Let it ask you the questions

The move that fixes under-described tasks

Most of this course has been about supplying context. Here is the shortcut for the times you do not know what is missing:

Before you answer, ask me the five questions whose answers would most change what you produce. Ask them one at a time. Do not answer until I have replied.

This is the highest-value single line in the whole course for people who feel their prompts are never quite enough, because it moves the burden of knowing what is missing onto the party that can see the gap.

Why it works

You do not know what you left out. That is the nature of leaving things out. But the model can see the shape of the task and where it is under-determined: the audience is unspecified, the timeframe is ambiguous, there are two readings of “cost”.

Asked to produce an answer, it silently picks. Asked to ask, it surfaces the choice.

Getting good questions rather than a form

Left unqualified, you get five bland questions: what is your target audience, what is your budget, what is your timeline. Constrain it:

Ask only questions you cannot reasonably guess the answer to, and where a different answer would produce a materially different result. If you can guess, state your assumption instead of asking.

That last clause is what stops the interrogation. You get two or three real questions plus a short list of assumptions you can correct, rather than a questionnaire.

One at a time matters too. Five questions in a block get answered in a rushed paragraph. One question gets a thought.

Where it earns the most

Anything you have not done before. Writing a job advert, a legal-ish letter, a grant application, a pricing page. The model has seen thousands; you have seen one. Its questions are a checklist you do not own.

Anything where you are the bottleneck. You know the answer and cannot work out what to type. Being asked is much easier than composing a brief.

Anything with a hidden decision. Half of “help me write X” requests contain a decision nobody has made: who is responsible, what the deadline actually is, whether this is a request or an instruction. The questions expose it, which is uncomfortable and useful.

Learning. Reverse the direction and it becomes a teaching tool: “ask me questions about this topic one at a time and tell me where my answer is wrong.” That is a different course, and the mechanism is the same.

The limits

The questions come from the shape of the task, not from understanding your situation. They are a good checklist, not insight. It will ask about budget and not about the colleague who will block this.

It will not ask about what it does not know it is missing. If your industry has a convention absent from the model’s sense of the task, no question will surface it.

Some models over-ask. If a model asks eight questions about a two-line request, it is treating clarification as helpfulness. Cap it: “at most three questions, then proceed with stated assumptions.”

The compact form

For anything substantial and under-specified:

Ask me up to three questions you cannot guess, one at a time. Then state your assumptions and proceed.

Fifteen words. It converts the hardest part of prompting (knowing what you failed to say) into answering questions, which almost everybody finds easier than writing a brief from nothing.

Keep the answers

Something worth doing the second time you use this. When the interview is over, ask for the brief:

Write my answers up as a briefing paragraph I can reuse for tasks like this.

You now have a reusable brief, assembled by being asked rather than by staring at an empty box. Do that three or four times across the kinds of work you actually do and you have built the standing brief from module two without ever sitting down to write one.

That is the general shape of getting good at this: the artefacts accumulate from ordinary work if you keep them, and they do not exist at all if you do not.