ZeroGPT Review: How NOT To Make An AI Detector

Let’s be honest — when it comes to AI detection, most of the time, you get what you pay for.

I’ve tested dozens of AI detectors over the years, and while some free tools manage to surprise me with their accuracy, others make me question if they’re even trying. ZeroGPT, unfortunately, falls into the latter category.

Now, I know what you’re thinking: “But they claim to have a 98% accuracy rate!” Well, that’s exactly what caught my attention. Because earlier this year, when we put ZeroGPT through our rigorous testing process, it barely managed to scrape together a 36.87% accuracy rate. And that’s being generous.

So, has ZeroGPT improved since our last encounter, or are they just throwing around numbers hoping nobody fact-checks them? That’s exactly what we’re going to find out today. 

What is ZeroGPT?

ZeroGPT is a free AI detection tool created using a trademarked detection model called DeepAnalyse Technology. They claim that they can detect AI-generated text from popular models like GPT, Claude, Gemini, and LLaMa with an accuracy rate of 98%.

To use it, you just need to copy and paste your input into their empty field. Once you press the “Detect Text” button, you’ll be able to see the likelihood that your input is machine-generated, along with highlights on which part of it is likely to be from AI.

Does ZeroGPT Have Any Features?

ZeroGPT’s free version only has the AI detection tool, but if you pay for their premium subscription, you can unlock some nifty features like:

• Batch File Upload
• API Access
• Multilingual Support

You can also use ZeroGPT on Telegram and WhatsApp using their bot, but if I’m being honest, I can’t imagine a scenario when this feature will be handy. 

Can ZeroGPT Detect AI Text?

Earlier in this article, I said that ZeroGPT claims to have a 98% accurate rate.

Now, that number striked me as odd since, earlier this year, we tested all the popular AI detection tools today — one of them being ZeroGPT. Out of all the detectors we tried, it has one of the lowest accuracy rates with 36.87%.

But maybe they improved their model and will prove me wrong. So, let’s test ZeroGPT.

Test #1

ZeroGPT: Wrongly detects ChatGPT text as human-generated.
AI Likelihood Score: 0%

Test #2

ZeroGPT: Wrongly detects ChatGPT text as human-generated.
AI Likelihood Score: 0%

Test #3

ZeroGPT: Wrongly detects ChatGPT text as human-generated.
AI Likelihood Score: 31.98%

Test #4

ZeroGPT: Wrongly detects ChatGPT text as human-generated.
AI Likelihood Score: 0%

Test #5

ZeroGPT: Wrongly detects ChatGPT text as human-generated.
AI Likelihood Score: 0%

Test #6

ZeroGPT: Wrongly detects ChatGPT text as human-generated.
AI Likelihood Score: 0%

Test #7

ZeroGPT: Wrongly detects ChatGPT text as human-generated.
AI Likelihood Score: 0%

Test #8

ZeroGPT: Wrongly detects ChatGPT text as human-generated.
AI Likelihood Score: 0%

Overall Tally

I didn’t expect much, but I’m still disappointed. This is much lower than our previous testing of ZeroGPT. It’s okay if an AI detector slips one through, but every single AI-generated text? What’s more is that I didn’t exactly try to make the text appear more human through prompt engineering with ChatGPT — I just wrote simple prompts like “Write an essay” or “Write a story.”

In my opinion, ZeroGPT is far too lenient to be considered reliable. Compare this to other detectors like Copyleaks and Sapling (both of which are also free) who are much more strict with AI-generated text, and you’ll see that the difference is night and day. Here’s how they did using one of the essays I used with ZeroGPT:

ZeroGPT vs. Undetectable AI: It’s Not A Contest

We’ve seen how ZeroGPT fares against traditional AI-generated content, so how about we switch into something more difficult?

Undetectable AI is an AI bypassing tool and humanizer that can paraphrase any given text into something that hides from AI detection. They’re able to do this because of a process that combines removing common AI words, adding some unpredictability to the text, and applying some human-like errors.

If you’re interested in learning more about Undetectable AI, I highly suggest our review here. We’ve also done a deep dive on their other features such as the AI Human Typer, Stealth Writer, and their own AI detector.

But for now, let’s see how ZeroGPT does against text tweaked by Undetectable AI.

Test #1

ZeroGPT: Wrongly detects Undetectable AI as human-generated.
AI Likelihood Score: 0%

Test #2

ZeroGPT: Wrongly detects Undetectable AI as human-generated.
AI Likelihood Score: 0%

Test #3

ZeroGPT: Wrongly detects Undetectable AI as human-generated.
AI Likelihood Score: 0%

Test #4

ZeroGPT: Wrongly detects Undetectable AI as human-generated.
AI Likelihood Score: 0%

Test #5

ZeroGPT: Wrongly detects Undetectable AI as human-generated.
AI Likelihood Score: 0%

Test #6

ZeroGPT: Wrongly detects Undetectable AI as human-generated.
AI Likelihood Score: 0%

Overall Tally

I mean, at this point, are we even surprised? We already know that Undetectable AI is very effective against good AI detectors, but against one that has proved itself (time and again) to be unreliable? It’s one-sided, to say the least.

ZeroGPT Pros and Cons

Should You Use ZeroGPT?

It doesn’t really matter if your AI detector is free if it doesn’t work.

ZeroGPT is a good example of false advertising. It just doesn’t add up. You can’t just claim that you have 98% accuracy when most AI-generated text that passes through your model comes out as human. 

And if that’s not enough to sway you, have I mentioned that their interface is riddled with way too many ads? I get that people are just here for the detector and not the aesthetics, but would it really hurt to put in some effort?

If someone needs a lesson on how not to make a free AI detection tool, then show them ZeroGPT. Otherwise, use more trustworthy platforms like Copyleaks and Sapling instead.