I did many (really, many) test with AI content. 90% failed

If you want to bypass these detectors, here are some tips that worked in tests:Adds human errors and variations Changes punctuation in a natural way (e.g. adds a few dashes, exclamation marks, ellipses).Purposely introduce minor typos or repetitions.Rewrite certain passages with shorter or chopped sentences, as a human would.
We’ve tested all of this on large-scale news sites before. We were using Jasper about three years ago, and it was completely useless. I don’t see a future for the content industry or these kinds of websites.
 
I've sent two Claude generated articles to https://copyleaks.com/ai-content-detector, one was 100% AI, the other was 0%. ZeroGpt and QuillBot have them at around 20% AI. This is how useful these tools are. You will only get fooled by them, 100% unreliable.

Google knows how to detect AI content, and also knows if the AI content is helpful or not. I'm still testing, results are mixed.

I can't edit all articles manually, I won't pass them through a humanizer because it comes out as total shit. From API to production directly. All articles are factually accurate, so in the end engagement should be OK. I'm going to burn all my Claude, Openrouter, Fireworks credits anyway...

OpenAI is one of the worst "writers" I've tested, Claude is much, much better. Deepseek is also good but unreliable in the output, Gemini is OK (I use if for parts of the text which need 2-3 edits along the way, because it's very cheap).

Edit: forgot to mention that I'm not expecting to succeed with no human editing, but it's worth a try :)
 
Last edited:
I've sent two Claude generated articles to https://copyleaks.com/ai-content-detector, one was 100% AI, the other was 0%. ZeroGpt and QuillBot have them at around 20% AI. This is how useful these tools are. You will only get fooled by them, 100% unreliable.

Google knows how to detect AI content, and also knows if the AI content is helpful or not. I'm still testing, results are mixed.

I can't edit all articles manually, I won't pass them through a humanizer because it comes out as total shit. From API to production directly. All articles are factually accurate, so in the end engagement should be OK. I'm going to burn all my Claude, Openrouter, Fireworks credits anyway...

OpenAI is one of the worst "writers" I've tested, Claude is much, much better. Deepseek is also good but unreliable in the output, Gemini is OK (I use if for parts of the text which need 2-3 edits along the way, because it's very cheap).

Edit: forgot to mention that I'm not expecting to succeed with no human editing, but it's worth a try :)
Have you tried https://originality.ai/ai-checker?
 
This is another AI content. Chatgpt with very High prompt. The page are correctly indexed, but traffic it's zero.

Schermata 2025-03-11 alle 00.24.09.png
Schermata 2025-03-11 alle 00.24.24.png
 
No.

But I know my texts are all 100% full AI, and any external tool besides what Google has won't show me what I need. I don't need to know if my text is AI generated or not (I know it is AI), I need to know if Google will accept it or not.

And for this matter, I need to access Google's own detection system. Which is impossible.

This is how stupid is to rely on AI detectors.
 
Back
Top