They're far from perfect, like grammarly, and I've managed to fool them myself. I would say with large samples of writing, though, they can be effective. For example, I was able to confirm beyond all reasonable doubt that a few writers were using AI by taking 5 samples of their own writing and comparing them against 5 samples of my own writing within GPTZero. And in each case I did that with a person actually using AI, their articles came back with uniformly higher percentages of AI generated content than my own. It's not a perfect system, but I've been confident in the results I've gotten from it when I've been a little more meticulous in my approach.