My GPT frameworks and workflow have continued to evolve the last two years, but I've narrowed down four simple tactics that turn AI from a vending machine into a true creative partner:
1. Embed expert processes
Tell the model not just to act like a creative director but exactly how that expert thinks and works — step by step, constraints, and goals — so it behaves like a real collaborator.
2. Build iterative feedback loops
Great creatives obsess over revision. Assign numbered options, collect quick quantitative feedback, then feed those choices back into the next round for rapid refinement.
3. Demand pattern-breaking and variety
LLMs default to safe predictions. Break the mold by asking for absurd, risky, even "bad" ideas to discover unexpected gems beyond the usual outputs.
4. Require strong opinions
Avoid polite hedging. Define personas with distinct POVs (the minimalist Tastemaker, the expressive Artist, the no-fluff Skeptic) to stress-test your concepts and reveal blind spots.
Stop treating AI like a slot machine and start training it with process, iteration, divergence, and debate. That's how you get truly creative results.