Honestly, for most of my cases I can chat with AI tools almost like talking to a human, especially since Claude 3.5 Sonnet and the projects and history feature therefore I don't have a special prompt strategy other than explaining what I want in sometimes more, sometimes less normal language and then adjusting from there

I never used any specific prompts except those integrated into the Keywords Research extension for SEO research.

In most cases, I just talk to it until I get satisfactory results. For some requests I use history to enrich the experience in old threads.
It's not ideal, but I don't believe a prompt will be sufficient for specific cases, either.

The trick here is to actually measure it. How do you define that something is better in an objective way?

But yes, I would be interested in!

I insert as much detail as possible, and experiment. Also, depends which tool you are using. Since discovering Perplexity.ai the process is much less frustrating than using ChatGPT. Sometimes I use both to compare, but perplexity seems to have nailed it on the back end to act more more like a real co-pilot.

But what about instructions you can use in almost every conversation? Like; 'only give me the answer, no explanation' or 'take a deep breath before you answer' or 'ask me questions if you are not sure'. It might depend on the conversation you're having though...

I recently discovered Fabric -> github.com/danielmiessler/fab…
It's a great library of prompts that are available in CLI and scripts