Have you heard of "The Police Sketch Challenge?" My family tried it this morning. The way it works is simple. One person describes to another, step by step, how to draw out a character. Similar to how an eyewitness would describe a suspect to a forensic sketch artist in order to get a facial composite drawn.
My wife gave the description of each character to myself, my daughter and my son. My daughter gave my wife the description for the composite she drew. The images on this post show all four sketches alongside the character described. Which one came the closest? (And who do you think drew each one?)
Now what does this have to do with AI prompting?
When you prompt AI to do something, there are two sides: the prompter (you) and the recipient (AI). Ideally, you tell AI what you need it to do, or what you need to know, and AI gives you a response, or takes action as appropriate.
The better your prompt, the better the results.
But here's the twist. Not all prompters are effective at what they prompt. And the various AI LLM's (large language models) and agentic AI are not created equally either.
You could give AI the best prompt, but if your AI isn't as powerful as other models, it won't be as effective.
Inversely, you could have the best AI model, but if you don't prompt it properly, it won't get the results you need.
Here are three best practices for AI prompts:
Be clear, explicit, focused and specific about what you want. Clarity produces the best results. Ambiguity gives messy results.
Provide examples (known as context). LLM's draw on knowledge and patterns. What you're trying to do has likely been done before in one way or another. Give an example to lead AI in the right direction.
Have a conversation with your AI. Yes, it sounds weird. You have to remember you're not talking to a human being. But as you iterate through what you're trying to accomplish, your AI becomes better at understanding what it is you're trying to do, and gets you closer to getting there.
Practice makes perfect!