The one idea
Do not ask the AI to solve your problem. First decide, with it, how the problem should be approached — then let it work.
This is the move nearly everyone skips. They treat the tool as a vending machine: insert question, receive answer. The people who get dramatically better results treat the first exchange as scoping, the way a good consultant does before quoting.
And the single highest-value question in that scoping is not something you tell it. It is something you ask it:
"Before you answer: what information are you missing, and what would you have to assume?"
Generic output is almost always the sound of unstated assumptions being filled in with averages. Make it state them and you find out exactly what your prompt left out — which you could not have guessed on your own, because you did not know what you had failed to say.
Tell it what you want, who it is for, and what would make it good — before asking it to produce anything. Then ask what else it needs to know.
Write a brief, not a request. The same brief you would give a competent new colleague who has never met your client and cannot ask you anything after today.
You are constraining the space of plausible continuations before generation. Official prompting guidance says to establish success criteria and a way to test against them before trying to improve a prompt; framing is that step.
The frame
Nine slots. You will not fill every one every time — for a quick rewrite, three is plenty. But when output keeps coming back generic, the missing piece is always one of these.
| Slot | The question it answers |
|---|---|
| Goal | What must be true after this is done? Not the task — the outcome |
| Context | Who is this for, what do they already know, what is the situation? |
| Inputs | What material am I providing, and what is authoritative in it? |
| Constraints | Length, tone, what must be included, what must be avoided |
| Process | How should it approach this? Steps, order, what to do first |
| Quality criteria | How will I judge the result? What separates good from acceptable? |
| Output format | Structure I can actually use — table, email, bullets, headings |
| Questions first | What is missing? What would you have to assume? |
| Verification | Which parts of this will I need to check, and how? |
The four most people skip are Goal, Quality criteria, Questions first, and Verification. Those four are most of the difference.
Goal is not the task
"Write a self-review" is a task. "Make my manager comfortable putting me forward for the senior title in this cycle, given she has three candidates and limited budget" is a goal. The second one has a shape. It tells you what to include, what to cut, and what would count as success — and it tells the AI too.
Quality criteria are what stop "fine"
If you do not say what good looks like, you get the average of everything written on the topic. Averages are always fine and never useful. "Good means: every claim is tied to a specific project with a number attached, and nothing in it could be said by a colleague on a different team."
Questions first is the lever
This one changes the interaction more than any wording trick. It moves the tool from answering to scoping, and it exposes what you left out.
You ask for a project update email to a client. Instead of producing one, it asks: is this the first update or a continuation, is the delay you mentioned already known to them, is the new date firm or provisional, and who else is copied.
You had not decided the third one. That is the actual state of your project — uncovered by a question you did not think to ask yourself, in ten seconds, before you wrote a paragraph committing to a date you cannot hold.
You ask for help structuring a dissertation chapter. Prompted for its assumptions, it says it is assuming a standard empirical structure, a word count around 8,000, and that your literature review is already written. Two of those are wrong. Had you not asked, you would have received a beautifully organised outline for somebody else's chapter and spent an evening trying to force yours into it.
Doing it
Framing a task properly
1 of 7Write the goal as an outcome, not an action. Not "summarise this report" but "give me enough to run a fifteen-minute discussion with the ops team about whether we adopt this, including what they will object to."
Try this
Here is a real request, badly framed:
"Help me prepare for my job interview."
Rewrite it as a frame. You do not need all nine slots — aim for goal, context, constraints, quality criteria, and a questions-first instruction.
Your challenge
Level 3 · IndependentTake a real task you have this week that you were going to hand to an AI in one line. Frame it instead, using at least six of the nine slots, and including the questions-first instruction.
You have succeeded when: the tool asks you at least two questions you had not considered, at least one of your answers to those questions changes what you would have asked for, and the final output needs less editing than your usual one-line attempts.
If it asks nothing useful, your frame was probably over-specified — you answered everything, including the things you had guessed at. Note which of your statements were actually assumptions.
What people usually get wrong
- Adding adjectives instead of specification. "Write a compelling, professional email" adds nothing. "Under 150 words, ends with one specific question, does not apologise" is specification.
- Giving context but no goal. Plenty of background and no statement of what success looks like produces a well-informed answer to an unclear question.
- Skipping the questions step because it feels slow. It costs one exchange and routinely saves four rounds of "no, not like that".
- Asking for questions but not saying "wait". You get questions plus a full answer that ignores them.
- Answering every question it asks. Some are noise. Answer what matters, instruct it to assume the rest, move on.
- Framing a trivial task. Nine slots to rephrase one sentence is theatre. Match the framing to the stakes.
How someone experienced does it
The strongest version of this is asking the AI to write the brief rather than the deliverable: "You are going to help me do X. Before we start, write the brief you would want if you were doing this job properly — the questions you would ask a client, the information you would need, and how you would know the result was good. Then I will fill it in."
You now get a checklist built from how this kind of work is usually specified, which routinely contains two or three items you would not have thought of. It is most valuable precisely when you are working outside your expertise and do not know what a good brief for this task contains.
The deeper reason experienced people do this: the framing is the thinking. Most of the value of the exercise arrives before any output does. When you are forced to state the goal as an outcome and name your quality criteria, you often discover you did not know what you wanted — and that was the real reason the last four attempts came back generic. The prompt was never the problem.
When not to use this
Skip the frame for small, self-evident, low-stakes tasks. Fixing grammar, converting a list to a table, explaining a term — ask directly. The frame costs more than the task is worth.
It is also the wrong tool when you genuinely do not yet know what you want. In that case start with an exploratory conversation — "what are the different ways people approach this?" — and frame properly once you know the shape of the problem. Framing a question you have not understood yet just locks in your confusion.
Prove it
Write one reusable frame for the task you hand to AI most often — the weekly report, the client email, the code review, the study summary. Save it as a text file with blanks to fill in.
Use it three times and edit it after each. What you will have at the end is worth more than any list of prompts someone else wrote, because it encodes what your work requires.
Keep learning this
Paste this into any AI assistant. It turns the assistant into a tutor that tests you instead of just answering you.
Act as an experienced practitioner who is good at teaching. I have just learned framing a task for an AI before asking it to produce anything. Assume I am intelligent but relatively new to this — treat me as beginner level. Work through this in order, and wait for my reply at each step: 1. Ask me 5 questions that test whether I actually understood framing a task for an AI before asking it to produce anything. Do not reveal the answers yet. 2. After I answer, tell me which parts I got right, which I got wrong, and which I only half-understand. Explain only what I misunderstood — do not re-teach what I already know. 3. Give me one practical challenge based on something I could genuinely encounter at work or in daily life. Do not solve it for me. 4. Evaluate my solution the way an experienced person would judge it, including what a professional would have done differently. 5. Tell me what to learn next, and why that comes next. 6. Give me trustworthy sources for deeper study — prefer official documentation, primary research or standards bodies over blogs and videos. Rules for you: no buzzwords. No motivational filler. Say "I'm not certain" when you are not certain, and tell me which parts of your answer I should verify myself. Clearly separate facts from your recommendations and your opinions.
Become independent at this
Use this when you want a path from where you are to actually good, with checkpoints you can test yourself against.
I want to become independently capable at specifying work clearly enough to delegate it — not permanently dependent on AI, tutorials or step-by-step guides. Design a progression for me with five stages: Beginner, Guided practice, Independent practice, Real-world application, Professional level. For each stage tell me: - what I must know - what I must be able to do without help - the mistakes people make at this stage - one practical challenge - one real project that would prove I reached this stage - one way I can test myself honestly Then tell me the signals that I am ready to move to the next stage, and the signals that I have skipped ahead too early. Keep the theory to the minimum I actually need. Focus on ability I can transfer to situations you and I have not discussed.