Verdict
I tested two methods to see which one brings the AI score down the most. The first is installing a GitHub project called humanizer into my AI agent, and the second is feeding the AI a humanizing prompt (this one comes from a YouTube creator, @Educraft). Both methods work and lower the AI score. But the best move is to use them together. OK, let's get into the actual steps, there's a tutorial included.
Real Use Cases
First, let's look at the humanizer project on GitHub. Open your browser and go to: https://github.com/blader/humanizer. Once you're on the project page, you'll see the project details and how it works right away.
OK, the skill offers a few ways to install and use it, but I just went with the most straightforward one: download it directly. I keep a dedicated folder on my computer for these awesome skills.
So once it's downloaded, how do we use it? Let's use hermes, which is pretty popular lately, as an example. Open the hermes chat box and tell it: I'll be sending you articles, and whenever I want one rewritten so it sounds more human, just run this skill to fix it up for me. It's at "D:\skills\humanizer-main".thank.
But I'll use a more set-and-forget method. Hermes loads its own SOUL.md as a system identity instruction every time you start a chat. I'll find that SOUL.md, open it in Notepad, and add this at the very bottom: I'll be sending you articles, and whenever I want one rewritten so it sounds more human, just run this skill to fix it up for me. It's at "D:\skills\humanizer-main".thank.
This does use up a bit more tokens, but the extra cost is tiny. If you humanize copy often, this approach is worth it.
OK, now let's see how well it works. Here, I'll send it a piece of text. ChatGPT wrote this text for me. I ran it through ZeroGPT and the AI score came out at 86%.
Then I ran it through humanizer. First I sent the GPT-written article to hermes, waited for it to send back the result, then ran the new version through ZeroGPT again. This time it showed 65.8%.
OK, next let's try the second method for more polishing. This one comes from a prompt shared by a YouTube creator, @Educraft. I'll put that prompt at the very end of the article. Let's test whether this prompt actually works. It's generated. Let's check how this prompt performed.
65% AI score. Looks like neither method can fully wipe out the AI score on its own. But then the site owner had an idea: what if we combine both methods? I opened SOUL.md, made it run the prompt first, then run the humanizer skill for a second pass. Let's see. After saving, let's check the result.
63.6%. Down from the original 86% to 63.6% now. The drop isn't huge. I ran a few more rounds of testing, some rounds dropped the AI score a lot, some only a little. Overall, the AI score does go down. But my advice: after using a tool to lower the AI score, you should still do a manual polish yourself, because human thinking jumps around in ways tools can't mimic. OK, that's the end of this review. Bye!
Pricing
Both methods are free. If you don't want to pay just to lower an AI score, I'd suggest giving these two tools a try.
Extra content
This is a sample article to verify the article system and Blog aggregation page.
Content supports HTML: bold, italic, lists, and more.
Section heading
Click through to see the full detail page.