The key word from GP is "reliably". Sure, you can probably tell when someone blatantly copied and pasted output from ChatGPT and when the prompt was not very so

The key word from GP is "reliably". Sure, you can probably tell when someone blatantly copied and pasted output from ChatGPT and when the prompt was not very sophisticated (write a comment about X). However for all other cases when someone put even the tiniest bit of effort into the prompt or did some post filtering the reliability of such detection plummets significantly.