So many people focus on how to prompt LLMs to obtain the right output, expecting the LLM to be able to produce a good version of the output with the right prompt.
At the current stage, as powerful as LLMs are, you’re still a more powerful model. Using an LLM to help you produce great content is more about using the LLM as a source of inspiration, as a devil’s advocate, as that buddy that will help you see things clearer than you ever could on your own. Not because your buddy knows the answers themselves, but because they help you frame the questions.
Of course, this also needs producing good prompts for the LLMs to act as this buddy, but I believe there is much more wiggle room here. Since you’re not trying to get the right output from one black box, but two, and you’re in much more control of the second one. You don’t need the LLM to be your best buddy, just an adequate one.