Write a content brief a writer can execute without asking questions
A prompt that turns one target query into a brief with the angle, the outline, the evidence the writer must gather, and the lines they are not allowed to invent.
- Works in
- ChatGPT, Claude
- You need
- The target query · What the business sells · Who is writing it and what they can access
- Written for
- content brief example
Scored by our own engine
This page, run through the audit we sell. Measured 6 August 2026.
Most content briefs are a title, a word count and a list of keywords, and the writer fills the gap with whatever they can find, which is usually the same article everybody else published. The prompt below produces a brief with the angle, a drafted opening answer, an outline marked with the evidence each section needs, and an explicit list of the things the writer is not allowed to invent.
The two fields that do most of the work
The first is the angle you are deliberately not taking. Every query has a default article, and ten sites have already written it. Naming it in the brief forces a decision before anybody starts, and it is the difference between commissioning a piece and commissioning the eleventh copy of a piece.
The second is the do not invent list. Writers do not fabricate out of dishonesty, they fabricate because a section needs a number and the brief did not say where to get it. Naming the figures, dates and product behaviours that must come from a source removes the pressure that produces them.
Why the brief drafts the opening paragraph
Section 4 asks the model to write the first 40 to 60 words rather than describe them. That paragraph is the answer to the query, and it is disproportionately the part an assistant lifts when it cites a page. Leaving it to the writer means it arrives as a warm up paragraph about how important the topic has become.
Drafting it in the brief also functions as a test. If the direct answer cannot be written in sixty words, the article does not have a clear question yet, and the honest outcome is to narrow the query rather than commission the piece.
What still needs a person
Every EVIDENCE marker. The model can tell you a section needs a measurement with a date on it, and it cannot take the measurement. Check each marker against what the writer can actually access before the brief goes out, because a brief that asks for evidence nobody can gather is a brief that gets satisfied with something invented.
You are a content editor writing a brief for one article. I will give you
the target query and some context. Produce a brief detailed enough that a
competent writer who knows nothing about my business can write the piece
without coming back with questions.
The brief must contain these sections and nothing else:
1. The one sentence promise. What the reader can do after reading that they
could not do before. If you cannot write this sentence, say so and tell
me the article should not be commissioned.
2. The reader. Who they are, what they already know, and the one thing they
are sceptical about. Scepticism is the useful field, because the article
has to answer it or it will be skimmed and closed.
3. The angle, in two sentences, plus one sentence on the obvious angle you
are deliberately not taking and why. Every query has a default article
that ten sites have already written.
4. The direct answer. Draft the first paragraph, 40 to 60 words, answering
the query outright with no throat clearing. Write it, do not describe it.
5. The outline. H2s in order, each with a one line note on what it does. For
any H2 that needs evidence, mark it EVIDENCE and say what kind: a
measurement, a source with a date, a worked example with real values, a
screenshot. Mark the ones the writer must gather rather than assume.
6. Do not invent. List the specific things a writer must not make up in
this article: figures, dates, quotes, product behaviour, anything a
reader could check. Be specific to this topic rather than generic.
7. Length and format, with a reason. Do not default to a word count from
nowhere. Say what the length is driven by.
8. Internal links. Which of my pages should be linked and from which
section, using the URLs I give you. Do not invent URLs.
9. Five questions the article must answer explicitly, phrased the way a
reader would ask them.
10. The check before publishing. Three specific things to verify in the
finished draft, written as pass or fail tests rather than as advice.
Constraints:
- Do not include a keyword density target, an LSI keyword list or a list of
terms to sprinkle. They are not how this works and they produce writing
that reads assembled.
- Do not reference competitor articles you cannot see. You have no live
view of the results page. Reason about the topic instead and say that is
what you are doing.
- Do not write the article. A brief that turns into a draft is a brief the
writer will ignore.
Target query: [THE QUERY THE ARTICLE IS FOR]
What we sell: [ONE SENTENCE ON THE BUSINESS AND WHO BUYS]
Who is writing it: [THEIR LEVEL, AND WHAT DATA OR ACCOUNTS THEY CAN GET TO]
Pages we could link to: [PASTE A FEW REAL URLS WITH ONE LINE EACH, OR WRITE "none"]What to change
Everything in square brackets is yours to replace. Nothing else needs editing.
[THE QUERY THE ARTICLE IS FOR]- One query, the one the page is meant to answer. Not a topic and not three related queries. A brief written for a topic produces an outline that covers everything shallowly, which is the single most common defect in commissioned content.
[ONE SENTENCE ON THE BUSINESS AND WHO BUYS]- This drives the angle section more than anything else. The same query briefed for a tool vendor and for an agency should produce different articles, because the reader is at a different point and the proof they want is different.
[THEIR LEVEL, AND WHAT DATA OR ACCOUNTS THEY CAN GET TO]- The EVIDENCE markers are only useful if they match what the writer can actually get. A brief that asks for a measurement from a dashboard the freelancer cannot log into produces an invented measurement, which is worse than no evidence at all. Say what they have.
[PASTE A FEW REAL URLS WITH ONE LINE EACH, OR WRITE "none"]- Real URLs with a line of description each, so the model places them in the right section rather than dumping them at the end. Without this the internal links section is either empty or invented, and invented URLs get published surprisingly often.
How to run it
- 01Decide the query before you open the prompt
One query per brief. If you are tempted to list three, you have three briefs, or you have a topic that has not been narrowed enough to commission yet. This decision is the whole difference between a brief and a wish list.
- 02Collect the URLs you want linked
Copy four or five real URLs from your own site with a line each on what they are. This takes two minutes and is the difference between an internal linking section that is usable and one the writer skips.
- 03Run the prompt and read the promise sentence
If the one sentence promise is vague, the article is not worth commissioning yet. That sentence is the cheapest possible test of whether there is anything to say, and it is much cheaper to fail it here than after two thousand words have been paid for.
- 04Rewrite the do not invent list yourself
The model produces a reasonable draft of this section and you know your topic better. Add the specific figures, product behaviours and dates a writer might reach for and get wrong. This is the section that prevents the retraction.
- 05Check the evidence markers against what the writer has
Go through every EVIDENCE marker and confirm the writer can actually get that thing. Any marker they cannot satisfy becomes either a supplied fact from you or a deleted section. Do not send a brief that asks for evidence nobody can gather.
- 06Send the brief and check the outline against the draft
When the draft arrives, run the heading structure check on it and compare against the outline you commissioned. Sections that quietly disappeared are the usual reason a piece answers less than the brief promised, and they are easier to see in the heading tree than in the prose.
Questions people ask
What should a content brief include?
A one sentence promise, the reader and their scepticism, the angle and the angle being avoided, a drafted direct answer, an outline with evidence requirements marked, a list of things the writer must not invent, the internal links with their placement, the questions the piece must answer, and pass or fail checks for the finished draft. Everything else is optional.
What is a good content brief example?
A good one is judged by whether the writer came back with questions. If they did, the brief was missing the answer, and the fix is to add that field to the template permanently rather than to answer it once by email. The most commonly missing fields are the angle being deliberately avoided and what evidence the writer is expected to gather.
Should a content brief include a keyword density target?
No. Density targets and term lists produce writing that reads assembled, and they solve a matching problem that search engines stopped having a long time ago. This prompt forbids them explicitly. Brief the question the page answers and the evidence it carries, and the vocabulary follows on its own.
How long should a content brief be?
Long enough that the writer does not need to ask anything, which for most articles is one to two pages. The length that matters is not the brief but the specificity of the evidence section. A short brief with exact evidence requirements outperforms a long one full of general guidance about tone.
Can ChatGPT write the article from the brief too?
It can produce a draft, and the brief is the part worth keeping either way. A model given this brief writes a more specific article than a model given the query, because the constraints do the work. The evidence marked sections still need a person, since the model cannot open your dashboard or take a screenshot.
Unsubscribe in one click. We never pass your address on.