Picture the first minute of someone trying your AI tool. They sign up, land on a blank box with a blinking cursor and wonder what on earth to type. Some type something vague and get something vague back. Many just leave. Nothing is broken. The screen asked too much of a stranger.
A blank prompt box asks users to do the hardest work first. This post is for product teams building AI tools. It covers five kinds of empty state, a framework we call the First Win Path and a fill in checklist you can apply to your own first run screen.
Key takeaways
In an AI tool, the empty state is the onboarding. Treat the first screen as the most important screen.
A blank prompt box is the hardest empty state. Give users starting points, not a cursor.
Aim for one real result in the first session, using the user’s own data where you can.
Ask for as little as possible before the first result. Set up can wait.
Measure time to first accepted output, not sign up completion.
Why is the blank prompt box a problem for AI onboarding?
A blank prompt box makes the user do the hardest part: knowing what to ask. Most people do not know what a new AI tool can do, so they type something vague, get a vague result and decide the product is weak.
This is the cold start problem, and it is a design problem, not a model problem. Google’s People + AI Guidebook asks teams to set expectations about what the system can and cannot do early. The same guidance appears in Microsoft’s human AI interaction guidelines, which start with making clear what the system can do and how well.
In a traditional SaaS app, an empty state is a list with no rows. In an AI tool, it is the whole product before the user has given it anything to work on.
What are the five types of empty states in an AI tool?
There are five, and each needs different content.
Empty state | When it appears | What it should do |
|---|---|---|
First run | Brand new user, no data | Get to one real result fast |
Blank prompt | Input waiting | Offer starting points, show scope |
No results | The model found nothing | Say why and suggest the next step |
No data yet | Connected, nothing synced | Show progress and what appears when it finishes |
Error or low confidence | The model could not answer | Be honest, give a way forward |
Most teams design the first one and forget the rest. The “no results” and “low confidence” states are where users decide whether the tool is honest.
Also read: AI Copilot UX Patterns: 7 Interface Patterns That Make Assistants Useful
What is the First Win Path?
The First Win Path is the shortest route from sign up to one result the user would actually use. It has four steps, and each removes one reason to leave.
Step | The user’s question | What the screen does |
|---|---|---|
1. Orient | What is this for? | One line of scope and one sample result |
2. Seed | What do I try? | Three to five starting prompts or an import |
3. Result | Is this any good? | A result on real or realistic data, with source or state shown |
4. Next | What now? | One clear next action, not a menu |
The path works because each step answers one question and then gets out of the way. Do not add a tour. A tour explains the interface before the user has seen a reason to care. A result explains the product.
How do you design a good blank prompt state?
Give three things: a scope line, starting points and a way to bring real data. Together they turn a blank box into a short menu.
Scope line. One sentence on what the tool is for, in the user’s words. “Ask about deals, forecasts and rep activity” beats “Ask me anything.”
Starting points. Three to five prompts that match real jobs. Make them clickable, and make each one produce a good result on the sample data.
Bring your own. Offer to connect or import data. A result on the user’s own data is worth ten on sample data.
Write the starting prompts from your support tickets and sales calls, not from your own imagination. They should sound like what users type.
Pro tip: Test every starting prompt on a fresh account before you ship. A suggested prompt that returns a weak result teaches users to distrust the product in the first minute.
How should an AI tool handle onboarding questions?
Ask only what the first result needs. Each extra question before the first result costs completion.
A role question can change the starting prompts, so it earns its place. A team size question rarely changes the first result, so it can wait until after. Move the rest of set up behind the first win, where the user already believes the product is worth it.
Ask before the first result | Ask after the first result |
|---|---|
Role or goal, if it changes the starting prompts | Team size and plan details |
One data source, if the result needs it | Extra integrations |
Nothing else | Notifications, invites, preferences |
What are the most common onboarding mistakes in AI tools?
Showing a tour first. Tours explain screens, not value.
Asking “How can I help?” and nothing else. An open question to a stranger gets silence.
Sample prompts that do not work. One bad suggestion costs more than none.
Hiding limits. If the tool cannot do something, say it up front.
Treating errors as an afterthought. The first low confidence answer is a trust moment.
Measuring sign ups. Sign ups say nothing about whether anyone got value.
How do you test your first run experience this week?
Watch five new users, with no help, from sign up to first result.
Recruit five people who match your buyer and have never seen the product.
Ask them to try the tool for ten minutes and talk aloud.
Time how long until each gets a result they say they would use.
Pass if four of five get there in under five minutes without help.
Note the exact screen where each hesitates. That screen is your first fix.
Pro tip: Track time to first accepted output as a product metric from day one. It makes onboarding visible to the whole team, not just design.
Also read: How to Explain Your AI Product on Your Website
Also read: AI Homepage Messaging Mistakes, How to Design Trust Into AI Interfaces.
To see how we approach interfaces for AI products, read our Simora AI simulation landing page and Nura healthcare platform dashboard case studies, or browse all of our case studies.
Keep reading: AI Copilot UX Patterns: 7 Interface Patterns That Make Assistants Useful, How to Design Trust Into AI Interfaces: 8 Moments Where Users Decide and AI Landing Page Design: Examples and Lessons From Three Pages We Designed
Does your first screen get users to a result?
Send us your first run screen and we will mark where the path from sign up to first result breaks. Book a design review · See our work
AI onboarding and empty states FAQs
An empty state is any screen where the product has nothing to show yet: a new account, a blank prompt box, a search with no results or a feed still syncing. In AI tools it carries more weight than in other software, because it is often the user’s first impression of the product.
Get them to one real result as fast as you can. State what the tool is for, offer a few starting prompts or an import of their data, show a result with its source or state, and then offer a single next step. Move set up and preferences until after that first result.
Often because the first result was weak or never arrived. A blank prompt box asks users to know what to ask, and a vague prompt gets a vague answer. Starting points, real data and honest limits fix more churn than extra features do.
Usually not at the start. Tours explain the interface before the user has a reason to care. A faster path is a good first result, with a short pointer to the next action. Use a tour later for advanced features, once people have seen value.
Yes. Kraftbase is a product messaging and UI/UX design studio for AI and B2B SaaS companies. We design first run screens, empty states and the marketing pages that set expectations before sign up, so the promise and the first screen match.
Time to first accepted output, share of new users who reach it in session one and day seven return rate for those who did. Sign up and tour completion rates are weak signals because neither shows that the user got value from the product.
Three to five. Fewer feels thin and more becomes a menu to read. Each one should map to a real job and return a good result on sample or connected data. Write them from support tickets and sales calls rather than guessing.
About the author
Jenish is a designer at Kraftbase, a product messaging and UI/UX design studio in Vadodara, India, that helps AI and B2B SaaS teams in the US, UAE and India explain and design what they’ve built.
Sources
Amershi et al., Guidelines for Human AI Interaction, Microsoft Research, CHI 2019
First Win Path: Kraftbase framework, not a published standard
Design review
Does your first screen get users to a result?
Send us your first run screen and we will mark where the path from sign up to first result breaks.
Picture the first minute of someone trying your AI tool. They sign up, land on a blank box with a blinking cursor and wonder what on earth to type. Some type something vague and get something vague back. Many just leave. Nothing is broken. The screen asked too much of a stranger.
A blank prompt box asks users to do the hardest work first. This post is for product teams building AI tools. It covers five kinds of empty state, a framework we call the First Win Path and a fill in checklist you can apply to your own first run screen.
Key takeaways
In an AI tool, the empty state is the onboarding. Treat the first screen as the most important screen.
A blank prompt box is the hardest empty state. Give users starting points, not a cursor.
Aim for one real result in the first session, using the user’s own data where you can.
Ask for as little as possible before the first result. Set up can wait.
Measure time to first accepted output, not sign up completion.
Why is the blank prompt box a problem for AI onboarding?
A blank prompt box makes the user do the hardest part: knowing what to ask. Most people do not know what a new AI tool can do, so they type something vague, get a vague result and decide the product is weak.
This is the cold start problem, and it is a design problem, not a model problem. Google’s People + AI Guidebook asks teams to set expectations about what the system can and cannot do early. The same guidance appears in Microsoft’s human AI interaction guidelines, which start with making clear what the system can do and how well.
In a traditional SaaS app, an empty state is a list with no rows. In an AI tool, it is the whole product before the user has given it anything to work on.
What are the five types of empty states in an AI tool?
There are five, and each needs different content.
Empty state | When it appears | What it should do |
|---|---|---|
First run | Brand new user, no data | Get to one real result fast |
Blank prompt | Input waiting | Offer starting points, show scope |
No results | The model found nothing | Say why and suggest the next step |
No data yet | Connected, nothing synced | Show progress and what appears when it finishes |
Error or low confidence | The model could not answer | Be honest, give a way forward |
Most teams design the first one and forget the rest. The “no results” and “low confidence” states are where users decide whether the tool is honest.
Also read: AI Copilot UX Patterns: 7 Interface Patterns That Make Assistants Useful
What is the First Win Path?
The First Win Path is the shortest route from sign up to one result the user would actually use. It has four steps, and each removes one reason to leave.
Step | The user’s question | What the screen does |
|---|---|---|
1. Orient | What is this for? | One line of scope and one sample result |
2. Seed | What do I try? | Three to five starting prompts or an import |
3. Result | Is this any good? | A result on real or realistic data, with source or state shown |
4. Next | What now? | One clear next action, not a menu |
The path works because each step answers one question and then gets out of the way. Do not add a tour. A tour explains the interface before the user has seen a reason to care. A result explains the product.
How do you design a good blank prompt state?
Give three things: a scope line, starting points and a way to bring real data. Together they turn a blank box into a short menu.
Scope line. One sentence on what the tool is for, in the user’s words. “Ask about deals, forecasts and rep activity” beats “Ask me anything.”
Starting points. Three to five prompts that match real jobs. Make them clickable, and make each one produce a good result on the sample data.
Bring your own. Offer to connect or import data. A result on the user’s own data is worth ten on sample data.
Write the starting prompts from your support tickets and sales calls, not from your own imagination. They should sound like what users type.
Pro tip: Test every starting prompt on a fresh account before you ship. A suggested prompt that returns a weak result teaches users to distrust the product in the first minute.
How should an AI tool handle onboarding questions?
Ask only what the first result needs. Each extra question before the first result costs completion.
A role question can change the starting prompts, so it earns its place. A team size question rarely changes the first result, so it can wait until after. Move the rest of set up behind the first win, where the user already believes the product is worth it.
Ask before the first result | Ask after the first result |
|---|---|
Role or goal, if it changes the starting prompts | Team size and plan details |
One data source, if the result needs it | Extra integrations |
Nothing else | Notifications, invites, preferences |
What are the most common onboarding mistakes in AI tools?
Showing a tour first. Tours explain screens, not value.
Asking “How can I help?” and nothing else. An open question to a stranger gets silence.
Sample prompts that do not work. One bad suggestion costs more than none.
Hiding limits. If the tool cannot do something, say it up front.
Treating errors as an afterthought. The first low confidence answer is a trust moment.
Measuring sign ups. Sign ups say nothing about whether anyone got value.
How do you test your first run experience this week?
Watch five new users, with no help, from sign up to first result.
Recruit five people who match your buyer and have never seen the product.
Ask them to try the tool for ten minutes and talk aloud.
Time how long until each gets a result they say they would use.
Pass if four of five get there in under five minutes without help.
Note the exact screen where each hesitates. That screen is your first fix.
Pro tip: Track time to first accepted output as a product metric from day one. It makes onboarding visible to the whole team, not just design.
Also read: How to Explain Your AI Product on Your Website
Also read: AI Homepage Messaging Mistakes, How to Design Trust Into AI Interfaces.
To see how we approach interfaces for AI products, read our Simora AI simulation landing page and Nura healthcare platform dashboard case studies, or browse all of our case studies.
Keep reading: AI Copilot UX Patterns: 7 Interface Patterns That Make Assistants Useful, How to Design Trust Into AI Interfaces: 8 Moments Where Users Decide and AI Landing Page Design: Examples and Lessons From Three Pages We Designed
Does your first screen get users to a result?
Send us your first run screen and we will mark where the path from sign up to first result breaks. Book a design review · See our work
AI onboarding and empty states FAQs
An empty state is any screen where the product has nothing to show yet: a new account, a blank prompt box, a search with no results or a feed still syncing. In AI tools it carries more weight than in other software, because it is often the user’s first impression of the product.
Get them to one real result as fast as you can. State what the tool is for, offer a few starting prompts or an import of their data, show a result with its source or state, and then offer a single next step. Move set up and preferences until after that first result.
Often because the first result was weak or never arrived. A blank prompt box asks users to know what to ask, and a vague prompt gets a vague answer. Starting points, real data and honest limits fix more churn than extra features do.
Usually not at the start. Tours explain the interface before the user has a reason to care. A faster path is a good first result, with a short pointer to the next action. Use a tour later for advanced features, once people have seen value.
Yes. Kraftbase is a product messaging and UI/UX design studio for AI and B2B SaaS companies. We design first run screens, empty states and the marketing pages that set expectations before sign up, so the promise and the first screen match.
Time to first accepted output, share of new users who reach it in session one and day seven return rate for those who did. Sign up and tour completion rates are weak signals because neither shows that the user got value from the product.
Three to five. Fewer feels thin and more becomes a menu to read. Each one should map to a real job and return a good result on sample or connected data. Write them from support tickets and sales calls rather than guessing.
About the author
Jenish is a designer at Kraftbase, a product messaging and UI/UX design studio in Vadodara, India, that helps AI and B2B SaaS teams in the US, UAE and India explain and design what they’ve built.
Sources
Amershi et al., Guidelines for Human AI Interaction, Microsoft Research, CHI 2019
First Win Path: Kraftbase framework, not a published standard
Design review
Does your first screen get users to a result?
Send us your first run screen and we will mark where the path from sign up to first result breaks.