypass guarantee. It also sits awkwardly beside a changelog whose entries exist precisely because the output stopped passing detectors.

Has any detector company ranked it?

Phrasly is not in the appendix table of GPTZero’s February 2026 paper, which names nine bypass services individually, though GPTZero has written about it separately. It is absent from the nineteen tools in Pangram’s DAMAGE paper and from the twenty scored in Pangram’s August 2025 benchmark. We cover the Pangram side in does Phrasly 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 should you take from this?

Phrasly is the most transparent company in this cluster. It keeps a dated versioned changelog while its competitors publish backdated blog posts, undated guarantees and screenshots with no model version on them. That practice deserves saying out loud.

It is also the practice that produced the clearest published evidence against its own product. Three dated entries record GPTZero catching Phrasly and Phrasly patching afterwards, the detector moving first each time. There is no GPTZero pass rate anywhere, on the one detector where producing one would have been easy. And when GPTZero ran its own test, it found the output flagged as AI-paraphrased.

Transparency and a good result are not the same thing. Phrasly has the first. On the evidence it published itself, it does not reliably have the second — and you can check your own text at gptzero.me, free, no account, in about a minute.

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 Phrasly review, and the same question against Turnitin and Pangram. Also in this cluster: StealthWriter, Rephrasy, WriteHuman.

Frequently asked questions