AI Product Design

AI Empty States

How to design the zero-state in AI-powered products — when there is no history, no output, and no data. The empty state is the first impression and the primary onboarding moment.

#empty state#zero state#ai onboarding#first run#ai ux#blank state

What is it?

An AI empty state is the UI presented to a user who has no conversation history, no AI-generated content, and no prior interaction with the AI feature. It is the starting point for every new user and the re-entry point after clearing history. Unlike traditional empty states (which might show a simple "No items yet"), AI empty states must also communicate capability, invite engagement, and set expectations.

Why it matters

The AI empty state is the single most important screen in an AI product. It is the first thing every new user sees. It must simultaneously communicate: "Here is what I can do," "Here is how to start," and "Here is what good results look like." Products that present a blank input box with minimal context have dramatically lower activation rates than products that guide users into their first successful interaction.

Best Practices

Show capability through curated example prompts, not just a text description. Concrete examples communicate ability better than abstract descriptions.
Include 3–6 starter prompts that represent the most valuable, most reliable use cases for the AI in this product context.
Use the empty state to demonstrate the output format — if the AI produces structured reports, show an example report structure.
Include a capability statement that also implies limits: "Ask me anything about your codebase" (scoped) vs. "Ask me anything" (overpromises).
Make the starter prompts clickable — tap to populate the input and submit immediately, not just to populate the input for manual editing.
Vary starter prompts by user context or role where possible — a marketing user and an engineer have different valuable use cases.
Keep the empty state visually engaging but not distracting. The primary goal is to get the user to submit their first prompt.
Include a brief, single-sentence statement of what the AI is and is not (not a full help document).

Common Mistakes

A blank input box with only "Ask a question..." as context — maximum user uncertainty, minimum activation.
Overpromising in the empty state: "I can do anything!" followed by limitations discovered painfully.
Example prompts that are too generic to be useful: "Write an email" (for what? to whom? in what context?).
Starter prompts that don't reflect the most reliable model capabilities — if they fail, trust is damaged immediately.
Empty states that require scrolling to see the most important content — the call to action should be immediately visible.
Not updating empty states as AI capabilities expand — stale examples underrepresent current capability.

Checklist

3–6 curated starter prompts are shown representing the most valuable use cases

Starter prompts are clickable to auto-submit, not just auto-populate

The capability scope is clearly implied by the starter prompts and framing

Starter prompts represent reliable, impressive use cases (not edge cases)

The primary CTA (input area) is visible without scrolling

The empty state communicates what this AI is for, not just that it exists

A brief capability statement is present

Starter prompts are tested with real users to confirm they produce good outputs

Research & Theory

First-Run Experience and Activation (Intercom research)

Research across SaaS products consistently shows that users who reach their "first value moment" within the first session have dramatically higher 7-day retention.

Why it's relevant

The AI empty state is the primary activation gate. Getting users to their first successful AI interaction is the most important conversion point in the product.

Paradox of Choice in AI Prompting

Users presented with a blank AI input have infinite choices, which causes decision paralysis. Offering 4–6 concrete options dramatically reduces time-to-first-prompt.

Why it's relevant

Starter prompts are not a crutch — they are the application of Hick's Law and Choice Architecture to the most critical user interaction.

Real-World Examples

Perplexity

Home screen shows focused starter categories (Writing, Coding, Research, Finance) with specific example queries inside each. Visual differentiation from search. Immediate access without login.

Claude (Anthropic)

Clean empty state with example capabilities shown as clickable suggestion pills. Categories match actual strengths: analysis, coding, writing, Q&A. New conversation shows fresh suggestions.

Copilot (Microsoft)

Context-aware empty states that change based on the host application. In Word, suggestions are writing-focused. In Excel, suggestions are data analysis-focused. Capability is surfaced in context.