I really don't think it's a prompt issue, more like it's a fundamental flaw in the construction of the LLM most especially with logic based questions.

For the most part AI generated text - just look at X or LinkedIn - is pretty easy to spot with its words, middle of the road, and overwhelmingly banal grammatical structure.

And I'm not even going to go into detail about hallucinations, suffice it to say it's convinced it's right until you tell it that it's wrong.

A remarkable achievement, yes indeed, but ultimately a cul-de-sac in my opinion - time to reverse out, take on board what's been learned, and leave the GPTs for the marketeers and people who want that free lunch.