Lol just tested an article I wrote (in literally 20 seconds) using wordplay and this tool says 99% real and 1% fake. I wrote another using the OpenAI custom model (DaVinci) and it says 99.7% real, and for 2nd article, it says 2% fake. I am going to ask for a refund.
Funny thing is that they have an ai detection tool and they themselves are offering an ai writing tool.
GPT-3 ai model is not public. Unless they made their model public or at least their dataset public and someone trains using the same dataset, no one can tell for 100% if the given article is written by gpt-3 (an AI) or not. (For more info on this... one must read about Grover AI - A State-of-the-Art Defense against Neural Fake News)