Give it what you'd give a new employee
If a new hire couldn't do the task with the information you gave, neither can the AI. Spell out the goal, the audience, the format, and the rules. Don't assume it knows your market, your voice, or your preferences unless you've told it.
Bad: "Write a listing description." Better: "Write a 700-1,000 character MLS listing description for this 3-bed/2-bath ranch in Columbia, MO. No em dashes, no clichés. Lead with the updated kitchen."
Context beats clever wording
What the AI can see matters more than how you phrase the request. A well-loaded Project with your files and instructions will outperform the most carefully worded prompt with no context behind it.
This is the whole argument for Projects, Knowledge files, and Connectors — load the right context and even a simple prompt produces strong output.
Tell it what you're solving, then ask it to write the prompt
Stuck on how to phrase something? Describe the problem and ask the AI to write the prompt for you.
"I want to turn listing details into an MLS description that follows Fair Housing rules. Write me a prompt for that."
Then refine it until it fits your workflow.
Show it what good looks like
One or two strong examples teach the AI more than a paragraph of instructions. When you build a GPT or Skill, put your best examples in the knowledge files or at the bottom of the instructions.
Four things to always include
Role — What job is it doing? ("You are an MLS copywriter who...")
Goal — What's the output? ("Write a 700-1,000 character description...")
Rules — What are the limits? ("No em dashes. No Fair Housing violations.")
Format — How should it be structured? ("Return only the final description, no commentary.")
Iterate, don't restart
You don't have to get it perfect on the first try. Start with a draft and refine it in the same conversation: "Make it shorter," "Lead with the backyard instead," "Remove the clichés." AI works best as a back-and-forth, not a one-shot request.