ai-onboarding-design

AI Onboarding Design

The Law

AN AI FEATURE WITHOUT DESIGNED ONBOARDING FAILS ITS MOST IMPORTANT USERS FIRST.
"They'll figure it out" is a blank input — users probe randomly, hit limits before they find value, and don't return.
Capability communication + trust signals + first-win design IS AI onboarding.

When to Use

Trigger when:

• Designing the first-run experience for any AI feature
• Designing empty states for a chat, AI assistant, or agent interface
• Adding AI capabilities to an existing product for the first time
• Reviewing whether existing AI onboarding communicates capability and builds trust

When NOT to Use

• Features with returning users who already know the interface (post-onboarding flows)
• Internal developer tools where the audience already knows the model's capabilities

The Three Onboarding Obligations

1 — Capability Communication

Users cannot use what they don't know exists. The empty state must teach the model's capabilities.

What to show:

• 4–6 example prompts, specific to this feature's domain
• Example prompts are clickable and submit immediately
• Prompts cover: simple query, complex analysis, creative use, edge of capability

What can I help you with?

┌─────────────────────────────┐  ┌──────────────────────────────┐
│ Summarise this contract     │  │ Compare these two proposals   │
└─────────────────────────────┘  └──────────────────────────────┘
┌─────────────────────────────┐  ┌──────────────────────────────┐
│ What are the key risks?     │  │ Draft an email about...       │
└─────────────────────────────┘  └──────────────────────────────┘

What not to show:

• "Ask me anything!" — too vague; users don't know what "anything" means for this feature
• Examples outside the model's actual capability for this use case
• More than 6 examples — choice paralysis

2 — Trust Building

AI features require more trust than deterministic features. Users don't know:

• Where the data comes from
• Whether the model makes things up
• What happens to their input