A few AI humanizers can produce text classified as human, but they make the writing clunky and lower quality; most do not work reliably.
🔒 2 more in the full analysis
ChatGPT Clever AI Humanizer GPTZero Grammarly Humanizer.com QuillBot Turnitin UndetectedGPT AI Humanizer UndetectedGPT AI Humanizer
Searchable transcript of 20 Hours Testing AI Humanizers—Do Any Actually Work? — Mateusz Jurga (04:23). Search for a phrase, then click its timestamp to jump straight to that moment in the video.
Captions sourced from the original video on YouTube, published by Mateusz Jurga. The video, its captions and all related intellectual property remain the property of their respective owners; AINotes claims no ownership. Provided for research, accessibility and search — see the Transcript Notice and Copyright Policy.
00:00 I spent about 20 hours researching a humanizers and it's really bad. Most of them don't work and the few that do work unfortunately come with a price. They create their own problems too. And then if you're really good at writing yourself, then you can still get detected as AI simply because you write very well. So I will show you a solution to that.
00:24 I'll show you the problems and the solutions and what you can potentially do. First of all, to check if AI humanizer works, you need to have an AI detection system that you trust. So, I ended up working with GPT0 here. And this is a text I created and it shows as human 100% human. And then I have this text here which was generated by chart GPT. And it comes with 100% AI.
00:48 So, we have a positive and a negative control. So, it seems fine. And what I will try to do now is to humanize this text. So this is Quillbot and it comes as 100% AI unfortunately. Then we have humanized AI. This is a mix of AI and human. So it's also a fail. Then this is grammarly.com comes as AI as well unfortunately. This is clever humanizer again all AI.
01:12 And now we have humanizer.com. And this is coming as human which is what we want. And I will leave a link in the description down below to this AI humanizer. But the problem here is that once you start reading this text, it's bit clunky. It's a bit weird. Especially the sentence that I highlighted, it also means investing wisely, protecting with wills and more.
01:37 What is that about? So the point is that this AI humanizer is making the text human by introducing mistakes by essentially lowering the quality of your written English. And that's how they get away from AI detection. And then we have AI humanizer and it also is 100% human here. But again, you know, feel free to pause this to read this. It is a bit clunky.
02:03 It is bit weird. And I wouldn't write my name under this. I would want to rewrite this myself honestly. And then it got me thinking, am I the only one? Is this an inherent problem with the language that I'm testing? So I found this thread started by SPAL on Reddit. I will leave a link in the description down below. Honestly, it's really good and clearly SP spent hours and hours researching this as well.
02:29 And it just bags up what I'm trying to tell you here. He used all of these prompts to try to humanize text. using chat GPT and then someone else came and they added even more prompts and then they also tried so many different AI humanizes. There you go. We have a long list of some apps that they tried here and the bottom line is if they work they just make the language unusable just really clunky.
02:56 So there is no point using any of them because they just make the language clunky and weird and you will have to rewrite it yourself anyway. So was the point. But then on the other hand, what if you write very well, right? Especially if you're writing something formal and academic, what is going to happen? And guess what? I found my own text to be detected as AI generated.
03:20 So again, I went back to Reddit to see if anyone have a similar problem. And this person here is having exactly the same issue with 80% of the actual writing being detected as AI. And there's few other people in the post here talking about the same issues. So this is clearly a common issue even with turn it in that is used by all the major universities.
03:44 So if someone comes to you and says oh turn it in showed that your writing is AI generated but I think honestly the only way to prove that it's not is to show them previous versions of a document. So if you're writing a literature review, just keep saving different versions of the document. So then you can show them, look, I actually worked on this.
04:04 This is my own work. Because right now, this is clearly just numbers game. And if something is very well written, well, it must be AI. If it's not well written, then it's human.