Does Clarity Bubble Bypass GPTZero? Read Its Own Screenshot (2026)

Does Clarity Bubble bypass GPTZero? It publishes one screenshot to prove it — and GPTZero is printing a warning on that screenshot saying the result is out of date.

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

Does Clarity Bubble bypass GPTZero?

Clarity Bubble is a useful case because, unlike most of this market, it does put an exhibit in front of you. There is a screenshot. It loads. It is real.

It also does not say what it is captioned to say, and reading it carefully takes about thirty seconds.

This page reports no first-party test of Clarity Bubble’s output. Everything below was fetched from claritybubble.com and from GPTZero directly on 3 September 2026, and every vendor claim is quoted verbatim.

Why GPTZero, and not the detector at the checkout?

We pulled Clarity Bubble’s sitemap. It contains 10,136 URLs, of which 10,126 sit under /tools/ — templated landing pages that differ from one another only in the keyword in the slug.

696 of them end in -vs-gptzero. Counting every URL that names a detector at all:

Source: claritybubble.com/sitemap.xml, fetched and counted 3 September 2026.

Set that against the pricing bullet, which names exactly two detectors: “Bypass all AI detectors (incl. Turnitin & GPTZero).” One of those two carries 696 landing pages and a screenshot that loads. The other has no pages at all, and its proof image — /detectors/turnitin.png — returns a 404, as it did when we checked for the Pangram version of this question six days earlier.

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 does the screenshot actually show?

We downloaded claritybubble.com/detectors/gptzero.png directly — 1,915 by 841 pixels, HTTP 200 — and inspected it at full resolution rather than trusting the caption.

It is authentic. A fabricated or recycled GPTZero image fails specific tests, and this one passes all of them. It shows the current three-class result — AI 0% / Mixed 0% / Human 100% — the verdict line “We are highly confident this text is entirely human”, a model version stamp reading Model 4.4b, the Basic Scan header, the AI Sentences and AI Words view toggles, and a signed-in account showing “720 credits of 10K remaining”. The document in the left pane is a generic essay titled The Impact of Artificial Intelligence on Daily Life, counted at 4,392 characters, 657 words.

So far this is better evidence than most of this market offers. Then you look at the bottom of the same image.

The consequence is narrow and it is decisive. The 100% Human verdict on the right cannot be attributed to the 657-word document on the left. One of them is out of date and the screenshot does not tell you which.

We are not claiming the image was faked — it plainly was not, and we checked in order to be able to say so. We are claiming that the single exhibit a company built 696 landing pages around does not verify on its own terms, and that nobody appears to have looked.

Is the wider proof wall a controlled test?

No, and it does not claim to be. Nine detector screenshots are served from /detectors/. Seven return 200 — GPTZero, ZeroGPT, Copyleaks, Originality, QuillBot, Winston and Smodin — and two return 404: turnitin.png and writer.png. The seven that load use two different source texts, 4,392 and 5,665 characters, so they are not the same document scanned across detectors.

One of them undercuts a headline claim on its own. The Originality.ai image reads “Model Lite 1.0.2 — 97% Confident That’s Original” — the weaker of Originality’s models, and 97%, not the “100% Human Score” the site advertises elsewhere.

No date, sample size, prompt, source model or scan mode is published for any of them.

What does the company promise in writing?

The marketing line is “The Only Humanizer That Passes All Detectors.” The contract says the opposite, in capitals, and it has not changed since we last checked:

“WE DO NOT WARRANT THAT OUTPUT WILL PASS ANY THIRD-PARTY AI DETECTOR OR MEET ANY PARTICULAR ACADEMIC OR PROFESSIONAL STANDARD.”Clarity Bubble Terms of Service, last updated 11 May 2026, read 3 September 2026

The homepage FAQ agrees with the lawyers rather than the banner: “We cannot guarantee that any output will bypass third-party AI detection tools.”

Refunds are seven days and conditional — a request is allowed “if you believe a charge was made in error or the Service was not usable as advertised”, reviewed “in good faith.” Since the Terms disclaim passing any detector, being detected is unlikely to qualify as not usable as advertised. Pricing is $5.99 a month or $35.88 a year for Basic, $19.99 or $119.88 for Pro, $39.99 or $239.88 for Ultra, with 300 free credits and billing via Polar.

There is still no legal entity, jurisdiction or postal address published anywhere on the site — only a Gmail address — and the footer describes the service as “SOC2-ready” while another page calls it “SOC2-compliant”, with no attestation published either way.

Have the detector companies said anything?

Nothing about Clarity Bubble. It 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 scored in Pangram’s August 2025 benchmark. Pangram’s July 2026 paper anonymises thirteen humanizers as Commercial A to M, so its presence there can be neither confirmed nor excluded.

How much should you trust either side’s numbers?

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?

Clarity Bubble picked the one detector anyone can verify for free, built 696 pages against it, and published a screenshot that is genuinely real. That is more than most of this market does, and the parts of it that are true deserve saying.

But the screenshot carries GPTZero’s own notice that the result is stale, the detector named beside GPTZero at checkout has zero pages and a missing image, the proof wall uses two different source texts with no dates, and the contract disclaims the entire proposition in capital letters.

The honest answer to the question in the title is that nobody knows, including Clarity Bubble, which is why its lawyers wrote what they wrote. You can find out for your own text in about a minute, free, at gptzero.me — and unlike the exhibit above, you will know the result belongs to the words you scanned.

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 Clarity Bubble review, and the same question against Pangram. Also in this cluster: StealthGPT, GPTHuman, Undetectable AI.

Frequently asked questions