When AI should ask before it answers

Most AI features are designed to answer immediately. That feels fast, until the answer is to a question the person didn’t mean. Then they have to read it, realize it’s wrong, explain again and wait again. One well-timed question at the start would have saved all of that.

The cost of a confident guess#

Requests are often shorter than the thought behind them. “Write a summary” could mean three bullet points for a manager or a page for a client. A system that guesses produces something plausible, and plausible-but-wrong is expensive: people may not notice the mismatch until they have used it.

When to ask#

Asking has a cost too: it delays the answer and adds a step. So ask only when the answer would genuinely change the result.

• The task is long or costly to redo, such as a full report or a batch of changes.
• The request has more than one likely meaning that leads to very different results.
• The action is risky or hard to undo, such as sending, deleting or buying.

For quick, cheap tasks, it is usually better to answer and make it easy to adjust.

Offer choices, not open questions#

An open question (“Can you clarify?”) hands the work back to the person. A small set of likely options is faster and shows that the system understood most of the request.

Ask once, then remember#

Nothing is more tiring than answering the same question every time. Remember preferences within a session, and where it makes sense, across sessions, with a way to change them later.

Always let people skip#

Some people know exactly what they want and just want the answer. A visible “Just answer” option respects that, and the system can state its assumption: “Assuming this is for your team; change it here.”

A quick checklist#

• Does the feature ask only when the answer would change the result?
• Are questions offered as a few clear choices?
• Can people skip, with the assumption shown?
• Is the answer remembered next time?