Give someone an empty text box with "Ask me anything" above it, and a surprising number of people don't know what to type. They try something simple, get a generic answer, and decide the feature isn't for them.
This is the blank prompt problem. The flexibility that makes AI features powerful also makes them hard to start using. With a normal feature, the buttons show you what's possible. With a text box, you have to already know.
At WiZR, getting people to actually use Frontline's AI features inside their daily work was a big part of adoption. A lot of what helped had less to do with the model and more to do with what we put around the input box.
Why it happens
- People don't know what the feature can do, so they don't know what to ask.
- They don't know how to phrase a request to get a good result.
- Their first attempt is often vague, so the first answer is mediocre.
- A mediocre first answer sets low expectations for everything after.
The first interaction matters a lot. If it goes badly, many people never try a second time. I wrote about measuring this kind of drop-off in the AI adoption metrics post.
Approaches that help
| Approach | How it works | When it's best |
|---|---|---|
| Starter prompts | Clickable example requests near the input | Early, when users don't know the scope |
| Task-based entry points | Buttons for specific jobs, like "Summarise this" | When there are a few common tasks |
| Examples using their data | Suggestions based on the document or page they're on | When context is available |
| Templates | Structured forms that fill in a good prompt | Repeated tasks with a consistent shape |
| Inline guidance | Tips that appear when a request is vague | Helping users improve over time |
| Showing good results | A sample output before they ask | Setting expectations about quality |
Start from the task, not the box
The most effective change is often removing the blank box as the main entry point. Instead of "Ask me anything," put the AI where the work happens. A "Draft a reply" button next to an email. A "Summarise" option on a long document. A "Create training material" action on a product update.
The user doesn't have to think of a prompt. They pick a task they already understand, and the product writes a good prompt behind the scenes. The open text box can still exist for people who want it.
Make starter prompts specific
Generic examples like "Write an email" don't help much. Specific ones that match the user's role and context do. "Turn these meeting notes into action items with owners" tells the user what's possible and how to ask.
If you know the user's role or the page they're on, tailor the suggestions. A sales manager and a trainer should see different examples.
Help when a request is vague
If someone types a very short or unclear request, the feature can ask a quick clarifying question or offer a couple of interpretations. That gets a better result and teaches them how to ask next time.
Don't overdo it. Asking clarifying questions on every request gets tiresome fast.
Set expectations honestly
Users need a rough sense of what the feature does well and where it struggles. A short line near the input, like "Works best with specific requests about your documents," helps more than a long tutorial.
Google's People + AI Guidebook has a useful section on helping users build an accurate mental model of what an AI system can do. 1
Measure the first few interactions
Track how many people who see the feature try it, how many try it a second time, and which entry points lead to repeat use. If starter prompts lead to much better repeat rates than the open box, that tells you where to invest.
Watch for a common pattern: people who start with a specific task often stick around, while people who start with an open-ended question often don't. If you see it, lean into task-based entry points.