From OrwellixVibe Writing Hub is published by Orwellix, a free writing workspace with an AI assistant that edits inside your document.Try it free

Vibe Writing HubBy Orwellix
By the editorial team ·

Mega prompt 01 / Session set-up

The Vibe Writing Mega Prompt

Turns a blank chat into a working session with rules for facts, voice, turn structure, and what done means, so every later prompt is shorter and every draft needs less repair.

Jump to the prompt577 words · 11 variables · Set-up stage
01

The prompt

Copy it whole, replace every bracketed variable, then paste it as your first message.

The Vibe Writing Mega Prompt577 words
You are my writing partner for a vibe writing session. The rule for the whole conversation: I direct, you propose, I judge, and I own the result. Read everything below before you reply.

## 1. The piece

- Writer: [WHO YOU ARE]
- Writing: [WHAT YOU ARE WRITING]
- Reader: [WHO WILL READ IT]
- The piece must: [WHAT THE PIECE MUST DO]
- It will appear: [WHERE IT WILL APPEAR]
- Length: [LENGTH]

## 2. Raw material (the only source of facts)

[RAW MATERIAL]

Rules for facts:
- Use only the facts, numbers, names, quotes, and sources in the raw material or that I add later in this conversation.
- Never invent a statistic, study, quote, example person, company, date, or source. If the piece needs something that is not here, write [VERIFY: what is needed] in its place and keep going.
- If raw material is contradictory or unclear, ask me rather than choosing silently.
- If you reword a quote, label it as paraphrase. Quotes marked verbatim stay verbatim.

## 3. How I write

[VOICE NOTES]

Banned words, phrases, and moves (never use these, even in options):
[BANNED LIST]

Additional voice rules:
- Match my sentence length and rhythm, including the uneven parts. Do not smooth my writing into a uniform register.
- Keep my point of view and stance. Do not hedge claims I made confidently, and do not strengthen claims I made carefully.
- Prefer the specific detail from the raw material over a general statement that would be true of anyone.
- Do not add a tidy summary paragraph, a motivational close, or a call to action unless I ask.
- Do not use three-item lists as a reflex. Use the number of items that is actually true.

## 4. How we will work

Before drafting, ask me up to [N] questions. Ask only the questions whose answers would change the draft. If you have none, say so and proceed.

Turn rules once we are drafting:
- When I ask for a draft, give me one draft, not several, and nothing else: no preamble, no explanation after.
- When I ask for options, give me exactly three clearly different versions. Label each with the tradeoff it makes in six words or fewer.
- When I say "change only X", change only X. Show me just the sentences you changed, marked OLD and NEW, then wait.
- When I say "keep", leave that text untouched for the rest of the session.
- When I paste text and say "mine", treat it as my writing: edit for the issue I name, not for taste.
- If you disagree with a direction I give, say so in one sentence, then do what I asked anyway.
- Never report that you have checked facts, run a search, or verified anything. You have not. Mark uncertainty with [VERIFY].

## 5. What done looks like

[DEFINITION OF DONE]

When I say "final check", do not rewrite. Instead, list every claim in the draft that would need a source, every place you changed my meaning, and every sentence you think still sounds generated. One line each.

## 6. Begin

Reply in this order:
1. Two sentences stating what you understand the piece to be and who it is for.
2. One line naming the single most important thing the raw material is missing, if anything.
3. Your questions, numbered, or the words "No questions."

Do not draft yet.

Best for
  • Any piece longer than a paragraph where facts, voice, or stakes matter
  • Sessions you expect to last more than two or three turns
  • Writers who keep re-explaining the same context every time they open a chat
  • Teams that want AI-assisted drafts to start from the same rules
Not for
  • A one-line rewrite or a quick subject line. Use a stage prompt instead. The brief is overkill
  • Anything containing confidential client, patient, student, or employee data you are not allowed to paste into a third-party tool
  • Replacing research. The prompt forbids invented facts, but it cannot supply true ones you have not gathered

Most disappointing AI drafts are not a model problem. They are a briefing problem. The assistant was never told whose facts to use, how the writer sounds, how big a change is allowed per turn, or what finished looks like. So it guessed, and it guessed generic.

This mega prompt is the briefing. It is long on purpose: you paste it once at the top of a session, fill in six blocks, and then spend the rest of the conversation in short, natural turns. It is the project-level instruction that every other prompt on this site assumes.

It does not write anything by itself. Its first job is to make the assistant ask you the questions a good editor would ask before a word is drafted.

02

Variables

What to put in each bracket. Required ones are marked. Thin answers here are the usual cause of generic output.

[WHO YOU ARE]Required

One line about you as the writer, including the role the reader will see you in.

ExampleA product manager at a 40-person logistics software company, writing under my own name
[WHAT YOU ARE WRITING]Required

The format and the subject in one line.

ExampleA 900-word blog post about why we stopped running quarterly roadmap reviews
[WHO WILL READ IT]Required

The specific reader, including what they already believe and what they are busy with.

ExampleOther product managers at small B2B companies who have been burned by roadmap theatre and skim at lunch
[WHAT THE PIECE MUST DO]Required

The change you want in the reader. Not 'inform' but what they should think, feel, or do after.

ExampleMake them willing to try one monthly 25-minute review in place of the quarterly ritual
[WHERE IT WILL APPEAR]

Channel and any constraints it imposes.

ExampleThe company engineering blog; plain prose, no images, one link to our docs allowed
[LENGTH]

A target and a hard ceiling.

ExampleAbout 900 words, never more than 1,100
[RAW MATERIAL]Required

Everything the piece may draw on: your notes, numbers, quotes with attribution, links, the messy voice memo transcript. The more real material, the less generic the draft.

ExampleBullet notes from three retros; the actual before/after meeting hours (14h per quarter down to 5h); one quote from our CTO (approved); a link to our internal doc
[VOICE NOTES]Required

How you write. Paste a Voice Sheet if you have one, or three short samples of your own writing with a line about what they have in common.

ExampleShort declarative sentences, one dry joke per piece at most, I say 'we' not 'the team', I never open with a question, I use concrete numbers over adjectives
[BANNED LIST]

Words, moves, and structures you do not want. Be specific. The assistant will obey this literally.

Exampledelve, leverage, game-changer, 'in today's fast-paced world', rhetorical questions, any paragraph that starts with 'Imagine', bullet lists in the body, a closing summary
[N]

How many questions the assistant may ask before drafting. Three is a good default.

Example3
[DEFINITION OF DONE]Required

What has to be true before you would publish or send it.

ExampleEvery number traces to my notes; the CTO quote is verbatim; a stranger could repeat the one-sentence argument; it sounds like my last three posts
03

Why it is built this way

Each block does one job. Knowing the job lets you shorten or adapt the prompt without breaking it.

  1. The rule in the first line

    Direct, propose, judge, own. Stating the division of labour once makes every later instruction shorter, and gives you a sentence to point back at when the assistant starts steering.

  2. Block 1: the piece

    Six lines the assistant will otherwise guess. 'What the piece must do' matters most. It is the difference between a draft that informs and one that lands.

  3. Block 2: raw material and the fact rules

    The single biggest lever against generic output. Real notes, real numbers, and an explicit ban on invention push the model toward your material and give it a safe way to flag gaps with [VERIFY] instead of filling them.

  4. Block 3: voice notes and the banned list

    Positive description of how you write plus a literal blocklist. The 'keep the uneven parts' rule counters the model's default pull toward uniform polish.

  5. Block 4: turn rules

    This is what makes the session conversational instead of a slot machine. One draft at a time, bounded edits shown as OLD and NEW, 'keep' as a lock, and a ban on fake verification claims.

  6. Block 5: definition of done and the final check

    Gives the session an end state and turns the last turn into an audit rather than another rewrite. The audit output is a list you can act on by hand.

  7. Block 6: begin, but do not draft

    Forces a restatement and questions first. If the restatement is wrong, you have caught the misunderstanding before it cost you a draft.

04

Example run

A representative first response, with the input summarized. Your results will differ by model and material.

Input

Filled for the roadmap blog post in the variable examples above: a product manager, 900 words, notes from three retros, the 14-hours-to-5-hours figure, one approved CTO quote, a banned list that includes rhetorical questions and closing summaries, and three questions allowed.

Assistant response, excerpt
I understand this as a first-person blog post, around 900 words, in which you explain why your company replaced quarterly roadmap reviews with a short monthly review, written for product managers at small B2B companies who are skeptical of roadmap rituals and reading quickly.

The raw material is missing what the monthly review actually covers. The retro notes describe the quarterly version in detail and the monthly one only by duration.

1. What happens in the 25 minutes? Even a rough agenda would let me describe it instead of asserting it works.
2. Was there a specific quarter where the old review failed badly enough to trigger the change, or was it gradual? The notes mention "the Q3 mess" without detail.
3. The CTO quote says the old process "optimized for feeling prepared." May I use it as the hinge of the piece, or do you want it lower down?
What to notice

The restatement is accurate, the gap it names is real, and all three questions would change the draft. This is the point of the brief. Answering those three questions took the writer two minutes and produced a first draft that used the Q3 detail as its opening and needed one [VERIFY] resolved rather than a rewrite.

05

Follow-up prompts

The second turn is where vibe writing happens. These are the short messages that keep the session moving.

After answering the questions, to get the first draft

Follow-up
Answers: [YOUR ANSWERS]. Draft it. One draft, nothing else.

When the draft is close but one section is wrong

Follow-up
Change only the [SECTION OR PARAGRAPH]. The problem: [WHAT IS WRONG IN ONE SENTENCE]. Show OLD and NEW for the sentences you touch, then wait.

When you want to see alternatives for one moment

Follow-up
Options for the opening sentence only. Three clearly different versions, labeled by tradeoff. Do not touch anything else.

When the draft sounds too polished

Follow-up
Mine: [PASTE YOUR OWN ROUGH PARAGRAPH]. Keep its rhythm and stance exactly. Fix only the two sentences that are grammatically broken, and tell me which two.

When you think it is done

Follow-up
Final check.

06

When it goes wrong

Symptoms you will actually see, the usual cause, and the correction that works.

The assistant drafts immediately instead of asking questions.

Cause
Some models treat a long prompt as a request to produce. Block 6 usually prevents this, but not always.
Fix
Reply: 'You skipped step 6. Do not draft. Restate the piece and ask your questions.' Then continue. It rarely happens twice in a session.

The draft is clean, well-organized, and sounds like nobody.

Cause
Block 3 was thin. 'Friendly but professional' is not a voice note. It is the default.
Fix
Paste three paragraphs you actually wrote and say 'Mine. Describe in five lines what these have in common, then apply that to the draft.' Better: fill in the Voice Sheet first.

Numbers or examples appear that were not in your raw material.

Cause
The model filled a gap because the raw material implied a figure should exist.
Fix
Ask: 'List every number, name, and example in the draft and where each came from in the raw material.' Delete anything it cannot trace. Then add the fact rules reminder: 'Replace, do not invent. Use [VERIFY].'

Every 'change only X' turn quietly rewrites other sentences too.

Cause
The model re-generates the whole piece and the untouched parts drift.
Fix
Insist on the OLD/NEW format and paste back your kept version: 'This is the current draft. Change only X. Return only the changed sentences.' Keep your own master copy outside the chat.

The final check says everything is fine.

Cause
Self-review by the same model is weak evidence. It is grading its own work.
Fix
Treat the final check as a starting list, not a verdict. Run the Critique Your Draft prompt in a fresh session, and do the fact-check by hand against your sources.
07

Before you publish

The prompt cannot do these. You can.

  • Every number, quote, and named example traces to your raw material or a source you opened yourself. The prompt forbids invention, but it cannot enforce it.
  • Resolve every [VERIFY] tag before publishing. Do not let the assistant resolve them.
  • Read the piece aloud once. If a paragraph sounds like someone else, that paragraph is where your voice notes were weakest. Rewrite it by hand.
  • Check that the stance is yours: nothing you said carefully got strengthened, nothing you said confidently got hedged.
  • If the piece goes somewhere with a disclosure rule, such as a classroom, a client, or a publication with an AI policy, follow it. The ethics lesson covers how.
08

Questions

Is a mega prompt the opposite of vibe writing?

No. A mega prompt front-loads the context the conversation would otherwise have to rebuild turn by turn. Vibe writing is the conversation itself. This brief sets the rules once so the conversational turns can be short and natural. A mega prompt that tries to specify the entire piece up front, so that one output is final, is the thing to avoid.

Do I need to fill in every variable?

Fill in the piece, the raw material, the voice notes, and the definition of done. Those four carry most of the value. Leave the others at their defaults if you are in a hurry, but expect more correction later.

Can I use this as a system prompt or a project instruction?

Yes. Blocks 3 and 4 are stable across pieces and work well as a saved project or custom instruction. Paste blocks 1, 2, and 5 fresh for each piece.

How long should the raw material be?

As long as it needs to be to contain every fact the piece will use. A page of messy notes beats a paragraph of tidy ones. If you have nothing, you are not ready to draft. Use the Discover Writing Ideas prompt first.

Next in this libraryThe Voice Capture Mega PromptExtracts how you actually write from samples you already have, and produces a paste-ready voice profile that makes every later AI draft need less correction.