Adding AI Features to Your App

Making Your App Smart

🔄 Quick Recall: In the previous lesson, you built a working app prototype — writing the foundation prompt, iterating through generations, connecting a backend, and testing on real devices. Your app works. Now let’s make it intelligent.

AI features are what separate an app that people use once from an app they use every day. A recipe app that just lists recipes competes with a million others. A recipe app that learns what you like, suggests meals based on what’s in your fridge, and adjusts portion sizes to your household? That’s sticky.

But adding AI features doesn’t mean building AI from scratch. It means integrating existing AI services smartly.

The AI Feature Menu

Not every app needs every AI feature. Choose based on what genuinely improves the user experience:

Rule of thumb: If an AI feature doesn’t directly solve a user problem or make a core workflow faster, don’t add it. AI for the sake of AI is a feature, not a benefit.

Adding a Chatbot

The Architecture

A chatbot in your app is simpler than it sounds:

User types message
    ↓
Your app sends to AI API (OpenAI, Claude, Gemini)
    ↓
API processes and returns response
    ↓
Your app displays response in chat UI

The complexity isn’t in the code — it’s in the system prompt that defines your chatbot’s behavior.

Crafting the System Prompt

The system prompt is instructions you send with every message that define who the chatbot is and how it should behave:

You are a meal planning assistant for the FreshPlate app.

Rules:
- Only suggest recipes using ingredients the user mentions
- Include estimated cooking time and difficulty (easy/medium/hard)
- Ask about dietary restrictions before suggesting
- Keep responses under 200 words
- Never recommend raw or undercooked meat preparations
- If asked about non-food topics, redirect to meal planning

Personality: Friendly, encouraging, practical. Like a helpful friend
who loves to cook, not a professional chef lecturing at you.

This prompt turns a general AI into a focused, branded assistant that fits your app.

Integration Options

For Lovable/Bolt apps: Most AI builders can integrate OpenAI or Claude APIs directly. Describe: “Add a chat feature in the bottom tab that connects to OpenAI’s API with this system prompt: [paste prompt].”

For FlutterFlow: Use FlutterFlow’s API integration to connect to AI providers. Custom functions handle the API calls and response display.

For any app: Use a chatbot-as-a-service platform (Voiceflow, Botpress) that provides pre-built chat interfaces and backend logic — you embed their widget in your app.

✅ Quick Check: Why is the system prompt more important than the AI model you choose? Because the system prompt defines your chatbot’s personality, knowledge boundaries, and response rules. Two apps using the same AI model with different system prompts will produce completely different user experiences. The model provides intelligence; the prompt provides identity.

Building Personalization

Start Simple, Scale Later

You don’t need machine learning to personalize an app. Start with rule-based personalization:

Level 1: Preference-Based (No AI Required)

If user selected "vegetarian" → show vegetarian content first
If user prefers "morning workouts" → surface AM routines at top
If user's goal is "weight loss" → prioritize calorie-conscious content

This is just database filtering based on profile data. Any AI app builder supports this.

Level 2: Behavior-Based (Simple Analytics)

If user completed 5 yoga sessions → suggest more yoga
If user always opens app at 7 PM → send reminders at 6:45 PM
If user skips Monday workouts → remove Monday from suggested schedule

This requires tracking user actions and making decisions from patterns. Supabase or Firebase analytics provide this data.

Level 3: AI-Powered (API Integration)

Send user's history + preferences to AI API
→ AI returns personalized content ranking
→ App displays content in personalized order

This is the most powerful but requires more integration work. Only implement if Levels 1-2 aren’t sufficient.

The Personalization Prompt

For Level 3, the AI API call looks like this:

Given this user profile:
- Age: 35, Goal: build strength, Experience: intermediate
- Completed: 20 upper body, 8 lower body, 3 cardio workouts
- Favorite time: evening, Typical duration: 30 minutes
- Last 3 sessions: chest press, shoulder press, bicep curls

Rank these available workouts by relevance:
[list of 50 available workouts]

Return the top 10 in order with a one-line reason for each.

The AI returns a personalized ranking that feels curated for each user.

Smart Recommendations

The Cold Start Problem

New users have no history — so what do you recommend?

Solution: The three-question onboarding

Ask three targeted questions during sign-up that give you enough data for immediate personalization:

1. Goal question: “What’s your main goal?” (weight loss / muscle building / flexibility / general fitness)
2. Experience question: “How would you describe your experience?” (beginner / intermediate / advanced)
3. Preference question: “What do you enjoy most?” (yoga / strength / cardio / variety)

Three questions take 15 seconds and give you enough data to personalize the entire first-day experience. Without them, every new user sees the same generic content.

✅ Quick Check: Why is the onboarding questionnaire critical for personalization? Because personalization requires data, and new users have none. Three targeted questions at sign-up provide enough signal to customize the first experience — which determines whether users come back. An app that feels “made for me” from day one has dramatically higher retention than one showing generic content for the first week.

Smart Notifications

The Three Dimensions

AI-optimized notifications personalize three things:

Timing: When is this specific user most likely to engage?

• Track when users open the app → send notifications 15 minutes before their typical open time
• Avoid sending during sleeping hours (infer from usage patterns)

Content: What does this user care about?

• Track which features they use → notify about related content
• Track what they ignore → stop sending those types

Frequency: How often before they uninstall?

• Start conservative (1/day max)
• Increase if engagement stays high, decrease if opens drop
• Some users want daily reminders; others want weekly summaries

Services for smart notifications: OneSignal (free tier), Firebase Cloud Messaging (free), Braze (enterprise).

Key Takeaways

• Add AI features only when they solve a real user problem — AI for the sake of AI adds complexity without value
• Chatbots are API calls with a good system prompt — the prompt defines personality and boundaries, not the model
• Start personalization with simple rule-based filtering (Level 1), then behavior-based (Level 2), then AI-powered (Level 3) — only escalate when simpler levels aren’t enough
• Solve the cold start problem with a three-question onboarding that enables immediate personalization
• Smart notifications personalize timing, content, and frequency per user — one relevant notification beats eight generic ones

Up Next: You’ll learn how to test and polish your app — ensuring it works flawlessly across devices, handles errors gracefully, and is ready for real users.