Stop settling for the model’s first draft. When you ask an AI to write copy, the first output is usually the most average, statistically probable response. That is not what converts.
Instead of asking for one draft, make the model generate multiple versions and then judge them against each other.
The Prompt
Write my landing page eight different ways. Different hook, different structure each time. Then create five judges: a skeptical CFO, a distracted founder scrolling at midnight, a competitor, my ideal customer, and a conversion copywriter. Have every judge score every version and explain the scores. Kill the losers. Merge what worked into a final version, and show me the scoreboard so I can see why it won.
[Attach your product or startup idea]
Why this works
This forces the model out of “helpful assistant” mode and into a multi-agent simulation. By assigning specific, highly critical personas (like a skeptical CFO or a distracted founder), you ensure the copy is tested against real-world friction before you ever publish it.