If you write with AI, you already know the loop.
Draft something, run it through a detector, watch it come back flagged, rewrite it, check again. Repeat until it reads human enough to send.
It’s not glamorous but it’s the reality of how a lot of content gets made right now. (Guilty)
The detector you use in that loop matters more than most people think. A bad one gives you false confidence.
You think you have humanized the content, and then your editor or professor runs it through something stricter and the whole thing unravels.
We tested two of the newest AI detection tools Undetectable AI and MyDetector.ai to catch AI texts.
Parts of this article were drafted with AI and run through both. You’ll see the results.
What is Undetectable AI?
Source: Undetectable.ai
Undetectable AI started as a humanizer, paste in AI text, get back something that reads more human.
The detector came later as a companion feature, and it’s now free with no signup required.
Paste your text, hit check, get a result instantly.
They say the detector is up to 99% accurate, trained to catch GPT-4, GPT-5-style output, Claude, Gemini, Llama, and paraphrased or rewritten content.
The paid plans are built around the humanizer side of the product.
- $19/month for 20,000 words
- $31/month for 35,000
- $42/month for 50,000.
All plans include unlimited AI detecting, which means you’re essentially paying for the humanizing and getting the detection for free on top of it.
For a writer like me, that’s a useful setup. Draft with AI, run it through the detector to see where you stand, humanize if needed, check again.
What is MyDetector.ai?
Source: MyDetector.ai
MyDetector.ai is a free AI text detector. You just paste your text or upload a file (PDF, Word, PPT, TXT) and get back a percentage showing how likely it is to be AI-generated. That’s it.
For me, that simplicity is actually the point.
There’s no subscription to manage, and no cap on how many documents you can check.
The detection approach looks for unnatural phrasing, repetitive sentence structures, and patterns that don’t read like something I would write.
Paid plans exist, but they’re built around the AI humanizer side of the product rather than detection.
Detection stays free regardless of which plan you’re on,
Starting from $0 up to $29.99/month for unlimited humanization. If you’re only here to detect AI text, you never need to pay anything (hopefully forever).
The Test Setup
Four text samples, both tools, same writing prompts, scored back to back.
Sample 1 was raw output from ChatGPT.
Sample 2 was raw output from Claude.
Samples 3 and 4 were humanized versions of those same texts run through Undetectable and MyDetector.AI’s detectors.
For context, this is probably the workflow of 99% of writer
Draft with AI, humanize, check if it passes. And it’s the same problem an editor or an educator would face on the other end.
Test # 1 ChatGPT ( Raw )
Verdict: Undetectable A.I identified the text as AI
AI Probability: 96%
Verdict: MyDetector AI identified the text as AI as Human
Test # 2 Claude ( Raw )
Verdict: Undetectable AI identified the text as AI
AI Probability: 63% with 37% Human Probability
Verdict: MyDetector AI identified the text as Human
Test # 3 ChatGPT ( Humanized )
Verdict: Undetectable AI identified the text as Human
AI Probability: 10%
Verdict: MyDetector AI identified the text as Human
Test # 4 Claude ( Humanized )
Verdict: Undetectable AI identified the text as Human
AI Probability: 28 %
Verdict: MyDetector AI identified the text as Human
Undetectable AI vs MyDetector.ai by The Numbers
| Test | Undetectable AI | MyDetector.ai |
| ChatGPT (Raw) | 96% AI | Human |
| Claude (Raw) | 63% AI | Human |
| ChatGPT (Humanized) | 10% AI | Human |
| Claude (Humanized) | 28% AI | Human |
The Verdict
MyDetector.ai called everything human. Not just the rewritten samples, every single one, including raw ChatGPT and raw Claude text that hadn’t been touched at all.
You can’t get a more unanimous result than that.
Either my writing prompts were just that good or the Model V2.0 that is on a BETA version currently has an effect on its performance.
But for now it simply couldn’t flag AI text in any form we gave it.
Undetectable AI text detector held up significantly way better, It caught the raw ChatGPT text at 96% and the raw Claude output at 63% (two solid flags on unedited AI writing).
Once the text was manually rewritten, AI scores dropped to 10% and 28% respectively, and both passed as human.
We’re giving this one to Undetectable AI by a wide margin.
Until that V2.0 beta matures into something more reliable or working for that matter,
Undetectable AI should be your choice.
