Turn a brief into a finished blog post, and find out what the brief was missing
A prompt that audits your brief before it writes, drafts only from what the brief contains, and lists every place it had to depart from it.
- Works in
- ChatGPT, Claude, Gemini
- You need
- A brief you have already written · The claims the post is allowed to make, with sources · Your house style rules, if you have any
- Written for
- chatgpt prompt for blog post writing
Scored by our own engine
This page, run through the audit we sell. Measured 4 August 2026.
Most blog post prompts fail on the input rather than the instructions. Hand a model a topic and it writes fluently about nothing in particular. The prompt below refuses to draft until your brief contains a question, a reader and a sourced claim list, then writes only from what the brief actually says.
A brief is a contract, not a topic
The seven item audit is the useful half of this prompt. It exists because most briefs are a title, a keyword and a word count, and everything a draft needs beyond that gets improvised by whoever writes it. When a model improvises, the improvisation looks like a fact.
Item 4 is the one that decides the outcome. Listing every claim the post may make, with where it came from, converts drafting from generation into assembly. Anything not on the list comes back as a [NOT IN BRIEF] marker, and a marker is a decision you have to make rather than a sentence you have to catch.
Item 7 is the one nobody writes and everybody needs. A brief with no out of scope line produces a post that wanders into the next three articles you planned.
Why it audits before it writes
Answer first: because the audit is cheap and the rewrite is not. Running phase one on its own takes seconds and tells you which field your briefing template keeps failing to collect.
Be honest about what this does not solve. The model still cannot tell whether your sourced claim is true, only that you supplied a source. It cannot judge whether the post is worth publishing at all, and it will happily write a well briefed post about something nobody asked. Those are judgements the brief carries, and this prompt only checks that the brief carries them.
The other thing it cannot do is supply the specific that makes the post worth reading. If your brief has no real example in it, the drafting prompt that refuses without one is the better starting point, and the brief comes second. Where the brief exists but the plan does not, work back from one topic into a set of pages first.
Then check what shipped. Run the published URL through the heading structure checker and compare the H2s against the sections you briefed, because that is the last place a well briefed post quietly loses its shape.
You are writing one blog post from a brief I have already written. The brief
is the source. You are not researching the topic and you are not improving on
the brief.
PHASE ONE. Audit the brief before you write a word.
Read my brief and say present or missing for each of these seven things:
1. The one question the post answers.
2. Who reads it and what they already know.
3. The sections it must contain, in order.
4. Every claim the post is allowed to make, and where each one came from.
5. The word count.
6. What the reader should do at the end.
7. What is deliberately out of scope.
Then give a verdict on a single line: READY, or INCOMPLETE followed by the
numbers that are missing.
If 1, 2 or 4 is missing, stop there. Do not draft. Name what to add and wait
for it. A post drafted from a brief with no question in it is a post nobody
briefed.
PHASE TWO. Only if the verdict is READY, write the post.
1. Every fact, figure, name, date, price and claim must come from the brief.
If a sentence needs something the brief does not contain, write
[NOT IN BRIEF: what is needed] and keep going. Never fill the gap yourself,
not even with something you are confident is true.
2. Follow my sections in my order. If one cannot work where it sits, write it
there anyway and raise it in phase three.
3. Answer first under every heading. One direct sentence of 40 words or fewer,
then the detail.
4. Hit the word count within ten per cent. If the brief gives you nothing to
say for a section, leave that section short and flag it. Do not pad it.
5. Write to the reader named in item 2 and cut anything they already know.
6. No closing section that restates the post. End on what to do next.
7. British spelling: optimisation, behaviour, analyse.
PHASE THREE. After the draft, give me exactly three lists and nothing else:
- Every [NOT IN BRIEF] marker, gathered into one place.
- Every point where you departed from the brief, quoting the line of the brief
you departed from, with one sentence on why.
- The three weakest sentences in the draft, quoted exactly, so I can cut them.
My brief: [PASTE THE WHOLE BRIEF, EXACTLY AS IT WAS WRITTEN]
Gaps I can fill right now: [ANSWER ANYTHING THE BRIEF LEAVES OUT, OR WRITE "nothing"]
House style: [YOUR STYLE RULES, OR WRITE "none"]What to change
Everything in square brackets is yours to replace. Nothing else needs editing.
[PASTE THE WHOLE BRIEF, EXACTLY AS IT WAS WRITTEN]- The brief unedited, including the parts you know are thin. Tidying it before you paste it defeats phase one, which exists to tell you what your briefing process keeps leaving out. If your briefs live in a template, paste the filled template rather than a summary of it.
[ANSWER ANYTHING THE BRIEF LEAVES OUT, OR WRITE "nothing"]- Your chance to answer the audit without rewriting the brief. "The reader is a marketing manager who has already tried an agency and left" or "the only figure we can source is our own 2026 pricing page" both belong here. Write "nothing" and the draft comes back full of markers, which is the honest outcome.
[YOUR STYLE RULES, OR WRITE "none"]- The rules you would otherwise fix by hand on every draft: no second person, sentence case headings, never say "solution", always spell out the product name. Five lines is plenty. This is the input that decides whether the draft arrives edited or arrives needing the same edit it needed last week.
How to run it
- 01Write the brief first, badly if necessary
This prompt has no value without one. A brief here means the question, the reader, the sections and the claims with their sources. Twenty minutes in a document beats an hour arguing with a model about a post neither of you defined.
- 02Put the sources next to the claims
Item 4 is the one that decides whether the draft is publishable. Next to each claim, write where it came from: your own data, a page you can link, a customer call. Claims with no source next to them come back as markers, and that is the prompt doing its job rather than failing.
- 03Run phase one on its own if the brief is old
Paste the brief and stop at the audit. Briefs written weeks ago by somebody who has left the team usually fail on items 2 and 7, and finding that out before a draft exists costs nothing. Fix the brief, then run the whole thing.
- 04Expect it to refuse, and do not talk it round
A missing question, reader or claim list stops the draft. The common move is to loosen the brief until the model agrees to write, which produces a post that reads fine and says nothing. Supply the missing item instead.
- 05Read phase three before you read the draft
The departures list is where the brief and the draft disagree, and one of the two is wrong. Sometimes the model was right to move a section. More often it drifted, and you have caught it in one line rather than in a second edit.
- 06Check the shipped headings against the brief
After the post is in the CMS, run it through the heading structure checker and compare the H2s with the sections in your brief. Templates, plugins and editors quietly demote headings, and an outline that arrived correct in the draft does not always survive publication.
Questions people ask
What is the best ChatGPT prompt for writing a blog post?
One that treats your brief as the only source and refuses to write without the parts that matter. The instruction doing the work is not the style guidance, it is the rule that every fact must come from the brief and that missing facts stay visible as markers rather than being filled in.
What does a content brief need before a model can write from it?
Three things are non negotiable: the single question the post answers, who is reading and what they already know, and the claims the post is allowed to make with a source for each. A word count and an outline help. Without the first three, any draft you get back is invented rather than briefed.
Why does the prompt stop instead of writing?
Because a draft built on a thin brief costs more to fix than to rewrite. Models never refuse on their own, so the stop condition has to be written in. When it triggers, it has told you something true about your briefing process, which is more useful than the draft you asked for.
How do I stop the model adding statistics I never gave it?
Ban invented evidence and give the gap somewhere to go. A rule saying no figures outside the brief, paired with a [NOT IN BRIEF] marker the model writes instead, means you either source the number or cut the sentence. Without the marker the model quietly supplies a plausible one.
Can I run this on several posts at once?
One at a time works better. Briefs pasted together bleed into each other and the drafts start sharing examples and phrasing, which is exactly the sameness the brief was meant to prevent. Run them separately and keep the audits, since the same item usually goes missing every time.
Unsubscribe in one click. We never pass your address on.