Free Generative AI Detectors: Which Ones to Choose? [Complete Test 2025]

•1250 words, 5 min. read

Avatar of Pierre-Nicolas Schwab

By Pierre-Nicolas Schwab

IntoTheMinds founder

In this article, I share the results of a test I conducted on 11 generative AI detectors. A clear winner emerges among the free tools. The results are mixed, or even frankly poor, for half of the AI detectors tested.

Free Generative AI Detectors: Which Ones to Choose? [Complete Test 2025]

Are generative AI detectors reliable? Since the surge of content produced by generative AIs (ChatGPT leading the way) on the internet, detecting this type of content has become a priority. Google has indeed announced in its latest search engine update that it will penalize low-quality content. I therefore tested 11 free generative AI detection tools to determine which are the most reliable. As you will see in this article, the results are far from consistent and often very disappointing. Compared to my 2024 test, I found that 2 tools no longer worked (plagiarismdetector.net and neuralwriter.com). I replaced them with ZeroGPT and GPTzero. I also enhanced my testing method to account for ChatGPT5.

Contact IntoTheMinds, market research agency

AI Detector: Results in 30 Seconds

  • The testing method was enhanced to account for ChatGPT 5
  • Quillbot achieves the highest overall score (91.25%) but shows significant shortcomings in detecting texts generated with ChatGPT 5
  • Copyleaks comes in second place (91.06%) but performs almost flawlessly on AI-generated texts, including those generated with ChatGPT 5. However, its ability to accurately detect human-written texts is less effective than Quillbot’s.
  • If your goal in 2025 is to detect texts generated by generative AI, my advice is to use Copyleaks, as ChatGPT5 will become more widespread and increasingly used to produce content. Quillbot is currently less effective at detecting content generated with ChatGPT 5.
  • The following AI detectors should be avoided due to their mediocre results: detecting-ai.com, contentatscale.ai, plag.fr, zerogpt.com

Content written by generative AIs has become a plague on the internet, prompting some to claim their content as 100% human. Google found itself caught in its own trap. It asked for “fresh” content, and that’s exactly what happened when ChatGPT was made available to the public. Some rushed to seize the opportunity to produce content that was original in name only. As I showed in another study, the similarity rate of texts produced by ChatGPT is very high. You’ve probably already had suspicions about the origin of a text, a social media post, or a comment while reading it. When it’s written by a generative AI, it shows.

Faced with the scourge of content generated by generative AIs, tools have emerged to detect them. I selected 11 and tested them (you’ll find the list at the end of this article).

Methodology

To test the ability of various tools to recognize texts written by a generative AI, I prepared a corpus consisting of:

  • 3 texts entirely written by ChatGPT 4.0 in English
  • 3 translations into French of texts written by ChatGPT 4.0
  • 1 text in French entirely written by ChatGPT 5
  • 1 text in English entirely written by ChatGPT 5
  • 3 texts from my blog, entirely written by me in French
  • 3 translations into English of texts written by me
  • 1 text written by a French author (Victor Hugo) before the era of artificial intelligence
  • 1 text written by an English author (Charles Dickens) before the era of artificial intelligence

In total, I had 16 texts distributed as follows:

 

FrenchEnglish
Written by a generative AI44
Written by a human44

I then ran each text through the tools listed at the end of this article.
I only used the free versions of the various tools. Since my 2024 test, Scribbr and Copyleaks now offer analysis of French texts. However, plagiarismdetector.net and neuralwriter.com were no longer functional during the tests. I replaced them with gptzero.com and Zerogpt.com. Unfortunately, gptzero.com only worked partially, so I don’t have complete results for this tool.

The results are summarized in the table below. Texts 1 to 6 were generated by ChatGPT 4.0; texts 7 and 8 by ChatGPT 5. Texts 9 to 14 were written by me in French and then translated into English. Texts 15 and 16 are excerpts from books published before the computer era. Next to the text number, you’ll find the language of the text in parentheses. I also added an indication of the ChatGPT model used.

The results reported in the tables below correspond to the percentage of text detected as having been written by a generative AI.

Detection of Texts Written by a Generative AI

The table below shows the results of the various tools in detecting texts entirely written by ChatGPT. The percentage indicates the portion of the text that the tool attributes to a generative AI.

1 (FR, GPT4)2 (EN, GPT4)3 (FR, GPT4)4 (EN, GPT4)5 (FR, GPT4)6 (EN, GPT4)7 (FR, GPT5)8 (EN, GPT5)
Quillbot72%94%81%91%100%100%47%75%
Copyleaks100%100%99%100%100%100%100%96,3%
Smodin79%81%91%100%100%0%12%
detecting-ai.com56%39%21%90,4%78%75,3%15,6%35,6%
freeaitextclassifier.com85%83,6%100%100%96%88%63%79%
contentatscale.aihumanhumanhumanhumanhumanhumanhumanhuman
corrector.app100%95%88%35%74%98,87%20%95,67%
plag.fr100%100%80%100%80%96%10%14%
scribbr.fr78%83%81%80%78%86%45%75%
gptzero.com98%64%
Zerogpt.com78,65%56,98%42,35%84,31%45,62%98,56%84,35%26,58%

In the English version, Copyleaks performs best with a 99% detection rate for English content created by ChatGPT. If you’re looking for an AI detector that works in both languages, Copyleaks offers the best compromise in this test. In 2024, Quillbot was the leader, but in this test, it made errors on the corpus of texts generated by ChatGPT 5.

At the other end of the spectrum, Contentatscale detects nothing at all. Detecting-ai’s score has plummeted since 2024 (barely 60% in English), and Zerogpt is only slightly better (67% in English, 63% in French).

Detection of Texts Written by a Human

In the second part of the test, the goal is to correctly detect texts written by a human. These should not be attributed to a generative AI. The desired values in the table below are therefore 0% in each column.

GPTzero could not be tested at all in this part despite all our attempts.

Here are the results.

9 (FR)10 (EN)11 (FR)12 (EN)13 (FR)14 (EN)15 (FR, Victor Hugo)16 (EN, Charles Dickens)
Quillbot0%0%0%0%0%0%0%0%
Copyleaks38,3%100%0%0%0%0%0%0%
Smodin0%0%0%0%0%0%0%0%
detecting-ai.com0%0%0%0%0%0%0%0%
freeaitextclassifier.com13%8%12%24%0%0%0%0%
contentatscale.aihumanhumanhumanhumanhumanhumanhumanAI
corrector.app0%0%0%0%0%0%0%0%
plag.fr10%12%14%12%16%9%0%57%
scribbr.fr69%0%0%0%0%0%0%0%
gptzero.com
Zerogpt.com12,4%45,78%22,32%5,6%78%11,3%45,65%8,96%

Regarding the detection of content written by humans, several generative AI detection tools achieve a perfect score:

  • Quillbot
  • Smodin (which has significantly improved since our 2024 test)
  • Detecting-ai.com
  • corrector.app

We can therefore clearly conclude that the detection of human content seems to be more refined. If your goal is to identify human content, choose one of these tools.

In general, generative AI detection tools make fewer mistakes when it comes to content written by humans.

Final Results

To determine the winner(s) of this test, both tasks must be considered. It’s not enough to accurately detect AI-written text; the tool must also not attribute human-written text to AI. Since the test is in two languages (French and English), the free version of the tool must also handle both languages.

As in 2024, Quillbot wins this test with an overall average of 91.25%. This AI detector achieves a perfect score on human-written content. While its detection capability for ChatGPT-generated content remains close to perfect, it makes significant errors on texts created with ChatGPT 5. The detection average for AI-generated texts dropped from 96% in 2024 to 83% in 2025.

This year, the outsider Copyleaks is almost tied with Quillbot (overall average: 91.05%). In 2024, Copyleaks did not support French in its free version. In 2025, this issue is resolved. Copyleaks made two errors on human texts, which lowers its average. However, its ability to detect AI-generated texts is much better than Quillbot’s, especially on texts generated by ChatGPT 5.

The results are therefore more nuanced than in 2024. This year, I would recommend choosing an AI detector based on your objectives:

  • Use Copyleaks to detect AI-generated texts.
  • Use Quillbot, Smodin, Detecting-ai, or corrector.app to detect human texts.

Contentatscale.ai remains to be avoided. Detecting-ai.com and zerogpt.com are at the bottom of the pack. Zerogpt, in particular, achieves an overall detection average of 68%, placing it last in this test of free AI detectors.

List of Tested Generative AI Detection Tools

You're at the end of this article
We think you will also like

Posted under the tags Algorithmic governance and in the categories Data & IT