ZeroGPT vs Pangram: Which AI Detector Was More Accurate in Our 160-Sample Test?
AI Detectors

ZeroGPT vs Pangram: Which AI Detector Was More Accurate in Our 160-Sample Test?

Shadab Sayeed
Written by Shadab Sayeed
September 28, 2026
Calculating…

AI detectors are everywhere now, especially around schools and universities. If you are a student, you may have already pasted an essay into a detector just to see whether it thinks your writing looks “AI-generated.” The problem is that two detectors can look at the same text and give very different answers.

To see how large that difference can be, we tested ZeroGPT and Pangram on the same set of 160 samples: 82 were AI-written and 78 were human-written. Both datasets recorded an AI score from 0 to 100, where a higher score means the detector believes the text is more likely to be AI-written. In the recorded results, a score of 50 or above always matched an “AI” verdict, so we used the tools' recorded verdicts when calculating accuracy.

The short answer

In this test, Pangram was substantially more accurate. It classified 149 of 160 samples correctly (93.1%), while ZeroGPT classified 118 correctly (73.8%).

Bar chart comparing overall accuracy of Pangram and ZeroGPT
Overall classification accuracy on the 160-sample dataset.

The difference becomes more important when you look at human-written work. Pangram correctly recognized 77 of 78 human samples (98.7%). ZeroGPT correctly recognized 62 of 78 (79.5%). In other words, ZeroGPT incorrectly flagged 16 genuinely human samples as AI, compared with only one for Pangram.

Also Read: Stealthwriter vs ZeroGPT: I Tested 100 Rewrites, and the Results Were Complicated!

Bar chart comparing AI-text and human-text accuracy for Pangram and ZeroGPT
Pangram performed better on both AI-written and human-written samples.

Why false positives matter so much for students

A false positive is a human-written piece that a detector wrongly labels as AI. For students, this is probably the most worrying type of error. You can write something yourself and still receive a high AI score.

In our test, Pangram produced 1 false positive. ZeroGPT produced 16. ZeroGPT also missed more AI-written text: it labeled 26 AI samples as human, compared with 10 for Pangram. That second mistake is called a false negative.

Bar chart comparing false positives and false negatives from Pangram and ZeroGPT
False positive means human text was flagged as AI; false negative means AI text was missed.

Looking at the paired results gives another useful picture. Both tools were correct on 113 samples. Pangram was correct while ZeroGPT was wrong on 36 samples. ZeroGPT was correct while Pangram was wrong on only 5. Both tools were wrong on 6 samples.

Also Read: [STUDY] ZeroGPT vs Winston AI: Which AI Detector Performs Better? 160-Sample Test

The scores themselves behave very differently

Another striking difference was how the tools used the 0–100 scale. Pangram was extremely decisive: 158 of its 160 scores were either exactly 0 or exactly 100. Only two results landed somewhere in between.

ZeroGPT was much more gradual. It returned 85 intermediate scores between 0 and 100. That can make the score look more nuanced, but a more detailed-looking number does not automatically mean a more accurate result. In this dataset, ZeroGPT's extra score variation came with more classification mistakes.

Also Read: [STUDY] Turnitin vs. ZeroGPT: How do they compare?

Bar chart showing how often Pangram and ZeroGPT returned zero, intermediate, or 100 scores
Pangram's scoring was mostly all-or-nothing, while ZeroGPT used intermediate percentages much more often.

The average AI score also showed stronger separation for Pangram. Real AI samples averaged 87.8 with Pangram, while human samples averaged just 1.4. With ZeroGPT, AI samples averaged 72.3, but human samples still averaged 30.3. A detector is easier to interpret when its scores for human and AI writing stay farther apart.

What the interfaces are like

Pangram's interface is more report-like. The screenshots show an overview panel, a clear Human Written or AI Generated verdict, a percentage gauge, AI highlighting, and separate Details and Evidence areas. That makes it easier to inspect why a piece received a particular result instead of seeing only a percentage.

Pangram human-written result
Pangram example: a human-written verdict.
Pangram AI-generated result
Pangram example: an AI-generated verdict with highlighted text.
Pangram AI-generated JPMorgan sample
Another Pangram AI result.
Pangram human-written Scrum sample
Another Pangram human-written result.
Pangram AI-generated hospital sample
Pangram marking a hospital-related sample as AI-generated.

ZeroGPT is simpler. You paste text into a large box, run the detector, and receive a percentage with a short interpretation such as “likely human written” or “mixed signals.” For quick checks this is straightforward, although the test results show that the percentage should not be treated as a precise measurement of how much of a document was written by AI.

Also Read: Which is Better GPTZero or ZeroGPT? An In-Depth Comparison!

ZeroGPT result showing 20.8 percent AI
ZeroGPT example with a relatively low AI score.
ZeroGPT result showing 100 percent AI
ZeroGPT example showing a 100% AI score.
ZeroGPT result showing 46.9 percent AI
An example of ZeroGPT's more gradual scoring.
ZeroGPT result showing 59.6 percent AI
A score just above the AI classification threshold in the dataset.
ZeroGPT result showing 44.1 percent AI
A mid-range ZeroGPT score.

So, which one performed better?

On these 160 samples, the numbers clearly favored Pangram. Its overall accuracy was about 19 percentage points higher, and the biggest practical difference for students was its much lower false-positive rate. ZeroGPT gave more finely graded percentages, but it also had much more overlap between scores for real human writing and AI writing.

That does not mean any AI detector can prove who wrote a piece of text. This was one dataset, detector models can be updated, and performance can change with writing style, text length, editing, subject area, or newer AI models. A detector result is best treated as one signal, not a final verdict.

About the Author
Shadab Sayeed

Shadab Sayeed

CEO & Founder · DecEptioner
Dev Background
Writer Craft
CEO Position
View Full Profile

Shadab is the CEO of DecEptioner — a developer, programmer, and seasoned content writer all at once. His path into the online world began as a freelancer, but everything changed when a close friend received an 'F' for a paper he'd spent weeks writing by hand — his professor convinced it was AI-generated.

Refusing to accept that, Shadab investigated and found even archived Wikipedia and New York Times articles were being flagged as "AI-written" by popular detectors. That settled it. After months of building, DecEptioner launched — a tool built to defend writers who've been wrongly accused. Today he spends his days improving the platform, his nights writing for clients, still driven by that same moment.

Developer Content Writer Entrepreneur Anti-AI-Detection