I also tried to generate longer text in "normal" English with prompts designed to produce output that is not identifiable by these detection tools (without putting too much work, no more than a spammer would) and it catches most of them pretty well. The lowest percentage I got was 60%. This is one of the better tools. Interested to see how it performs when GPT 5 comes out.
Well, again, most spammers using this tool are too lazy to create a prompt to make the output look different from the usual AI blather. That's why they're using AI: they are lazy and uncreative.
At the end of the day what really matters whether the post is considered "spam" or not. If its not adding anything to the conversation then its most likely spam and should be deleted either way.