Prompt Engineering: Getting Better Results from AI Tools

You ask an AI: "Write me a scene." It produces something generic and forgettable. Your friend asks: "Write a scene where a reserved architect reveals vulnerability through an argument about design philosophy." That output is specific and usable. The difference? Prompt engineering—the skill of asking questions in ways that get better answers.

Prompt engineering for creative work isn't complicated, but it's countintuitive. Most people assume detailed questions get detailed answers. True, mostly. But the real skill is knowing which details matter and which ones distract the AI.

Here are the principles that actually work: First, be specific about what you want, but vague about how to get it. "A mysterious character" works better than "A character who is mysterious, wearing black, with sad eyes, who speaks quietly but knows secret information." The second version over-constrains. The first sets intent and lets the AI generate variations.

Second, include context about why you need it. "I need dialogue for a confrontation scene where my character reveals she's been lying—the stakes are her relationship, not the lie itself" gives the AI narrative context. Compare that to "Write confrontation dialogue." The AI now understands emotional weight.

Third, specify tone or reference material. "Write this scene in the style of minimalist noir dialogue, like early Cormac McCarthy" anchors the AI to a specific aesthetic. Without that anchor, you get generic contemporary tone.

Fourth, constrain length or form helpfully. Not "Write a short scene" (too vague) but "Write this in 150-200 words, focusing on subtext rather than exposition." Now the AI has a target.

The misconception: that good prompts are longer. They're often shorter. The worst prompts are verbose and try to over-specify everything (character appearance, exact dialogue structure, specific plot outcomes). The best prompts are tight: clear intent, relevant constraints, space for AI creativity.

Many creators write prompts like they're emailing a freelancer. They include everything they can think of. What works better: write prompts like you're giving a collaborator a creative challenge. "I'm stuck on dialogue that feels exposition-heavy. What if we focused on what they don't say instead?" This engages the AI as a thinking partner, not an executor.

Another pro technique: use examples. "Write dialogue that captures the subtext in this exchange [paste example]" often produces better results than describing what you want. The AI's embeddings (remember that concept?) understand patterns better than descriptions.

Try this: Take a creative prompt you've used before. Rewrite it to be 30% shorter while keeping the essential intent. Remove appearance descriptions, exact dialogue requirements, and step-by-step instructions. Keep intent, tone, constraints, and context. Use it with your AI tool. Compare the output to the verbose version—you'll likely find the tighter prompt generates more useful, more surprising work.