All promptsAI visibility

Structure an article so an assistant can lift one passage out of it

A prompt that restructures a draft you already have into question shaped headings with a standalone answer under each, and invents no figures.

Works in
ChatGPT, Claude, Gemini
You need
A draft you have already written, headings included · The single question the article answers · One line on who is asking it
Written for
chatgpt prompt for seo article
98A

Scored by our own engine

This page, run through the audit we sell. Measured 4 August 2026.

Score your own page →

Most prompts for writing an SEO article generate the article. This one refuses to, because the property that decides whether an assistant can quote you is not in the prose, it is in the structure. The prompt below takes a draft you have already written and returns question shaped headings, a 40 word standalone answer under each, and a list of every passage that cannot survive being lifted on its own.

What actually gets lifted

A model quotes a passage, not a page. That passage has to answer the question by itself, without the sentence before it and without the heading three sections up explaining what “the second option” refers to.

That is a structural property and it is testable by hand. Take any paragraph from your draft, paste it into a blank document, and read it as a stranger. If it opens with “This is why” or “As mentioned above”, or if it contains a number whose unit lives in the previous sentence, it does not survive. Block two of the prompt does that test on every paragraph you have, which is the part that is tedious rather than difficult.

None of this is a documented ranking rule. No operator publishes one. It is an argument about what a retrieval system can and cannot use, and the reason it is worth acting on is that the fix is free: the same edit makes the page better for a reader who wanted one fact.

The constraint that keeps the article yours

Rule four is the one people delete and it is the one holding the piece together. A model restructuring your draft will meet a sentence that needs a figure, and it will supply one.

The supplied figure is always plausible. It is in the right units, it sits in a sensible range, and by the time the piece is published nobody remembers which numbers were yours. That is how a fabricated statistic ends up on a page with your name at the bottom of it.

So the prompt writes [NEEDS A FIGURE] instead, deliberately ugly, and lists every unsupported claim in a third block it is forbidden from filling. You do that work, from your own records. It is the slow part and it is the part that makes the article worth quoting rather than merely quotable.

Run the finished page through the AI content readiness check to see whether the restructure moved anything measurable. If the wider question is why an assistant names a competitor instead of you, start with the mechanism rather than with one article.

The prompt 492 words
You are an editor preparing an article to be quoted by an AI assistant, not
only ranked by a search engine. I am going to paste a draft I have already
written. Restructure it. Do not write me a new article.

Rules you must follow:

1. Every H2 becomes a question a reader would genuinely type. Not a label.
   "How much does a structural survey cost" is a question. "Pricing" is a
   label, and a label gives a retrieval system nothing to match against.
2. Under each H2, write one direct answer of 40 words or fewer as the first
   sentence. It has to make sense to somebody who has read nothing above it.
   If it needs the previous section to be understood, it has failed and you
   rewrite it before showing it to me.
3. Every figure sits in the same sentence as the thing it describes. Never
   leave a number in one sentence and its unit, currency, date or scope in
   another.
4. Use only the facts, figures and claims present in my draft. If a section
   needs a number I did not give you, write [NEEDS A FIGURE] and move on. Do
   not estimate it, do not offer a typical range, do not substitute an
   industry average.
5. Do not add sources, studies, statistics, dates or expert names that are not
   already in my draft. If a claim needs a citation, list it in block three
   rather than producing one.
6. Keep my section order unless a section cannot be understood where it sits.
   If you move one, say which and why in a single line.

Return exactly three blocks and nothing else.

BLOCK ONE, THE SKELETON. The H1, then every H2 in order with its 40 word
answer underneath it. Headings and opening answers only, no body prose. This
is the thing I am going to paste back into the draft.

BLOCK TWO, THE PASSAGES THAT FAIL. Quote each passage from my draft, exactly
as I wrote it, that cannot survive being lifted on its own. One line each on
what it depends on: the paragraph above it, an unexplained pronoun, a number
with no unit attached, or a heading that does not ask anything.

BLOCK THREE, THE UNSUPPORTED CLAIMS. Every claim in my draft that a careful
reader would want a source for, listed and left empty. Do not fill any of them
in and do not suggest where a source might be found.

Refuse rather than produce output in these two cases:
- The draft is under 300 words. Say it is too short to restructure and tell me
  what is missing.
- The draft has no headings and no natural section breaks. Say so and ask me
  what the article is meant to answer.

The one question this article answers: [THE SINGLE QUESTION]
Who is asking it: [WHO THE READER IS AND WHAT THEY ALREADY KNOW]
The draft: [PASTE THE FULL DRAFT INCLUDING EVERY HEADING]

What to change

Everything in square brackets is yours to replace. Nothing else needs editing.

[THE SINGLE QUESTION]
The one thing the article settles, written as a question. Without it the model restructures around whatever it thinks the piece is about, which on a rambling draft is usually the second section. "How long does a commercial EPC last" produces a different skeleton from "commercial EPC guide", because the second one is a title rather than a question.
[WHO THE READER IS AND WHAT THEY ALREADY KNOW]
Decides how much the 40 word answers can assume. "A landlord who has never commissioned one" and "a facilities manager who commissions twelve a year" need opposite opening sentences, and a model given neither writes for a reader who half knows everything.
[PASTE THE FULL DRAFT INCLUDING EVERY HEADING]
Your actual draft, pasted as text with the heading levels intact. Do not hand over a URL: some assistants cannot fetch it and the ones that can usually return the page with its headings flattened, which is exactly the structure this prompt is meant to judge.

How to run it

  1. 01
    Write the draft first, badly if necessary

    This prompt restructures, it does not generate. Give it a real draft with your own figures, prices, dates and examples in it, however untidy. A draft with nothing specific in it produces a skeleton full of [NEEDS A FIGURE] markers, which is an accurate report on the draft rather than a failure of the prompt.

  2. 02
    Decide the one question before you paste anything

    Say out loud the single question the article settles. If you need two sentences, you have two articles, and restructuring will not fix that. This is also the sanity check that stops you writing a second page that competes with one you already published.

  3. 03
    Run it and read block two before block one

    The failed passages are the useful output and the part everyone scrolls past. They tell you where your writing depends on the paragraph above it, which is the exact property that decides whether a passage can be lifted whole. Block one is just the tidy version of that finding.

  4. 04
    Check every [NEEDS A FIGURE] yourself

    Each marker is a place your draft asserted something without a number behind it. Fill them from your own records or delete the claim. Do not ask the model to fill them in a follow up message, which is where the constraint quietly stops applying.

  5. 05
    Paste the skeleton back into the draft, one heading at a time

    Put each 40 word answer directly under its heading, as the first thing a reader meets, then let your existing prose follow. Do not collect the answers into an FAQ block at the bottom. The answer has to sit at the boundary a model chunks on, which is the heading.

  6. 06
    Measure the result rather than trusting it

    Run the finished page through the free AI content readiness check. It scores whether sections are bounded by headings, whether answers arrive in the first line and whether an author is named. A restructure that does not move the score changed the layout and not the property that matters.

Questions people ask

What is the best ChatGPT prompt for writing an SEO article?

The most useful one edits a draft you already wrote rather than generating a new article. A model asked to write an article from a keyword produces confident generic prose with invented specifics in it. A model asked to restructure your draft, forbidden from adding facts, produces the one thing that is hard to do by hand: an honest report on which of your passages can stand alone.

Why should every H2 be a question?

Because a heading is the boundary a retrieval system chunks on, and a question shaped heading gives it something to match a user question against. A label such as "Pricing" carries no query in it. This is a structural argument rather than a documented ranking rule, and it costs nothing to adopt.

Does answer first writing make an article worse to read?

It reads blunter to the writer and usually better to the reader. The cost is real: you lose the build up, and some pieces genuinely need one. The benefit is that a reader who came for one fact gets it in the first line instead of the fourth paragraph, which is the same reason the structure is extractable.

Why does the prompt refuse to fill in missing numbers?

Because a model asked to complete a sentence that needs a figure will produce a plausible one, and a plausible figure inside your own article is indistinguishable from a real one three weeks later. The [NEEDS A FIGURE] marker is deliberately ugly so it cannot survive a proofread.

Will this get my article quoted by ChatGPT?

Nobody can promise that, and any tool that does is guessing. What structure controls is whether a passage on your page can be lifted without the paragraph above it. That is a necessary condition rather than a sufficient one, and it is one of the few parts of the problem you can edit this afternoon.

A new prompt, most days One working prompt for a real SEO or AI visibility job, what to change in it, and a worked example. No sequences, no offers dressed as newsletters.

Unsubscribe in one click. We never pass your address on.

Run your first audit
in about a minute

Free account, no card. Paste your URL and get a real, scored report of your AI and search visibility.

Measuring rankings in