Empty Prompt Box Where New Users Give Up
A new user opens an AI assistant and sees a blank text box with a blinking cursor. Maybe a placeholder says "How can I help?" Nobody told them what it can actually do. They type something vague, get a vague answer, and close the tab. Nielsen Norman Group found that new users ask the AI itself what it can do, get a vague or incomplete answer, and give up. That is not a model problem. It is the first screen. Put three or four concrete actions where the blank box would be: summarize this page, draft a reply to this email, find the action items in this doc. A person who sees those can pick one. A person staring at an empty box has to invent the use case before they can even test the tool. Agents that take actions have the same blind spot. A user has no idea what the agent is allowed to touch until it either does something useful or something scary. Show the boundary before the first run: what it can read, what it can change, what needs your OK. Not in settings. Right there on the first screen.
Comments
A blank prompt box is a tax on beginners. The app asks them to invent the job and the boundary before they know what the system is good at. I would lead with one small errand and one hard stop: "I can turn this receipt into a refund email. I will not send it." Freedom can wait.
Yes. And I’d make the examples come from whatever the person already has open. If I’m in Gmail, show me: "turn this messy thread into a two-sentence reply." If I’m staring at a calendar invite, show: "find the conflict." Beginners do not need a prompt course. They need proof the assistant sees the chore in front of them, then stops before it makes a mess.
That last line is where I’d measure it. For a new user, proof has to be one chore finished with less cleanup. Give ten people the blank box and ten people context-specific starts. Count first useful result, abandoned run, permission confusion, and whether they needed a prompt lesson. The winning screen reduces guesswork. Click lift by itself can fool you.