Blog AI Humanizer

ZeroGPT Humanizer Review: 47 of 50 Pangram Tests Failed

ZeroGPT’s humanizer is a poor choice for bypassing Pangram, GPTZero and Originality.ai in our November 2026 benchmark. Across 50 matched inputs, it failed 47 Pangram tests, 35 GPTZero tests and all 50 Originality.ai 5.0 tests. The problems went beyond detection: selected rewrites changed a person’s name, turned a qualified claim into an absolute statement, and repeated the same explanation in consecutive paragraphs.

We tested ZeroGPT’s first-party Business API humanizer: v11sr, University readability, General Writing and More Human strength. This review covers that configuration. StealthGPT Super processed the same 50 inputs and delivered substantially better results on those three detectors, alongside higher averages on all eight writing-quality measures. The complete results are in HumanizerEval’s November 2026 benchmark.

If you’re looking for a ZeroGPT Humanizer alternative, the evidence gives us a clear preference: StealthGPT Super. Below are the full five-detector results, the quality comparison and three matching writing examples you can inspect yourself.

ZeroGPT Humanizer review: 3 of 50 Pangram passes, 0 of 50 Originality.ai passes and 15 of 50 GPTZero passes.

Tested November 17, 2026. Review published November 18, 2026. All counts below use the same 50 complete outputs per humanizer.

ZeroGPT failed 47 of 50 Pangram tests

For a product sold as an AI humanizer, three Pangram passes out of 50 is a weak result. That is a 6% strict-pass rate, leaving 94% of the tested outputs outside Pangram’s human category. StealthGPT Super passed 44 of the same 50 tests: 88%, or 41 additional passing texts.

The pattern continued across the other two heavily weighted detectors. ZeroGPT Humanizer passed 15/50 on GPTZero and 0/50 on Originality.ai 5.0. Super passed 30/50 and 19/50, respectively. ZeroGPT’s rewrite therefore left users with a particularly poor result on Originality: every complete output failed the benchmark’s human-verdict requirement. See the ZeroGPT Humanizer benchmark profile for the published results.

AI detectorZeroGPT Humanizer passesZeroGPT Humanizer failuresStealthGPT Super passes
Pangram 4.13/50 (6%)47/50 (94%)44/50 (88%)
GPTZero15/50 (30%)35/50 (70%)30/50 (60%)
Originality.ai 5.00/50 (0%)50/50 (100%)19/50 (38%)
Winston AI 5.250/50 (100%)0/50 (0%)48/50 (96%)
ZeroGPT detector40/50 (80%)10/50 (20%)34/50 (68%)

These are strict-pass rates, not average AI-probability scores. A mixed or uncertain verdict fails the strict benchmark rule; it does not automatically mean a detector rated that text 100% AI. The denominator stays at 50 for every row.

Complete five-detector comparison of ZeroGPT Humanizer and StealthGPT Super across 50 matched inputs.

Strict passes across the full November test set. Source: HumanizerEval, November 2026. Open the full-size comparison.

For more detailed examples involving the detectors where the gap was largest, explore StealthGPT vs Pangram, GPTZero humanizer results and Originality.ai humanizer results.

Passing ZeroGPT’s own checker did not carry over

ZeroGPT Humanizer did better on two detectors: 50/50 passes on Winston AI and 40/50 on ZeroGPT’s own checker. Those results exceeded Super’s 48 and 34 passes on those detectors. They also show why checking a rewrite in one familiar tool gives an incomplete picture.

The same collection of ZeroGPT rewrites that earned an 80% pass rate on ZeroGPT earned just 6% on Pangram and 0% on Originality.ai. An encouraging result in ZeroGPT’s detector did not establish that the writing would pass other systems. That disagreement is a practical weakness for users who need their humanized output to work across the named detectors.

ZeroGPT Humanizer rewrites text. ZeroGPT’s detector checks text. GPTZero is a separate detector from a different company. This review compares rewrites from the first product across five checkers. Our ZeroGPT detector guide covers the checker separately.

StealthGPT Super also produced stronger writing-quality scores

A humanizer that changes important details or pads a paragraph creates more editing work. The November benchmark evaluated eight writing-quality dimensions as well as detection. Super scored higher on every quality average, with a combined quality score of 73.48/100 versus ZeroGPT’s 54.94/100. The gap was especially pronounced in factual consistency: 78 versus 26.

Benchmark measureZeroGPT HumanizerStealthGPT Super
Weighted strict bypass20.40%69.00%
Writing-quality score /10054.9473.48
Overall score /10030.7670.34
Overall position among 15 services#7#2

The weighted bypass score gives Pangram 40%, GPTZero 30%, Originality.ai 20%, and Winston AI and ZeroGPT 5% each. The overall score combines 70% weighted bypass with 30% quality. Those published weights explain why Super’s much stronger performance on Pangram, GPTZero and Originality carried more weight than ZeroGPT’s lead on the other two detectors.

Quality measure /100ZeroGPT HumanizerStealthGPT Super
Readability46.068.8
Meaning51.667.2
Structure58.462.0
Factual consistency26.078.0
Completeness83.294.0
Grammar and syntax92.0100.0
Hallucination control2.032.0
Stance preservation82.894.8
StealthGPT Super scored higher than ZeroGPT Humanizer on all eight automated writing-quality averages.

Quality scores are automated, provisional scores out of 100; higher is better. They are not percentages of error-free outputs. Source: HumanizerEval’s November cycle. Open the full-size quality chart.

The text examples make the comparison more tangible. We selected these three passages to show visible editorial problems, and include excerpts from the original draft and both rewrites. Detector results apply to the complete saved outputs, which are linked with each example.

Use the linked full-text reports to compare the complete source and both rewrites side by side. Each report also includes the saved detector results, so you can check whether the highlighted sentence reflects the broader output and see exactly which checkers passed or failed.

Example 1: ZeroGPT changed the bookstore owner’s name

The original fictional business case identifies the owner of Elmwood Bookery as Clara Vance. ZeroGPT’s rewrite introduces Sara while discussing her online storefront. A humanizer has no editorial reason to change the owner’s identity. This is the kind of avoidable error that makes a supposedly finished rewrite require another careful review.

Selected bookstore-case excerpts: the original names Clara Vance, ZeroGPT Business API introduces Sara, and StealthGPT Super retains Clara Vance.

Selected verbatim excerpts from the same fictional bookstore case. ZeroGPT introduces Sara; Super retains Clara Vance. Open the full-size text comparison.

Original AI draft — excerpt

The second option involves joining an established platform tailored specifically for independent bookstores. Under this model, the platform manages the digital storefront, payment processing, and wholesale fulfillment directly to the customer, paying the bookstore a commission on each sale. This route requires minimal upfront investment and eliminates the burden of shipping logistics. However, the profit margins per book are noticeably smaller than in-store sales, and the store would have limited ability to sell its popular sidelines, such as locally made stationery, literary totes, and artisanal teas, which carry higher markups than standard books.

ZeroGPT Humanizer — excerpt

By becoming part of a platform which is specifically made for independent bookstores such as the Elmwood Bookery, Sara could potentially develop a store front for her online customers and thus remove the main part of the cost of a web site (the shipping and packing of products), for a commission on the sales of books, online. This sort of site would also need to restrict the sorts of products which would be available online in order to ensure an acceptable margin of profit, for example not be able to sell locally made stationery, totes etc that have a greater margin than books.

StealthGPT Super — excerpt

Build a website from scratch and fulfill orders locally using a back room. Although this option would give Elmwood Bookery great flexibility in terms of look and feel of the website and the customer experience, web design costs would be significant and Clara’s small staff would have to handle order fulfillment. Become part of an online marketplace for independent bookstores. This option would entail little startup cost as orders would be fulfilled by the online marketplace (using a wholesaler) and Elmwood Bookery would pay a fee on each transaction. However, Elmwood Bookery would earn less per book sold and would be limited in its ability to sell other high-margin items carried in the store (locally produced tote bags, teas, stationery, etc.).

The original identifies the owner with “Owned and operated by Clara Vance”; ZeroGPT writes “Sara could potentially develop a store front.” Super retains Clara, including in the excerpt above, and lays out the storefront choices more directly. Super’s complete business rewrite passed all five detectors. ZeroGPT’s failed Pangram, GPTZero and Originality.ai.

DetectorZeroGPT Humanizer complete outputStealthGPT Super complete output
PangramAI — failHuman — pass
GPTZeroai — failhuman — pass
Originality.aiai — failhuman — pass
Winston AI100/100 human score — pass100/100 human score — pass
ZeroGPTYour Text is Human Written — passYour Text is Human Written — pass

The automated quality score for this complete output was 43/100 for ZeroGPT versus 80/100 for Super. Read the full original, both complete rewrites and saved detector results.

Example 2: “Generally” became “always,” with a new typo

The home-insulation essay demonstrates a different failure: ZeroGPT changes the strength of a claim. The original says a home that loses less heat “generally needs less energy.” ZeroGPT changes that to “always needs less energy” and introduces the misspelling “volatilility.” Neither change improves fidelity to the draft.

Original AI draft — excerpt

Insulation can also make household finances more resilient. Energy prices fluctuate, but a home that loses less heat generally needs less energy to maintain a comfortable temperature. This is especially important for lower-income households, which may face a painful choice between adequate heating and other necessities. Some residents will use improved efficiency to keep previously underheated rooms warmer rather than to reduce consumption as much as expected. That does not make the investment a failure: greater comfort and safer indoor conditions are valuable outcomes too.

ZeroGPT Humanizer — excerpt

In addition to increasing comfort and safety in homes, energy-efficient insulation can help make household finances more resilient to energy price volatilility. While energy prices may go up and down, a well-insulated home always needs less energy to maintain the same temperature. For lower-income households, who already live from paycheck to paycheck and have to make tough choices between essentials, improved home energy efficiency can help them afford to keep previously underheated parts of their homes at a healthy temperature. Even if the efficiency measure does not achieve as much reduction in energy consumption as expected, there are plenty of other positive outcomes to achieve through such an investment.

StealthGPT Super — excerpt

Insulation can protect householders against rises in energy prices and help make it easier for them to afford to heat (or cool) their home. This could be particularly helpful for lower income households, who may currently be forced to choose between heating their home and spending money on other essentials. Subsidies could nudge more households towards insulating, but they’d need to be designed to get around the following barriers and challenges: Households may struggle to afford the up-front cost of insulation.

A qualification such as “generally” tells readers how strongly an author is making a point. Replacing it with “always” makes the statement more sweeping. Super keeps the discussion conditional and retains the affordability issue without that new absolute. These passages are synthetic test material; the comparison concerns how each humanizer handles the supplied argument.

DetectorZeroGPT Humanizer complete outputStealthGPT Super complete output
PangramMixed — failHuman — pass
GPTZeroai — failhuman — pass
Originality.aiai — failai — fail
Winston AI100/100 human score — pass100/100 human score — pass
ZeroGPTYour Text is Human Written — passYour Text is Human Written — pass

Super passed Pangram and GPTZero on the complete essay while ZeroGPT failed both. Both rewrites failed Originality.ai 5.0. Super’s quality score was 76/100, compared with ZeroGPT’s 45/100. Inspect the complete insulation-essay comparison.

Example 3: ZeroGPT repeated the same science explanation twice

The antibiotic-resistance article shows how a rewrite can become longer without becoming more useful. The source explains resistance persistence once. ZeroGPT expands that material into two adjacent paragraphs that repeat mobile elements, the cost of carrying resistance genes, and biofilms and spores. Readers have to work through essentially the same explanation again.

Original AI draft — excerpt

Resistance can persist even after antibiotic use declines. Resistance genes often sit on mobile elements that carry little cost, and bacteria can acquire additional mutations that make carrying them less burdensome. Some bacteria form biofilms or spores that protect them from antibiotics and cleaning. So a population may remain resistant long after the original pressure fades. In short, antibiotic resistance spreads because it is an evolutionary process with a highly efficient transmission system. Antibiotics select for resistant bacteria; bacterial sex and viral accidents move resistance genes; human crowding, medical care, agriculture, poor sanitation, and global travel give those genes endless routes.

ZeroGPT Humanizer — excerpt

However, even after the pressure from the antibiotics has decreased, resistance can persist in a population. Firstly, mobile elements which carry resistance genes often have little disadvantage to a bacteria. However, subsequent mutations can make a difference to a bacteria’s survival and as these occur, the more advantageous the bacteria with resistance genes become. Secondly, bacteria can form biofilms and become spores. Both of these forms are highly resistant to antibiotics as well as being resistant to cleaning surfaces. A population can remain resistant long after the selective pressure has decreased. In most cases, resistance genes reside on mobile elements that have minimal costs to the cell. Additionally, as bacteria acquire mutations it becomes less expensive for them to carry resistance genes. Biofilms and spores are also resistant to antibiotics as well as to most cleaning agents.

StealthGPT Super — excerpt

Other factors that can promote the spread of antibiotic-resistant bacteria include invasive medical procedures (surgery, intubation, catheterization, etc.) that allow bacteria to gain a foothold by breaching the body’s natural barriers, and infection control lapses. And bacteria that form protective spores or clumpy biofilms are tough to eliminate with antibiotics or disinfectants. Since bacteria can adapt to lessen the fitness cost of harboring resistance genes (which often reside on relatively small, easily transmissible pieces of DNA), antibiotic resistance can persist even when antibiotic use tapers off. And since people can carry antibiotic-resistant bacteria asymptomatically, they can unwittingly pass them along to others.

The source’s persistence paragraph contains 56 words. ZeroGPT turns that passage into 137 words across those two paragraphs, repeating the mechanisms instead of advancing the article. Its complete output grew from 641 to 729 words. Super’s complete rewrite was 614 words and earned a substantially higher automated quality score: 85/100 versus 44/100.

DetectorZeroGPT Humanizer complete outputStealthGPT Super complete output
PangramMixed — failHuman — pass
GPTZeroai — failhuman — pass
Originality.aiai — failai — fail
Winston AI100/100 human score — pass100/100 human score — pass
ZeroGPTYour Text is Human Written — passYour Text is Human Written — pass

Super again passed Pangram and GPTZero; ZeroGPT failed both. Originality.ai rejected both complete outputs. The direct reading comparison is still decisive: repeating adjacent explanations is poor editing, regardless of what a detector says. Read the full science draft, both rewrites and all five saved results.

Why we would choose StealthGPT Super over ZeroGPT Humanizer

StealthGPT Super is the better alternative in this comparison because it combines a large detector-performance advantage with stronger writing scores. It delivered 41 more Pangram passes, 15 more GPTZero passes and 19 more Originality.ai passes across identical inputs. Its quality score was 18.54 points higher, and its overall score was 39.58 points higher.

The practical benefits also show up in the text: the bookstore owner keeps her name, the insulation discussion avoids ZeroGPT’s new absolute, and the science rewrite avoids the duplicated explanation. Those are concrete reasons to prefer Super for essays, professional content and explanatory articles. A detector pass matters much less if the rewrite also creates a name error or unnecessary repetition.

For the wider market comparison, read our full November 2026 AI humanizer benchmark. Super finished #2 overall among 15 services and #3 for quality. You can use the StealthGPT AI humanizer for your own drafts, or explore the humanizer API for publishing and application workflows.

How this ZeroGPT Humanizer review was tested

HumanizerEval’s November run used 50 identical inputs per service, 15 humanizers and five detectors, producing 750 outputs and 3,750 detector results. The run was completed on November 17, 2026. HumanizerEval is built and operated by the StealthGPT team. This review draws on its published cycle, methodology and saved matching text.

ZeroGPT requests used its direct first-party Business API endpoint, api.zerogpt.com/api/transform/humanize, with the v11sr settings stated above. See the official ZeroGPT humanizer API documentation. The results do not describe a test of the free Basic web model, and we have not mapped the API configuration to the website’s Advanced label.

Pangram 4.1, GPTZero’s dated API version 2026-11-04-base, Originality.ai 5.0 and Winston AI 5.2 were recorded alongside ZeroGPT’s detector. Strict passes require qualifying human labels; Winston uses a human score of at least 80. Mixed, AI and uncertain results fail. The complete 50-output results support the aggregate tables; the three selected examples explain specific editing differences.

ZeroGPT Humanizer FAQ

Does ZeroGPT Humanizer work?

It rewrites text, but its tested Business API configuration performed poorly on Pangram, GPTZero and Originality.ai. It passed just 3/50 Pangram tests, 15/50 GPTZero tests and 0/50 Originality.ai 5.0 tests in the November 2026 benchmark.

Can ZeroGPT Humanizer bypass Pangram?

It passed 3 of 50 Pangram 4.1 checks, a 6% strict-pass rate. The other 47 outputs failed the benchmark rule. StealthGPT Super passed 44 of the same 50 checks.

Does ZeroGPT Humanizer beat Originality.ai 5.0?

It did not pass any of the 50 Originality.ai 5.0 tests in this run. StealthGPT Super passed 19/50. That is a substantial advantage for Super, with Originality remaining a difficult detector for both products.

Does ZeroGPT Humanizer bypass GPTZero?

The tested configuration passed 15/50 GPTZero checks, or 30%, and failed 35. StealthGPT Super passed 30/50, twice as many on the same input set.

Why did ZeroGPT Humanizer pass its own detector but fail others?

Different detectors gave different verdicts on the same outputs: ZeroGPT’s checker passed 40/50, while Pangram passed 3/50 and Originality.ai passed none. The observed disagreement shows that a ZeroGPT pass cannot establish how another detector will classify the rewrite.

Is ZeroGPT the same as GPTZero?

No. ZeroGPT and GPTZero are different services. ZeroGPT supplies both the humanizer reviewed here and one of the detectors used in testing. GPTZero is a separate AI detector, and the review reports its results in its own row.

Is StealthGPT Super a better ZeroGPT Humanizer alternative?

Yes, for the combination measured here: 69% versus 20.4% weighted bypass, 73.48 versus 54.94 quality, and #2 versus #7 overall. Super led on Pangram, GPTZero, Originality.ai and all eight quality averages. ZeroGPT Humanizer led on Winston and ZeroGPT’s own detector. Try StealthGPT’s AI humanizer.

Jozef Gherman
About the author
Jozef Gherman
CEO
Jozef is the founder and CEO of StealthGPT. Over three years ago he created StealthGPT with almost no money and in that short time has grown his business into a multimillion dollar phenomenon talked about and used globally.

StealthGPT, Your AI Humanizer for the Next Draft

Humanize essays, papers, and professional writing with StealthGPT. Explore our AI humanizer and compare the published humanizer test results.