The analogy

Walk into a carpenter's and tell them "make me a table". They'll bombard you with questions: how big, for how many people, what wood, indoors or out, budget. They're not being dense: without measurements and use they can't cut anything. If instead you come in and say "a dining table for six, in oak, for the kitchen, about three hundred euros", they get straight to work.

The AI is that carpenter. The more boundaries you give it at the start, the less it has to ask you and the less it risks guessing wrong. The clarifying questions aren't a snag: they're the sign that the measurements were missing.

How it really works

A vague request opens many plausible interpretations, and the AI has two roads: ask, or choose itself and risk taking the wrong one. The more carefully tuned models tend to ask when the ambiguity is high, instead of guessing blindly. When instead it guesses and gets it wrong, it's usually because the request didn't give it the footholds to understand what you really wanted.

What you can do in practice

  • Put in the request the elements that narrow the field: purpose, who it's aimed at, format, length, tone, constraints to respect.
  • Compare the two versions. Vague: "write me a text for work". Targeted: "write me a three-line email to my boss to move Thursday's meeting, courteous and direct tone".
  • If you don't feel like anticipating everything, flip the logic: end the request with "if you're missing anything to answer well, ask me before you start". You make it work for you.
  • When the answer is off-target, don't start from scratch: add the detail that was missing.

A common misconception

People think that if the AI asks for clarification it's not very capable, that an "intelligent" system should understand at once. On the contrary: asking instead of guessing blindly is often the better behavior, the same as a good professional who doesn't start until they've understood what you need. The problem isn't the clarifying question, it's the vague request that makes it necessary.

Frequently asked questions

How do I keep the AI from guessing wrong?

By giving it enough context and, if you want, explicitly inviting it to ask you what it's missing. Most off-target answers come from requests that were missing a piece, not from the AI's incapacity.

Better one long request or many short ones?

A complete, targeted request beats both the skimpy message and the river of words. It's not about writing a lot, but about putting in the right elements: what you want, for whom, in what form, with what limits.

Can I tell it to ask me things before answering?

Yes, and it's one of the most useful tricks. Add "ask me the questions you need before writing": instead of firing off an answer on invented premises, it gathers the missing details from you and then works on the concrete.

The shorter I am, the better, because the AI understands at once?

No, and it's the illusion that generates the most wrong answers. Vague brevity is the number-one cause of off-topic results: the fewer details you give, the more the AI has to fill the gaps its own way. You don't need to write an essay, you need the few data points that matter. Useful conciseness is the precise kind, not the impoverished kind.