Quick Answer
We write every post on this blog with a four-stage prompt: outline, interview, draft, read-back. You paste one stage at a time into ChatGPT or Claude. The interview stage is the one that matters, because it pulls your own facts out of you instead of letting the AI invent them.
- Outline. It turns your topic into your reader's questions.
- Interview. It asks you one question at a time.
- Draft. It writes from your answers only.
- Read-back. It checks each section like a search engine.
What is this prompt?
It is four blocks of text, printed in full below. We run them to draft every post on this blog, word for word. Nothing is held back.
Most AI writing prompts are one block. You hand over a topic and get back a finished post that reads like every other post on the internet, because the AI filled it with general knowledge anyone can get for free.
This one splits the job into four. Each stage stops and waits for you. That is the whole difference. The rules behind it are set out in how to write content that ranks in SEO, AEO and GEO.
How do you use it?
Paste one stage at a time. Fill in the brackets in stage one. Finish each stage before you paste the next.
- Outline. It turns your topic into the questions your reader asks, then waits for your OK.
- Interview. It asks you one question per section and waits for each answer.
- Draft. It writes from your answers only, and tells you which sections it had to fill on its own.
- Read-back. It checks each section the way a search engine reads it, and reports what can be quoted alone.
It works in ChatGPT, Claude, Gemini, or any AI chat that holds a conversation. Most of your time goes into stage two, because that is the stage where you do the talking.
Why does the interview stage matter most?
Because it is the only stage that puts something on the page the AI could not have written without you.
The interview is the part people skip, and it matters most. Ask seven questions at once and you get seven thin answers. Ask one and you get a story. Your story is the one thing no other site can be quoted for.
Answer out loud if you can, with your phone's voice typing. Speech keeps the plain words that typing edits out. The AI writes in whatever voice you feed it.
Stage one, outline
CONTEXT
I am writing a blog post. You are not writing it yet.
Topic. [what the post is about]
Reader. [who reads it, in plain words]
Goal. [what I want them to do after reading]
Angle. [how it works, what is possible, or why it pays. Pick one]
Me. [what my business does, where, how long]
Existing. [posts I already have that touch this, and what each one covers]
TASK
1. Give me the exact question a real person types into Google or asks an AI
to land on this post. Then five follow-up questions they ask right after.
2. Turn those into five to seven section headings.
CONSTRAINTS
Use the words a customer uses, not the words my industry uses.
Every heading is a question in the reader's own words.
Order the headings the way the reader thinks, not the way I would present it.
If a heading does the same job as a post I already have, drop it and tell me which
post it collided with.
Do not write any body copy.
OUTPUT
The search question. The five follow-ups. The numbered outline.
STOP
Stop after the outline. Ask me for my take on it. Do not continue until I
approve it.
CHECK
Before you send it, confirm every heading is a question, and confirm a
stranger could tell from the outline alone what the post argues.
Stage two, interview
CONTEXT
The outline is approved. Now you collect the material. I am the source.
TASK
Ask me one question per section, in outline order.
CONSTRAINTS
One question at a time. Wait for my answer before asking the next one.
Ask only for what I would know and you would not. What I have actually seen,
the number I actually have, the mistake I watched someone make, the thing
customers say that surprises people.
Do not suggest an answer. Do not offer options to pick from. Do not comment
on my previous answer beyond one line.
When I give you a number, ask me where it came from before you move on. A number
with no source does not go in the post.
If my answer is thin, ask one follow-up for a specific example, then move on.
OUTPUT
One question. Nothing else.
STOP
Stop after each question and wait.
CHECK
After the last section, list back the specific facts, numbers and examples I
gave you, each one with the source I gave for it. If a section has none, say so
now. If a number has no source, say so now.
Stage three, draft
CONTEXT
You have my answers. My answers are the content. You are the structure, not
the source.
TASK
Write the post.
CONSTRAINTS
Do not add a fact I did not give you.
If a section is thin, say it is thin. Do not fill it in.
Never invent a number, a client, a quote or a study. If you need one and I
did not give it, leave [need a real number here] in the text.
Short sentences, one idea each. Every paragraph is one to three sentences,
then a break.
A smart 12 year old reads it once, out loud, and gets it. If they would stop
on a word, use the smaller word.
Use my words from my answers, including the plain ones.
Hold the angle I named. How it works runs spare and technical. What is
possible runs short and vivid. Why it pays runs plain and concrete.
No em dashes. No colons.
None of these words. Unleash, empower, transform, journey, holistic, synergy,
AI-powered, leverage, cutting-edge, next-generation, seamless, robust,
innovative, passionate, solutions provider. If one of them fits, the sentence is too
vague. Say the concrete thing instead.
No "it is not just X, it is Y". No "whether you are a startup, an agency or a
brand". No "in today's landscape". No closing line that restates the headline.
No three item list where all three run the same shape. No "crafted with care"
and no "thoughtfully designed".
OUTPUT
1. A title in plain words, and a one sentence description under 155
characters that says what the post covers.
2. A Quick Answer heading. A 40 to 60 word paragraph that answers the title
question outright and holds one specific fact, and makes full sense on its
own. Then three to five bullets, one line each, a bold term and a plain
explanation. Three to twelve words a bullet, every bullet the same shape.
Then a horizontal rule into the body.
3. The approved sections. Each opens with a one or two sentence direct
answer, then 130 to 160 words of detail. Section one says plainly what the
thing is before it says anything else. X is, or X means. Every section
carries one specific fact of its own. A number, a threshold, a named tool, a
dated event.
4. Five to eight questions and answers. 50 to 90 words each, answer first.
STOP
Stop after the draft.
CHECK
Count the Quick Answer paragraph. State the number. Rewrite it if it is
outside 40 to 60.
Then list every section and mark it MINE if it ran on my answers, or YOURS if
it ran on your general knowledge.
Then scan the whole draft for em dashes, colons and the banned words. State a
count for each. Every count has to read zero before you send it.
Stage four, read-back
CONTEXT
The draft is written. Now read it the way a search engine reads it.
TASK
Go section by section.
CONSTRAINTS
Do not fix anything. Report only.
OUTPUT
For each section. The question it answers. Its word count, and whether that
count sits between 130 and 160. The one specific fact it carries. Whether it can
be lifted and quoted with nothing above it for context. PASS or FAIL on each.
STOP
Stop after the list.
CHECK
Count the FAILs. State the number. If it is zero, say what you checked for,
so I can tell you read it and did not skim it.
What can't the prompt do?
It gets you to a solid draft, not a finished post.
It cannot tell you whether a number you gave it is true. It will not add your internal links or check the post against pages you already have, unless you list them in stage one. It does not know your competitors, and it will not tell you the topic is already covered better elsewhere.
Read the draft yourself before you publish. The stage-three CHECK block marks every section MINE or YOURS. Any section marked YOURS is a section you have not actually said anything in. Either go back and answer it properly, or cut it.
Where this fits
The prompt is the working method. The reasoning behind each rule sits in the guide it came from.
- How to write content that ranks sets out the answer-first structure the prompt enforces.
- How AI engines choose what to cite explains why the interview stage exists.
- SEO, AEO and GEO explained covers the three results one good page earns.
Frequently asked questions
Is there an AI prompt for writing blog posts that rank?
Yes. This page publishes the exact four-stage prompt we use on our own blog. It outlines the post as the questions your reader asks, interviews you one question at a time, drafts only from your answers, then checks each section the way a search engine reads it. Paste it into ChatGPT or Claude one stage at a time. It gets you to a solid draft, and you still check the facts and links yourself.
Does this prompt work in ChatGPT or Claude?
Both, and any AI chat that holds a conversation across turns. The prompt does not use tools, file uploads, or custom instructions. It is plain text. The one thing it needs is a model that will actually stop and wait, so if your AI runs ahead and answers its own questions in stage two, tell it to ask one question and stop, then paste the stage again.
How long does one post take with this?
Most of it is stage two, and that is the part you cannot rush. You are answering five to seven questions yourself, and the fuller your answers, the less work the draft needs later. Stage one is quick. Stages three and four take as long as the model needs to write and re-read the draft. The editing pass afterwards is yours and sits outside the prompt.
Why does the prompt ask questions one at a time?
Because batched questions get thin answers. Ask someone seven things at once and they skim all seven. Ask one and they tell you the story, the number, and the mistake they watched someone make. That detail is the only part of the post an AI could not have written on its own, and it is the part a search engine or an AI answer has a reason to quote.
Will content written this way get flagged as AI content?
Google does not penalise AI-assisted content. It penalises unhelpful content, whoever wrote it. This prompt is built to avoid the second problem: it refuses to invent facts, it marks any section it filled from general knowledge, and it runs on your answers rather than its own. What comes out reads like you because the material is yours.
Can I change the prompt?
Yes, and you should. The banned-word list in stage three is our house style, not a rule of the web. Swap it for yours. The parts worth keeping are the shape: one question at a time in stage two, a hard no on invented facts in stage three, and a report-only read-back in stage four.
About &7: We are an SEO marketing and web design studio in Singapore. This prompt is the method behind our own blog, and part of how we run SEO, AEO and GEO services for clients getting found in Google and cited by AI. If you would rather the studio write and structure it for you, start a conversation.