People do not know what to ask
A traditional app shows its features in menus. An AI product often shows one empty box. New users stare at it, unsure what is possible.
Google's People + AI Guidebook includes onboarding as a core question: how to introduce people to new AI features. The answer is less about explaining and more about letting people succeed once, quickly.
A first-run flow in four steps
Ask one question
Capture the user's goal or role so examples can be relevant to them.
Run a real example
Show a prefilled prompt with a live result, using sample or connected data.
Invite a change
Let them edit the example. Making a change teaches the interaction better than reading about it.
Name the limits
Say what the AI is good at and where to double-check, in two short lines.
Use sample data to skip the cold start
If your AI needs the user's data to be useful, the first run is empty and unimpressive. Offer a sample workspace so people see value before they connect anything. Then make connecting real data an obvious next step.
Avoid the tutorial trap
Product tours teach people to click Next. They rarely change behavior. Replace them with in-context hints that appear when a user is about to do something and disappear once they have done it.
- Show one hint at a time, tied to the screen the user is on.
- Let users dismiss everything and find it again in a help menu.
- Retire a hint after the user performs the action.
What to measure
Track time to first useful result, the share of new users who complete one real task in their first session, and how many return within seven days. Tour completion is not a goal. It is a distraction from these.



