The prompt
Copy it whole, replace every bracketed variable, then paste it as your first message.
You are helping me find what to write. You are not writing it. I direct, you propose, I judge. Do not draft anything in this conversation. ## 1. Where I am starting - Who I am: [WHO YOU ARE] - What I might write: [FORMAT AND LENGTH] - Who would read it: [WHO WILL READ IT] - What I have already seen written on this: [WHAT ALREADY EXISTS] ## 2. The interest, or the notes [INTEREST OR NOTES] ## 3. First, ask what I know firsthand Before you propose anything, ask me up to [N] questions, numbered, about what I have seen, done, measured, argued over, or been wrong about on this subject. Ask only questions whose answers would change which angles are possible. Do not ask what I want to write about; that is what we are finding out. If the notes already answer a question, do not ask it. Then wait for my answers. ## 4. Then propose angles After I answer, propose three to five angles. An angle is a claim an informed reader could disagree with, resting mostly on material I actually have. For each angle give: - The claim, as one sentence in plain words. Not a title. - The material that supports it: quote or point to the lines in my notes or my answers. - Who would push back, and their best objection, in one line. - What I would need to know or verify before I could write it honestly. Mark each item [VERIFY: what is needed]. - Rough split: how much of the piece would be my own material versus research I have not done yet. ## 5. Rules - Do not invent facts, figures, examples, or sources to make an angle look stronger. If an angle needs something I do not have, it goes in the [VERIFY] line. - Do not hand back the angles I said already exist unless you can show what my material adds to them. - Do not choose for me. After the angles, write one line naming which angle rests most on firsthand material, then stop. - No summary, no encouragement, no offer to draft. - Never say you have researched or checked anything. You have not.
- Best for
- A subject you keep circling but cannot turn into a piece
- A folder of notes, voice memos, or half-finished drafts that needs a spine
- Work writing where you have the experience but not yet the argument: a post, a talk, a memo, a newsletter issue
- Writers who tend to pick the safe, general version of a topic and regret it halfway through drafting
- Not for
- Finding a topic you know nothing about. The prompt is built to surface what you already hold. With no material, it can only generate categories
- Fiction premises. It can be bent to that, but the Short Story mega prompt asks the right questions for story
- Keyword research or content calendars. This is one piece at a time, and the point is that you could argue it
The weakest moment in most AI-assisted writing is the first one. You type 'give me ten blog post ideas about X' and get ten titles that anyone could have written and nobody needs. They are not ideas. They are categories with a verb in front.
An angle is different. An angle is a claim an informed person could disagree with, resting on something you have actually seen, done, measured, or been wrong about. You cannot get one by asking the assistant what is interesting about a subject. You get one by making the assistant ask you what you know, and only then letting it propose.
This prompt does that in two turns. First it interrogates you, briefly, about firsthand material. Then it proposes angles, each with the evidence you already hold, the person who would push back, and a plain list of what you would need to know or verify before you could write it honestly. It never drafts. It ends with one line about which angle leans hardest on your own material, and then it stops so you can choose.
Variables
What to put in each bracket. Required ones are marked. Thin answers here are the usual cause of generic output.
[WHO YOU ARE]RequiredYour role and the position you would be writing from. The angles depend on what you have standing to say.
ExampleAn independent bookkeeper who has worked with eleven small restaurants over nine years
[FORMAT AND LENGTH]What the piece might be, even if you are not sure yet. It changes how big an angle can be.
ExampleA 1,000 to 1,400 word post for my own site, or possibly a newsletter issue
[WHO WILL READ IT]RequiredThe reader, including what they already believe about the subject.
ExampleOwners of one- or two-location restaurants who think their cash problems are about sales volume
[WHAT ALREADY EXISTS]What you have already read or seen on this subject, so the assistant does not propose it back to you.
ExampleEvery piece I have seen is either 'know your food cost percentage' or 'build a six-month reserve', and neither matches what I see in the books
[INTEREST OR NOTES]RequiredThe vague interest in a sentence, or the pile of notes pasted as they are. Mess is fine. Include the half-thoughts.
Example'Something about restaurant cash flow.' Plus forty lines of notes: which clients hit a crunch in February, the two who did not, a line about a supplier who switched to weekly billing, three things I told owners that turned out to be wrong
[N]How many questions the assistant may ask before proposing angles. Four is a good default. The questions are the point.
Example4
Why it is built this way
Each block does one job. Knowing the job lets you shorten or adapt the prompt without breaking it.
'You are not writing it' in the first line
Fixes the mode. Without it, most assistants will have drafted an opening paragraph by the second turn, and once a draft exists it anchors everything after it. The discover stage is about choosing. Nothing gets written.
Block 1, especially 'what already exists'
The cheapest way to stop the assistant proposing the generic version of your subject. If you name the two articles everyone has already written, the proposals have to beat them, or show what your material adds.
Block 3: questions about firsthand material
The hinge of the prompt. The questions are pointed at experience (seen, measured, argued over, been wrong about), not at preference. 'Do not ask what I want to write about' closes the dodge the assistant otherwise takes, which is to ask you to do the discovering.
Block 4: the five-part angle
Claim as a sentence, not a title, so you can tell whether it says anything. Evidence quoted from your notes, so a hollow angle is visible. An objector, so it is arguable. A [VERIFY] line, which becomes your research list. A rough split, which is the cost of writing it.
Block 5: 'do not choose for me' and the one closing line
Keeps the decision human while giving you a tie-breaker that is about material rather than about what sounds exciting. The ban on encouragement matters more than it looks: a warm recommendation is a thumb on the scale.
Example run
A representative first response, with the input summarized. Your results will differ by model and material.
Filled for the bookkeeper in the variable examples: eleven restaurant clients over nine years, a reader who blames cash problems on sales volume, two existing articles she wants to avoid repeating, forty lines of messy notes, and four questions allowed.
Four questions before I propose anything: 1. Of the eleven restaurants, how many hit a cash crunch last February, and did the ones that did not do anything differently in the fall, or were they simply better capitalized? 2. Your notes mention a supplier who moved one client to weekly billing. Did that help or hurt her cash position over the following quarter, and do you have the numbers? 3. You list three things you told owners that turned out to be wrong. What were they, and what did you believe at the time that made them seem right? 4. When an owner calls you in a panic, what is the first number you look at, and is it the number they expected you to look at? (After the writer answered: seven of eleven hit the crunch; the two that did not had both prepaid January rent in December; the weekly billing cut one owner's payable balance by about $9,000 but made her weekly swings worse; the three wrong beliefs included "a reserve fixes this"; the first number she looks at is the payables aging, not the sales report.) Angle 2. The February crunch is a timing problem, not a sales problem, and the fix happens in December. - Material: seven of eleven clients hit it; the two that did not had prepaid January rent in December; your line "nobody's sales fell off a cliff, their bills all landed at once." - Pushback: an owner whose January sales really did drop will say timing is a luxury you can only manage when revenue is fine. Fair, and the piece has to answer it. - [VERIFY: whether the two prepaying owners did it deliberately or by accident; you said you were not sure.] [VERIFY: the share of annual payables that lands in January and February for at least one client, from the books, before you call it "all at once".] - Split: roughly four fifths your material, one fifth checking your own numbers. Angle 4. The advice I gave for six years, "build a reserve", was wrong for most of my clients, and here is what replaced it. - Material: your three wrong beliefs, the reserve one in particular; the owner who built a $15,000 reserve and still ran short because it sat in the same account as operating cash. - Pushback: anyone who has been saved by a reserve. You are arguing against standard advice, so you need to be exact about who it fails and who it does not. - [VERIFY: how many of your clients actually built a reserve when told to; your notes say "a few".] - Split: almost entirely your material. The risk is confidence, not research. The angle that rests most on firsthand material is angle 4. It is also the one only you could write.
The questions went to experience, not preference, and every one of them changed which angles were possible. Angle 2's [VERIFY] list became the writer's research list for an afternoon. She chose angle 4, not because the assistant's closing line pointed at it, but because the pushback line named the objection she most wanted to answer. The closing line is a tie-breaker about material. The choice stayed hers.
Follow-up prompts
The second turn is where vibe writing happens. These are the short messages that keep the session moving.
When one angle is close but the reader is wrong for it
Angle 2 again, but the reader is [NEW READER]. Same five parts. Do not touch the other angles.
When two angles overlap
Angles 1 and 3 are the same piece. Merge them into one claim and show me which material the merged version drops.
When every angle feels safe
These are all defensible. Give me the two angles you held back because they were riskier, with the same five parts.
When you have chosen
Angle 4. Lock it. I will bring notes for the outline next. Do not outline yet.
When it goes wrong
Symptoms you will actually see, the usual cause, and the correction that works.
It returns titles ('The Hidden Cost of February') instead of claims.
- Cause
- Title-shaped output is the assistant's default for anything called an idea.
- Fix
- Reply: 'Rewrite each angle as a sentence I could say to a client, one they could disagree with.' Titles come later, if at all.
It skips the questions and proposes angles immediately.
- Cause
- A long prompt with notes pasted in reads as a request to produce.
- Fix
- Reply: 'You skipped section 3. Ask your questions first, then wait.' It rarely happens twice.
The [VERIFY] lines are longer than the material lines on every angle.
- Cause
- Either your notes were thin or the assistant reached for the general version of the subject.
- Fix
- Ask: 'Which of these could I write this week from what I gave you alone? Drop the rest.' If none survive, you are not ready. Go gather material.
A statistic appears under 'material' that is not in your notes.
- Cause
- The slot demanded support and the assistant supplied some.
- Fix
- Reply: 'Every line under material must quote my notes or my answers. Remove anything that does not.' Then reread the angle without it.
It recommends one angle warmly and calls it exciting.
- Cause
- Assistants close with encouragement unless told not to, and sometimes even then.
- Fix
- Ignore the warmth. Reread the pushback lines and the split lines, and choose the angle whose objection you most want to answer.
Before you publish
The prompt cannot do these. You can.
- Check each 'material' line against your notes. If you cannot find the line it points to, the angle is weaker than it looks.
- Copy the [VERIFY] items into a research list outside the chat before you move on. They are the honest inventory of what you do not yet know.
- Do not pick the first angle because it came first. Read the pushback lines. The angle whose objection you most want to answer is usually the one.
- Say the chosen claim aloud in your own words. If you cannot, it is not yours yet.
- An angle that only you could write is worth more than an angle that is easier to write. The assistant cannot weigh that. You can.
Questions
What if I have no notes, just a topic?
Paste the topic as the interest and answer the questions as carefully as you can. Your answers become the notes. If you find you cannot answer them, you do not have a piece yet. You have a subject to go and learn something about, and the questions tell you what.
Why does the prompt make the assistant ask questions before proposing anything?
Because angles come from what you know, and the assistant cannot see that unless it asks. The questions also keep the first thing on the screen from being a generic list you then anchor to. Two minutes of answering saves a round of generic proposals.
How is this different from asking for ten ideas?
Ten ideas are categories. This prompt returns claims, each with the evidence you hold, the person who would object, and the list of what you would need to verify. Fewer, and each one is something you could actually argue.
Can I use it for fiction?
Partly. It will surface what you know firsthand, which good fiction draws on. But premise, character, and shape are different questions, and the Short Story mega prompt asks them properly.