Skip to main content
Back to Documentation

Writing Great Prompts

How to write prompts that get the best results from the Mobiforge AI builder.

Why prompts matter

The better your prompt, the better your app. Mobiforge uses your description to generate real React Native code — screens, navigation, backend tables, and business logic. Specificity is everything. A vague prompt produces a vague app; a detailed, structured prompt produces production-ready code with the screens, flows, and data models you actually need.

What makes a great prompt

Be specific about screens

List every screen your app needs — home, profile, settings, detail views. The AI generates one file per screen, so naming them upfront ensures nothing is missed.

Describe user flows

Explain what users do step by step — login, browse, add to cart, checkout. Flows reveal the navigation structure and the state the app needs to manage between screens.

Mention data

What data does your app store? Users, products, orders, bookings, reviews? Naming your entities lets the AI build the right Supabase tables and relationships from the first generation.

Specify features

Auth, payments, push notifications, maps, camera — name them explicitly. Each feature requires different libraries and backend configuration that the AI sets up only when you ask.

Prompt examples

These prompts give the AI everything it needs to build a complete, working app on the first generation.

Good — Fitness tracker
A fitness tracker where users log daily workouts with exercise name, sets, and reps. Users can create custom workout plans, see a weekly progress chart, and follow friends. Include auth, a home feed, workout log screen, and profile page.
Good — Food delivery app
A food delivery app with a restaurant catalog, menu browsing, cart, Stripe checkout, and real-time order status. Restaurants have a separate dashboard to manage their menu and see incoming orders. Users can leave reviews after delivery.

Common mistakes

Too vague

"Build me a social app"

No screens, no data model, no flows. The AI has nothing to reason about and will produce a generic shell with placeholder content.

Too broad at once

Don't try to describe 50 features in one prompt. The AI performs best when you build iteratively — get the core working, then add features one conversation at a time. Attempting too much at once leads to incomplete implementations.

Skipping data

If you forget to mention what data your app stores, the AI will guess — and its guess may not match your mental model. Always name your key entities (users, products, orders, etc.) and describe their relationships.

Iterating with AI

Your first prompt doesn't need to be perfect. After Mobiforge generates the initial build, every message you send in the chat is an edit — not a rebuild. The AI reads your existing code, understands what's already there, and makes targeted changes.

Use follow-up prompts to add features, fix screens, change styling, restructure navigation, or modify your data model. Each iteration is fast and versioned, so you can always roll back if something goes wrong.

Add a screen

Add a notifications screen with a list of recent activity

Change UI

Make the home screen use a horizontal card carousel instead of a list

Add a feature

Add Stripe payments to the checkout screen