Start a project

The empty prompt box problem

Many AI features greet people with an empty box and a blinking cursor. Infinite choice turns out to be no choice at all. Here’s how to design the first move.

On this page
  1. Why an empty box is hard
  2. What people do instead
  3. Suggest a first move
  4. Show what it can, and can’t, do
  5. Help people shape their request
  6. Keep the door open for experts
  7. A quick checklist

Key takeaways

  • An empty prompt box gives people unlimited choice and no guidance, which slows them down.
  • Suggested starting points, based on context, get people moving.
  • Show what the feature can and can’t do, and keep free typing open for experts.

Open almost any new AI feature and you’ll meet the same thing: an empty text box, a blinking cursor and a friendly line such as “Ask me anything”. It looks simple and open. For many people, it is the hardest possible place to start.

Why an empty box is hard#

Hick’s law tells us that more choices slow decisions down. An empty prompt box is the extreme case: the choices are unlimited, and none of them are shown. People have to work out, on their own, what the feature can do, what it is good at and how to phrase a request. That is a lot of thinking before anything useful has happened.

It is the blank-page problem that writers know well, moved into a product. The more capable the system, the harder it can be to know where to begin.

What people do instead#

Faced with a blank box, many people type something short and vague, get a short and vague answer, and conclude that the feature isn’t very good. Others hover for a moment and leave. Either way, the problem isn’t the AI’s ability. It is the first step.

Suggest a first move#

The simplest fix is to offer a few starting points, chosen for where the person is and what they are likely to want. Suggestions do two jobs at once: they get people moving, and they quietly teach what the feature is good at.

Before

An empty box: “Ask me anything…”

After

The same box, with three suggestions underneath, based on the page the person is on:

  • Summarize this report in five bullet points
  • Draft a reply to this customer
  • Find the three biggest risks in this plan

Good suggestions are specific, and they come from context. A suggestion that mentions the document people are looking at is far more useful than a generic “Write a poem”.

Show what it can, and can’t, do#

People trust a tool more when they understand its limits. A short line near the box, such as “Works best with documents under 50 pages” or “Can’t see files you haven’t shared”, saves frustration later. So do examples of a good request, shown in the placeholder text or a small help link.

Help people shape their request#

Not every request has to be typed from scratch. For common jobs, a few simple controls can do the heavy lifting.

  • Quick options for things people often adjust, such as length or tone.
  • Follow-up suggestions after each answer, such as “Make it shorter” or “Turn this into a table”.
  • Remembered starting points for tasks people repeat every week.

Keep the door open for experts#

Suggestions should help beginners without slowing down people who already know what they want. Keep the text box ready to type into straight away, keep suggestions out of the way of typing, and let experienced users hide them if they prefer.

A quick checklist#

  • Does the empty state offer two to four specific, context-based suggestions?
  • Does it say, briefly, what the feature is good at and what it can’t do?
  • Are common adjustments available without typing?
  • Can experts ignore all of it and just type?

Share this article

Keep reading

Related articles

All articles

Newsletter

New writing, once a month.

Practical notes on design, search and research. No spam, unsubscribe anytime.