Prompt Engineering

Good prompting isn’t about magic words — it’s about being clear. Aurora responds best when you say exactly what you want, one thing at a time.

​

Core Principles

​

Be specific

Vague requests get vague results. Say what you want, where you want it, and how it should behave.

Weak: “Make the login better” Strong: “Add password validation to the login form — minimum 8 characters, at least one number. Show errors in red below the input.”

​

One thing at a time

Break big requests into steps. Ask for the layout first, then the styling, then the logic. This gives the AI room to get each piece right and makes it easier to catch issues early.

​

Describe the problem, not just the fix

When something is broken, describe what’s happening and what you expected. “The form refreshes the page on submit — it should stay on the same page and show a success message” is far more useful than “the form doesn’t work.”

​

Show, don’t just tell

Upload a screenshot or paste an image of what you’re going for. Visual context helps Aurora understand layout, spacing, and design intent much faster than words alone.

​

Iterate, don’t restart

If the output is close but not right, tell Aurora what to adjust. “Move the sidebar to the left and make it narrower” is better than re-describing the entire page. Build on what’s already there.

​

Prompt Patterns

​

Adding a feature

“Add a search bar to the top of the products page that filters results as the user types. Show a loading spinner while searching.”

​

Fixing a bug

“The signup button does nothing when clicked. It should submit the form and redirect to the dashboard.”

​

Styling changes

“Make the hero section full-width with a dark background. Use white text and center everything vertically.”

​

Requesting a full page

“Create a pricing page with three tiers: Free, Pro, and Team. Each tier should show the price, a list of features, and a call-to-action button.”

​

What to avoid

• “Make it look better” — too vague to act on
• Cramming multiple features into one message — leads to half-finished results
• Ignoring the preview — always check the output before your next message
• Repeating the same prompt — if it didn’t work, rephrase or add detail instead