How to write prompts that ship

Good prompts are specific enough for the model to execute and open enough to allow good design choices. Here’s a simple framework that works across Claude, Cursor, ChatGPT, and similar tools.

Start with the outcome, not the tech

Describe what the app does and who it’s for before listing tech. For example:

• Weak: “Make a React app with a form.”
• Strong: “A waitlist page for a B2B tool. Visitor enters email and optional company size; we show a thank-you and send the lead to our CRM. Mobile-first, one clear CTA.”

The second gives the model a clear goal and constraints so the output is usable.

Give one clear “hero” flow

Spell out the main user path: what they see first, what they do, what happens next. That becomes the backbone of the generated app. Extra features can be listed after.

Call out layout and UX when it matters

If you care about layout, say so: “Single column on mobile, two columns on desktop” or “Sticky header, one CTA above the fold.” One or two sentences prevent generic “AI slop” layouts.

Avoid vague or conflicting instructions

• Vague: “Make it look professional.”
  Better: “Clean SaaS style: white background, one accent color for buttons, plenty of whitespace.”
• Conflicting: “Minimal design” plus “Lots of features on the homepage.”
  Pick one priority so the model doesn’t average them into mush.

Use (or buy) blueprints for repeatable patterns

Proven prompts for whole app categories — landing pages, dashboards, blogs — save time and improve consistency. Our Blueprints are tested with real tools so you can ship faster.

Iterate in small steps

Start with the smallest version that’s useful, then add: “Add a pricing section with three tiers” or “Add dark mode using the same accent color.” Small, concrete follow-ups usually work better than one giant prompt.

---

For more terms and definitions, see the glossary.