The prompt
Copy it whole, replace every bracketed variable, then paste it as your first message.
You are my drafting partner for a blog post. I direct, you propose, I judge, I own the result. Read everything, ask your questions, and do not draft until I say so. ## 1. The search job - Query this post must satisfy: [SEARCH QUERY] - What the searcher wants to get done: [WHAT THEY WANT] - My angle, the thing only I can say: [YOUR ANGLE] - Who I am: [WHO YOU ARE] - Length: [LENGTH] - Links allowed: [LINKS ALLOWED] ## 2. Raw material (the only source of facts and examples) [RAW MATERIAL] Fact rules: - Every example, number, story, and step in the post comes from the raw material. Nothing else. - Never invent a statistic, study, expert, tool, customer, or quote. Where the post needs one and I have not supplied it, write [VERIFY: what is needed] and keep going. - Never claim a link or source that is not in "Links allowed". - If the raw material contradicts my angle, say so before drafting. ## 3. Voice [VOICE PROFILE] Banned words, phrases, and structures (never use, even in options): - [BANNED LIST] - "In this post", "In this article", "Let's dive in", "Without further ado" - A section titled "Conclusion", "Final thoughts", "Key takeaways", or "Why it matters" - Rhetorical questions as openers - Three-adjective lists, "it's not about X, it's about Y" - Any paragraph that could be pasted into a different post on the same topic without change ## 4. Structure rules - The first sentence answers or directly addresses the query. No warm-up paragraph. - Every header is a claim or an instruction with content, not a topic label. "Ask about their last bookkeeper on the first call" is a header. "The first call" is not. - Use the number of sections the material supports. Do not pad to a round number. - Each section contains at least one specific detail from the raw material: a number, a step, a thing that happened. - Lists only where the content is actually a list of parallel items. Steps are a list. Reasons usually are not. - End on the last useful sentence. No summary, no "I hope this helps", no call to action unless I ask for one. ## 5. How we work Before drafting, ask me up to 3 questions whose answers would change the post. Also tell me, in one line, which part of my angle the raw material supports least. When I say "draft": give me one draft, with a title and up to three alternative titles after it, then stop. No commentary before or after. When I say "change only X": return only the changed sentences marked OLD and NEW. When I say "options for X": three clearly different versions of X only, labeled by tradeoff in six words or fewer. Never say you have checked facts, searched, or verified anything. Mark uncertainty with [VERIFY]. ## 6. Begin Reply with: two sentences restating the search job and my angle; one line on what the raw material supports least; then your questions, numbered. Do not draft.
- Best for
- Posts between 700 and 2,000 words aimed at a query people actually search
- Company blogs where the author has firsthand experience to draw on
- Solo writers who publish regularly and want a consistent starting draft
- Updating an old post with new material without losing its original voice
- Not for
- Mass-producing posts on topics you have no experience with. The prompt forbids invented examples, so it will return [VERIFY] tags where your knowledge should be
- Pure news or announcements. Use the Professional Document prompt for a release
- Personal essays where the search query is irrelevant. Use the Personal Essay prompt
The web is full of posts that were obviously assembled: a definition, five bullet points, a 'why it matters' section, a conclusion that restates the title. A reader can smell it in the first paragraph, and so can a search engine's quality systems. The posts that survive are the ones where a specific person said a specific thing from specific experience.
This mega prompt is built around that. It makes you state the search intent you are serving and the angle only you can bring, it forces the raw material to carry the examples, and it bans the structures that mark a post as generated. Headers must be claims. Examples must come from your notes. The ending is the last useful sentence, not a summary.
It assumes you have already chosen the angle. If you have not, run the Discover Writing Ideas prompt first. This one will not rescue a post that has nothing to say.
Variables
What to put in each bracket. Required ones are marked. Thin answers here are the usual cause of generic output.
[SEARCH QUERY]RequiredThe exact phrase a reader would type to find this post. One query, not a cluster.
Examplehow to onboard a new bookkeeping client
[WHAT THEY WANT]RequiredWhat the searcher is trying to get done, in their words. This is the intent you serve.
ExampleThey have just signed a client and want a checklist so they do not forget documents and look unprepared in month one
[YOUR ANGLE]RequiredThe claim only you can make, or the thing the top results get wrong. One or two sentences.
ExampleMost onboarding checklists are about documents. The real risk is expectation mismatch, and I fix it with a 20-minute 'how we will argue' call in week one
[WHO YOU ARE]RequiredYour role and why you have standing to write this.
ExampleA bookkeeper of nine years, running a two-person practice with 31 clients
[RAW MATERIAL]RequiredYour notes, numbers, stories, and the actual steps. Every example in the post will be drawn from here.
ExampleMy current onboarding checklist; the story of the client who ghosted in month two; the three questions I ask on the first call; numbers on how many clients churn in the first 90 days before and after I added the call
[VOICE PROFILE]RequiredPaste your Voice Capture profile or three lines about how you write.
ExampleShort sentences, dry, numbers before adjectives, no rhetorical questions
[LENGTH]A target and a ceiling.
ExampleAbout 1,200 words, never over 1,500
[LINKS ALLOWED]What the post may link to. Prevents invented sources.
ExampleMy own pricing page and one link to the IRS recordkeeping page I will supply
[BANNED LIST]Words and structures you do not want, beyond the defaults the prompt already bans.
Examplestreamline, 'in this post we will', bullet lists longer than four items
Why it is built this way
Each block does one job. Knowing the job lets you shorten or adapt the prompt without breaking it.
The search job block
Query, intent, and angle are separated on purpose. Most weak posts have a query but no angle, or an angle that ignores what the searcher wanted. Writing all three forces the mismatch into view before drafting.
Raw material as the only source of examples
Generated posts feel generated because the examples are generic. Tying every example to your notes is the single most effective rule in the prompt, and the [VERIFY] fallback means gaps are flagged rather than filled.
Links allowed
Assistants invent plausible sources with confidence. An explicit allow-list makes any other link an obvious error on review.
The banned structures
Most are classic tells of assembled content. The last item, 'a paragraph that could be pasted into a different post', is a test the model applies to itself and does reasonably well with.
Headers as claims
Topic-label headers invite filler beneath them. Claim headers force each section to say something, and they read better in search snippets.
Ending on the last useful sentence
The summary-conclusion is the model's strongest reflex and the reader's most skipped paragraph. Banning it by name works, but banning 'filler' does not.
Questions before drafting, plus the weakest-support line
The weakest-support line is an early warning. If the model says your central claim rests on one anecdote, you can add material or soften the angle before a draft bakes it in.
Example run
A representative first response, with the input summarized. Your results will differ by model and material.
Filled for the bookkeeping onboarding post in the variable examples: query, intent, the 'how we will argue' call angle, nine years of practice, a checklist, one ghosting story, and 90-day churn numbers before and after the call was added.
This post has to give someone who just signed a bookkeeping client a practical first-month process, and your angle is that document checklists miss the real risk, which is mismatched expectations, fixed by a short early call about how disagreements will be handled. Supported least: the claim that the call reduces churn. Your notes give 90-day churn of "about 1 in 5" before and "2 of the last 14" after, but no dates or client counts for the before period. I will write it carefully unless you add numbers. 1. Is the ghosting story one I can tell in detail, including what the client said in month one, or only in outline? 2. Do you want the actual document checklist in the post, or a link to it? Including it makes the post longer and more searchable; linking keeps the focus on the call. 3. What do you call the call internally? "How we will argue" is memorable; if you use it with clients, I would use it as the section header.
The weakest-support line caught a real problem: the before-and-after churn figures were not comparable. The writer added client counts for both periods. Question two is a real editorial fork, and the writer chose to include the checklist as a short table, which later became the section most linked by other sites.
Follow-up prompts
The second turn is where vibe writing happens. These are the short messages that keep the session moving.
After answering, to get the draft
Answers: [YOUR ANSWERS]. Draft.
When a section reads like it could be in anyone's post
The section '[HEADER]' is generic. Rewrite it using only the detail from my notes about [SPECIFIC THING]. If that detail is not enough, say so and leave a [VERIFY] instead of filling it.
When you want stronger headers
Options for the headers only: rewrite each header as a claim the section proves. Show OLD and NEW for each. Do not touch body text.
When the opening is slow
Change only the first paragraph. It must address the query in the first sentence. Options: three versions, labeled by tradeoff.
Before publishing
List every claim in the draft that a skeptical reader could challenge, with where it came from in my raw material or 'unsourced'. Then list any sentence you think sounds generated. Do not rewrite.
When it goes wrong
Symptoms you will actually see, the usual cause, and the correction that works.
The draft is accurate but reads like a help-center article.
- Cause
- The voice block was thin, or the raw material was all process and no story.
- Fix
- Add one story to the raw material, told in your own words, and say: 'Mine. Use this as the opening and keep its rhythm.' Then re-draft only the first two sections.
Headers are back to topic labels after an edit.
- Cause
- Each regeneration drifts toward the default. The header rule is the first thing lost.
- Fix
- Use the headers-only follow-up. Keep your master copy outside the chat and paste headers back as 'keep' lines.
A tool, statistic, or expert appears that is not in your notes.
- Cause
- The model filled a gap the structure implied.
- Fix
- Ask for the claim-source list in the last follow-up. Delete anything unsourced or resolve it with a real source you open yourself.
The post is the right length but half of it is throat-clearing.
- Cause
- Length targets invite padding, especially 'why this matters' preambles.
- Fix
- Cut the length target by 30 percent and say: 'Draft to this length. Remove any paragraph that does not contain a detail from the raw material.'
The assistant ends with a summary anyway.
- Cause
- The reflex is strong, and some models treat 'no conclusion' as 'no section titled Conclusion'.
- Fix
- Reply: 'Delete the last paragraph. End on the previous sentence.' Then add 'no final paragraph that restates earlier points' to your banned list for next time.
Before you publish
The prompt cannot do these. You can.
- Open the query in a private search window and read the top three results. If your draft says nothing they do not, you have a structure, not a post. Add the thing only you know.
- Check every number against your source. Then check that the comparison it implies is fair, as with the churn figures in the example.
- Resolve every [VERIFY] with a real source you opened, or cut the claim.
- Read the headers alone, top to bottom. They should form a rough argument. If they are a table of contents, rewrite them.
- Confirm every link is on your allowed list and actually resolves.
- If your blog has an AI disclosure policy, follow it. If it does not, decide what you would be comfortable saying if asked.
Questions
Will a post written this way rank?
No prompt can promise that. What this one does is remove the things that keep posts from ranking: no answer to the query, no original material, filler structure, and invented sources. Whether it ranks then depends on the query, your site, and whether anyone links to it.
Should I include the keyword in headers?
Only where it is the natural way to say the claim. The prompt asks for claim headers, and a claim about the topic usually contains the topic's words anyway. Forcing the exact phrase into every header is a tell for readers and a weak signal for search.
Can I use this for a post that is mostly a list?
Yes, if the list is genuinely the content, such as a set of steps or a comparison. Tell the assistant in the structure block that the post is a list of N items and that each item needs a detail from your material. The ban on reflexive lists is about lists that replace reasoning, not lists that are the reasoning.
How do I update an existing post with this?
Paste the existing post as raw material, mark it 'mine', add the new material below it, and say what changed in the world. Ask for a plan of which sections change before asking for any rewriting, then work section by section with 'change only'.