Another thing I noticed with AI checker tools is that sometimes the result depends on how long the text pasted is, the Huggingface GPT2 detector tool actually states clearly that it needs a certain number of tokens (so a certain lenght of the text) to be more reliable, but other tools don't.
Some ChatGPT detector tool also heavily fluctuates in the results, sometimes they give a 70% human content ratio for content that I literally copy-pasted in there from ChatGPT itself (and I didn't use any prompt to make it more human-like such as rewriting the text, spinning or any prompt specifically made for tricking tools, I just told "tell me X", it gave me the output and I just copy pasted that same output into Ai content detector tools).
Also tried searching for articles made way before the Ai copywriting tools, restricting search to articles from 2018, and copy pasted something in there and they detected AI content lol.
Those "AI detection tools" are easy to trick but can also (and that's even worse) generate a lot of false positives imo, you can trick the tool by adding some white spaces inside or just add some weird punctuation and then you go from 40% human content to 90% but then flag as AI content some random text that was human written (I mean, maybe poorly written but still human written).