Top 5 AI Content Detectors That Consistently Identify AI-Generated Text

Back in January 2023, I delved into the fascinating world of AI-generated plagiarism and the tools designed to combat it. At that time, the buzz around generative AI was just starting to pick up steam. Fast forward to now, and I've got a fresh update on my hands, reflecting the latest developments in this rapidly evolving field.

When I first put GPT detectors to the test, the best I got was a 66% accuracy rate from one of the three available checkers. Fast forward to February 2025, and I've expanded my testing to include up to 10 different checkers. The results? Three of them nailed it perfectly. Just a couple of months later, that number jumped to five. It's clear that these tools are getting sharper by the day.

What I'm Testing and How I'm Doing It

Before we dive deeper, let's clarify what we mean by plagiarism. According to Merriam-Webster, to "plagiarize" means "to steal and pass off (the ideas or words of another) as one's own; use (another's production) without crediting the source." This definition certainly applies to AI-generated content. If someone uses an AI tool like Notion AI or ChatGPT and then claims the output as their own without giving credit to the AI, it's still plagiarism, plain and simple.

To evaluate these AI detectors, I use five blocks of text: two written by me and three generated by ChatGPT. I run each block through each detector separately, noting whether the tool correctly identifies the text as human or AI-generated. If a detector gives a percentage, I consider anything above 70% as a strong indicator of either human or AI origin. If you're curious to try this out yourself, you can find the text blocks in this document.

The Overall Results

I've run a series of five tests across 10 different detectors, which means I've conducted a whopping 50 individual tests. Yes, it required a lot of coffee! The detectors I tested include BrandWell, Copyleaks, GPT-2 Output Detector, GPTZero, Grammarly, Monica, Originality.ai, QuillBot, Undetectable.ai, Writer.com, and ZeroGPT.

For this update, I added Copyleaks and Monica to the mix, while dropping Writefull since it discontinued its GPT detector. Content Guardian wanted in, but I couldn't get the accounts set up in time for testing.

Here's a snapshot of the overall results. As you can see, five detectors managed to correctly identify all five text blocks as either human or AI-generated.

David Gewirtz/ZDNET

I also tried to spot any trends in improvement over time by comparing results from six different test runs. While I've increased the number of detectors and swapped a few out, there's no clear pattern of improvement. The only consistent result is that Test 5 was reliably identified as human across all detectors and dates.

David Gewirtz/ZDNET

I'll keep testing and hope to see more consistent reliability in the future. However, even with some perfect scores, I wouldn't trust these tools entirely to validate human-written content. As it turns out, writing from non-native speakers often gets mistaken for AI-generated content.

Even though my own writing was mostly identified as human this time around, one detector (GPTZero) was too uncertain to judge, and another (Copyleaks) wrongly labeled it as AI-generated. The results can be quite inconsistent across different systems.

How Each AI Content Detector Performed

BrandWell AI Content Detection (Accuracy 40%)

Originally developed by Content at Scale, this tool now operates under BrandWell.ai, a marketing services company focused on AI. Unfortunately, its accuracy was pretty low. It couldn't differentiate between human and AI content in Test 2, as you can see in this screenshot:

Screenshot by David Gewirtz/ZDNET

Copyleaks (Accuracy 80%)

It's ironic that Copyleaks boasts "the most accurate AI detector with over 99% accuracy" when more than half of the detectors I tested did better. But hey, marketers will be marketers, right? The company mainly caters to educational institutions, publishers, and enterprises looking to ensure content originality and academic integrity.

Screenshot by David Gewirtz/ZDNET

GPT-2 Output Detector (Accuracy 60%)

This tool was developed using a machine-learning hub managed by Hugging Face, an AI company based in New York. While the company has secured $40 million in funding to develop its natural language library, the GPT-2 detector seems to be a user-created tool using the Hugging Face Transformers library.

Screenshot by David Gewirtz/ZDNET

GPTZero (Accuracy 80%)

GPTZero has grown significantly since my first test. Initially, it was a bare-bones site, and it wasn't even clear if it was a company or a passion project. Now, it's a full-fledged operation with a mission to "protect what's human." They offer AI validation tools and a plagiarism checker. However, its performance has dipped a bit. It correctly identified my text as human-generated in my last two runs, but this time, it mistakenly flagged it as AI-generated.

Screenshot by David Gewirtz/ZDNET

Grammarly (Accuracy 40%)

Grammarly is renowned for helping writers produce grammatically correct content, but that's not what I'm testing here. It also checks for plagiarism and AI-generated content. In the grammar checker, you'll find a Plagiarism and AI Text Check button in the lower-right corner:

Screenshot by David Gewirtz/ZDNET

Although Grammarly's AI-check accuracy was poor, it correctly identified the test text as previously published.

Monica (Accuracy 100%)

Monica is a newcomer to the scene, offering an all-in-one AI assistant with various services. Users can select from different large language models. The company claims Monica is the "Best AI Detector Online," but it seems to run content through other detectors like ZeroGPT, GPTZero, and Copyleaks. Interestingly, both GPTZero and Copyleaks didn't fare well in my tests, but Monica—and ZeroGPT—did. It earned a perfect score, but I'll keep an eye on it in future tests.

Screenshot by David Gewirtz/ZDNET

Originality.ai (Accuracy 100%)

Originality.ai is a commercial service that doubles as an AI and plagiarism checker. The company sells usage credits; I used 30 credits for this article. They offer 2,000 credits for $12.95 per month. I ran 1,400 words through the system and used just 1.5% of my monthly allocation.

Screenshot by David Gewirtz/ZDNET

QuillBot (Accuracy 100%)

In my previous tests, QuillBot's results were all over the place—multiple runs of the same text yielded wildly different scores. This time, however, it was consistently accurate and scored a perfect 100%. I'll check back in a few months to see if it can maintain this performance.

Screenshot by David Gewirtz/ZDNET

Undetectable.ai (Accuracy 100%)

Undetectable.ai boasts the ability to "humanize" AI-generated text so detectors won't flag it. As a professional author and educator, I find this feature a bit troubling—it feels like cheating. However, they also have an AI detector, which performed exceptionally well.

Screenshot by David Gewirtz/ZDNET

The AI detector passed all five tests. It's interesting to see the indicators showing flags for other detectors. The company explained, "We developed multiple detector algorithms modeled after those major detectors to provide a federated and consensus-based approach. They do not directly feed into the listed models; rather, the models are each trained based on results they've generated. When it says those models flagged it, it's based on the algorithm we created and updated for those models."

I did have a question about the OpenAI flag since OpenAI's content detector was discontinued in 2023 due to low accuracy. Nonetheless, Undetectable.ai detected all five tests correctly, earning a perfect 100%.

Writer.com AI Content Detector (Accuracy 40%)

Writer.com is a service that generates AI writing for corporate teams. Its AI Content Detector tool can scan for generated content, but its accuracy was low. It identified every text block as human-written, even though three of the five tests were generated by ChatGPT.

Screenshot by David Gewirtz/ZDNET

ZeroGPT (Accuracy 100%)

ZeroGPT has come a long way since my last evaluation. Back then, it lacked a clear company identity and was cluttered with Google ads. Now, it's a polished SaaS service with transparent pricing and contact information. Its accuracy has also improved, jumping from 80% last time to a perfect 100% this time around.

Screenshot by David Gewirtz/ZDNET

Is It Human, or Is It AI?

What about you? Have you tried AI content detectors like Copyleaks, Monica, or ZeroGPT? How accurate have they been in your experience? Have you used these tools to protect academic or editorial integrity? Have you encountered situations where human-written work was mistakenly flagged as AI? Are there detectors you trust more than others for evaluating originality? Let us know in the comments below.

Get the morning's top stories in your inbox each day with our Tech Today newsletter.

You can follow my day-to-day project updates on social media. Be sure to subscribe to my weekly update newsletter, and follow me on Twitter/X at @DavidGewirtz, on Facebook at Facebook.com/DavidGewirtz, on Instagram at Instagram.com/DavidGewirtz, on Bluesky at @DavidGewirtz.com, and on YouTube at YouTube.com/DavidGewirtzTV.