How to write a content brief a writer can actually use

Most content briefs are a title, a target keyword, a word count and a list of related terms to include. Occasionally a competitor URL to “use as a reference”.

That is a specification for average content, and it is followed faithfully. The writer produces something that hits the word count, includes the terms, and closely resembles the reference. Then everyone is disappointed by an article that does exactly what the brief asked for.

A brief is not a set of constraints. It is the transfer of everything the researcher understood about a reader, into the head of the person who has to write for them. Judged that way, most briefs transfer almost nothing.

Figure 1

Known by the researcher Received by the writer

1. 01 The exact query Survives, usually flattened into a tidy noun phrase. Cost: the frustration, the assumed knowledge, what they already tried
2. 02 Search volume Survives intact. It is a number, so it travels well. The least useful thing in the brief, and the most reliably transmitted
3. 03 Who is asking, and what they already know Almost never survives. Cost: the piece opens at the wrong level and loses the reader in a paragraph
4. 04 What already ranks, and what it gets wrong Reduced to a competitor URL, if it appears at all. Cost: the writer produces another version of what is already there
5. 05 The angle only you can take Rarely written down anywhere. Cost: the piece reads like everything else, because nothing in it is new
6. 06 Word count Survives, and becomes the target. Cost: padding, which is how a brief actively makes content worse

Here is what earns its place.

1. The exact query, untranslated

Not the topic. The query, in the words people actually type, including the awkward phrasing.

“How do I stop my sourdough being so dense” is a brief. “Sourdough troubleshooting” is a topic, and it will produce a general article about sourdough that never answers anybody’s question.

The specific wording carries information — the frustration in it, the assumed knowledge, what the person has already tried. Translating it into a tidy noun phrase throws all of that away, and it is exactly the information the writer needs.

2. What the searcher already knows

One or two lines. It changes everything about how the piece opens.

“They have baked sourdough four or five times. They know what a starter is. They have already read the beginner guides and are frustrated that the standard advice hasn’t worked.”

Now the writer knows not to spend four hundred words explaining what a starter is, which is what every ranking article does, and which is the reason this person is still searching.

Getting this wrong in either direction is fatal. Pitch too basic and the reader — who is further along — leaves in the first paragraph. Pitch too advanced and you lose them just as fast. A word count cannot tell you this. A sentence about the reader can.

3. What is already ranking, and what it gets wrong

The single most valuable section, and the one most often reduced to a list of URLs.

Whoever wrote the brief should have read the top results. Write down:

What they all cover. The table stakes — the things this piece has to include just to be a credible response.

What they all miss. The shared blind spot. There is nearly always one: an assumption every page makes, a step they all skip, a question they all raise and never answer, a trade-off none of them will state honestly.

That second point is the reason the piece exists. If a writer knows that every ranking article on this query assumes you have a stand mixer, they can write the version for people who don’t — and that version will be better than all of them for a large slice of the audience.

If nobody can find a gap, the brief should not be written. That is a legitimate and useful outcome, and almost no content process is set up to produce it.

4. The angle: the thing only you can say

What does this organisation know that a competent writer with a search engine could not assemble?

• First-hand experience of doing the thing at scale
• Proprietary data — your own numbers, aggregated and anonymised
• A contrarian position you are prepared to defend
• The failure modes you have watched happen repeatedly
• The trade-off everyone else is too commercially cautious to state

This is the part that cannot be outsourced, and it is the part most briefs leave blank. If the angle box is empty, the piece will read like everything else, because there is nothing in it that is not already published somewhere.

It is also the section that matters most now that competent prose is free. A model can produce a well-structured, fluent, comprehensive article on almost any topic in seconds. What it cannot do is know what happened on your client calls last quarter. The brief is where that knowledge enters the process, or fails to.

5. The questions that must be answered

A list. Real questions, phrased the way a reader would ask them.

This replaces the word count. It is a far better specification of scope, because it describes what the piece has to do rather than how much space it should take doing it. A writer who answers eleven real questions well will land somewhere near the right length without being told, and without padding.

Word count targets invite padding, and padding is the most common way a brief actively makes content worse. Google’s guidance on creating helpful, reliable, people-first content puts the question to publishers directly, in its list of things that signal content written for search engines rather than people: “Are you writing to a particular word count because you’ve heard or read that Google has a preferred word count? (No, we don’t.)”

Source the questions from the SERP’s related questions, from your sales and support teams, from community threads, and from the search bar on your own site. The last one is consistently under-used and is the highest-signal source available, because those are people who arrived on your site and could not find something.

6. Evidence the piece needs

Where claims must be supported, say so, and say with what.

“Every performance claim needs a linked source, and at least one should be primary research rather than another blog citing another blog.”

This is what prevents the vague, unfalsifiable prose that content built purely from other content tends to produce — the sentences that sound authoritative and assert nothing.

It also protects you: a claim invented to fill a gap is a liability, and a brief that demands sourcing is the cheapest place to prevent it. This site publishes the sourcing rules it holds its own pages to for exactly that reason: it is easier to enforce a standard that is written down somewhere than one that lives in a reviewer’s head.

If there are things the piece must not claim — regulatory limits, competitor comparisons legal won’t approve, results you cannot substantiate — those belong here too. It is far cheaper to say so in the brief than in the third round of edits.

7. The one action at the end

Not a list of possible CTAs. One.

What should a convinced reader do next, and is that action proportionate to where they are in a decision? Somebody who arrived on a troubleshooting article is not ready for a demo. They might be ready to read the next thing, or to download something, or to subscribe. The site’s FOUND, TRUSTED, CHOSEN framework is one way of asking the question: a reader still working out whether they trust you is not at the stage where the choosing happens.

Mismatched CTAs are the most common reason otherwise good content converts at nothing. A brief that specifies the next step forces the question of whether one actually exists. What happens after the click is a separate problem with its own evidence, covered in the guide to local website conversion.

What to leave out

Keyword density and term lists. A writer aiming at a term count writes for the tool. If the piece genuinely answers the query, the vocabulary follows. Include the primary query for the title and heading — leave the rest.

A word count, as a target. As a rough signal of expected depth, fine. As a number to hit, it produces padding.

A full H2 structure, dictated. A rough shape is useful. A locked outline removes the writer’s judgement about how the argument should flow, and the person who reads all the source material usually has better judgement about that than the person who wrote the brief.

A competitor URL as a “reference”. It becomes the ceiling.

To put it side by side:

On the reference-URL row above: the writer will produce a slightly different version of it. If you must include a competitor link at all, include it under “what this gets wrong.”

The test

Give the brief to two competent writers who have never spoken to each other.

If they come back with recognisably the same piece — the same argument, the same structure, the same audience, differing only in prose — the brief did its job. It transferred an understanding.

If they come back with two unrelated articles that happen to share a title, the brief was a topic and a word count, and everything that mattered stayed in the researcher’s head.

A complete brief, worked

Everything above is a set of principles. Here is what they look like assembled into an actual brief, for a real query in this site’s own field, written the way it would go to a writer who has never touched the topic. Nothing in it is decorative — every field earns its place under the same test used throughout this piece: cut it, and see what breaks.

Two things about this brief are easy to miss on a first read. The reader-context field and the evidence field are doing more work than their length suggests — search intent is a business question, not a keyword attribute, and getting the reader’s position wrong by even one stage is enough to lose them in the opening paragraph regardless of how good the rest of the piece is.

And the “what this must not claim” field exists at all because this category is unusually prone to a specific kind of overclaim — worth naming plainly in the brief rather than trusting a writer to catch it unprompted.

Give this brief to two writers who have never spoken to each other, and the test above should hold: recognisably the same argument, the same five questions answered, differing mainly in how each of them writes a sentence.

Briefing for citation by ChatGPT and AI Overviews, not just ranking

A specific kind of brief has started appearing over the past year: the word count is unchanged, the keyword list is unchanged, but a new line has been added near the top — “optimise for AI Overviews” or “structure for AI citation” — usually followed by an instruction to add FAQ schema, break the content into short chunks, or produce a companion llms.txt file. Almost none of that instruction survives contact with what the platforms themselves say they need.

Google’s own documentation is direct about this. Its guidance on AI features in Search states plainly that “there are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary,” and goes further on the specific tactic most often sold as the fix: “there’s also no special schema.org structured data that you need to add.”

Its separate guide to optimising for generative AI features is equally blunt about content chunking, which shows up in a large share of “AEO” advice: “there’s no requirement to break your content into tiny pieces for AI to better understand it.” The same page disposes of the companion file as well — “you don’t need to create new machine readable files, AI text files, markup, or Markdown to appear in Google Search.” Both pages were current when checked on 9 September 2026.

That is not a loophole in the guidance. A controlled study by Ahrefs tracked 1,885 pages that added JSON-LD schema between August 2025 and March 2026, each matched against three control URLs on other domains with similar prior citation levels, and found the change made no measurable difference on ChatGPT or Google’s AI Mode.

On AI Overviews specifically it produced a small, statistically significant decline of 4.6%. Every page in the set already had 100 or more AI Overview citations before any schema was added, so the finding is about pages already inside the consideration set rather than about pages trying to enter it. Whatever is driving AI citation there, more markup is not obviously the lever — the site’s guide to local business schema sets out what schema is still genuinely worth doing on that basis.

So the honest answer to “what changes in the brief” is smaller than the industry selling AI-optimisation services would prefer: mostly nothing structural, because the structural fixes on offer are not supported by the platforms’ own documentation or by the one controlled test available on the question. What changes is a matter of emphasis, not mechanism.

Every section the piece answers needs to be able to stand on its own if it’s the only sentence an assistant lifts. That is not a new discipline invented for AI — it is the same standard this site holds every FAQ answer to, because an FAQ is read out of context by an answer engine as often as it is read on the page.

It is worth extending that same test to the body of a piece written with citation in mind: could this paragraph be quoted alone and still make sense to someone who never saw the rest of the article?

The second thing that changes is which existing brief fields matter most. Whitespark’s 2026 local search ranking factors report had a panel of local-search practitioners score 187 factors for their predicted impact on visibility in AI search responses, and the highest-scoring one is not on the page at all: presence of the business on expert-curated “best of” and similar lists. Prominence on key industry-relevant domains sits third, and the quality and authority of unstructured citations — newspaper articles, blog posts, government sites, industry associations — fourth.

The pattern is not that on-page work stopped counting. A dedicated page for each service ranks second on the same list, which is squarely inside a writer’s control.

It is that three of the top five factors are mentions somewhere other than your own site, which is closer to what this piece has already called the angle — a finding or argument distinctive enough that other people choose to reference it — than to anything a “chunk this for AI” instruction produces.

The full dataset on AI and local search has the sourcing behind this, including where each assistant’s local business data actually comes from and how quickly that changed through 2026, and the guide to local SEO for AI search works through what to do about it.

Put plainly: a brief written to be cited by an AI system and a brief written to earn a genuine citation from another publisher are, on the current evidence, asking the writer for close to the same thing. The line worth adding to a brief is not a formatting rule. It is a reminder that the angle and evidence fields were never optional, and that AI citation has raised the cost of skipping them rather than created a shortcut around them.

Briefs that fail even when they look complete

Word count as a target and an undifferentiated keyword list are the two failures this piece has already named, and they are the two most written-about failures in the whole content-brief genre. Three others show up just as often and get discussed far less, because on the page a brief carrying one of them can look entirely competent.

The brief describes a competitor’s structure instead of the actual gap. A brief for “why is my cart abandonment so high” that lists “causes, fixes, tools, case study” as the section order, because that is what the top three results all do, is not a brief — it is a transcription.

It hands the writer a structure that already encodes somebody else’s decisions about what matters and, more damagingly, what to leave out. A shared blind spot copied forward is still a blind spot. The fix is the one this piece has already argued for above: write down what the ranking pages get wrong, not what they get right, and let the structure follow the gap rather than the other way round.

Word count stands in for a depth requirement. This piece has already made the case against word count as a target; it is worth stating plainly what happens when a brief specifies one anyway.

A writer told to produce 2,500 words on a topic with 1,400 words of honest material has three ways to close the gap, and all three make the piece worse: restate the introduction as a summary near the end, explain a term the intended reader already understands, or add a section answering a question nobody searching this actually has. None of that is a writer failing to follow the brief. It is a writer following it precisely, and the brief manufacturing the padding by design.

Nobody read the SERP before writing the brief. This is the hardest of the three to spot, because a brief written this way can still contain a query, an audience note and a list of questions — everything looks present.

What gives it away is that the instructions could sit on top of almost any query in the category without changing: “cover pricing, process and results” is not specific to this business’s problem or this reader’s question — it is the shape of an article on the topic, not an answer to the question. A brief built this way can only ever specify the generic version of the piece, which on a competitive query is the version the SERP is already saturated with.

A related failure happens one step earlier, before a brief exists at all: why keyword research keeps producing the wrong content is usually not a failure of the research itself but of what survives the handover into a brief. These three failures are what that handover looks like when it goes wrong even after the research was sound.

The same gap explains why entire page types go unbriefed and stay thin: category pages are the clearest example, commissioned as templates and populated automatically, with nobody ever asking what question one of them is answering.

Briefing a subject-matter expert is a different job from briefing a generalist

The seven fields above are the same regardless of who is writing. What changes is which ones need active work from the person writing the brief, and which can be left thinner.

An expert briefed to fill in the angle field usually produces the obvious version of it. This sounds backwards, and it is one of the more consistent failure patterns in briefing someone genuinely senior. An expert asked to write down what makes their view non-obvious tends to write down what is obvious to them — because obvious-to-them and obvious-to-the-reader stopped being the same thing years before the brief was written.

The fix is not to leave the field for them to complete cold. It is to draft two or three candidate angles from things the expert has said informally — on a call, in a message, in an aside during an interview — and hand those back for the expert to confirm or correct, rather than handing them a blank field and a deadline.

The same overcorrection shows up in the reader-context field. An expert routinely overestimates what a reader already knows, because their own baseline moved years ago and rarely gets checked against a beginner’s. A brief for an expert writer should state the reader’s starting knowledge more explicitly than usual, not less, precisely because the expert is the person most likely to skip past it.

A generalist writer needs the evidence field to do the work the expert’s judgement would otherwise do. A subject-matter expert can often tell, mid-draft, when a claim sounds wrong and needs a source; a generalist covering unfamiliar ground usually can’t make that call reliably, and has two ways to fail when they can’t — write with unearned confidence, or hedge every sentence into mush. Neither is fixed by better prose instructions.

It is fixed upstream, by a brief that supplies pre-checked facts the writer can attribute rather than a general instruction to “cite sources,” and by naming, in the brief itself, who reviews the draft for accuracy before it goes further. That reviewer is very often the expert who was too close to the topic to write it well themselves — a genuinely common and entirely sensible division of labour, and one worth writing into the brief explicitly rather than assuming it will happen.

The practical rule is short: never leave the angle field thin for an expert, and never leave the evidence field thin for a generalist. Each is compensating for the failure mode the other doesn’t have.

The handoff: reviewing a draft against the brief

A brief that never gets checked against the draft it produced is a wish list, not a working document. The review step is where most of the value of a good brief either gets banked or thrown away, and it is thrown away more often through vague feedback than through a bad draft.

The test for feedback is the same as the test for the brief itself: does this note trace back to a specific line in the brief, or is it a preference showing up dressed as a fault? A note that cannot be traced to a question the brief asked, a claim it required, or an angle it specified is usually the reviewer’s taste on the day, and taste dressed as feedback is how a competent draft gets rewritten three times without getting better.

Two different situations produce feedback, and it is worth telling them apart before writing a note. One is the writer missing something the brief asked for, which is a note about the draft.

The other is the draft revealing that the brief itself was wrong — the angle didn’t survive contact with the material, or a question turned out not to matter once the research was actually written up. That is a note about the brief, and the right response is to say so plainly and fix the brief before it produces the same problem on the next piece, not to quietly patch the draft and let the brief stand uncorrected.

Whoever wrote the brief is usually the right person to review against it, for the same reason they were the right person to write it: the observations that justified the piece existing live in their head, not fully in the document, and a reviewer meeting the topic for the first time can check spelling and flow but not reliably whether the piece did the thing it was written to do.

The judgement a brief has to carry is the subject of search intent as a business question; this page is about getting it onto the page without losing it.

A brief that is checked against what it produced, with feedback specific enough to act on, is what turns one good piece into a process that keeps producing them. A brief that is filed and forgotten the moment a draft lands is a slightly better title and word count than the alternative, and not much more.

Questions

What should an SEO content brief include?

The exact query in the words people type, who is asking it and what they already know, what currently ranks and specifically what those pages get wrong, the angle only you can take, the questions the piece must answer, the evidence required, and the single action at the end. Keyword lists and word counts are the least useful parts of a brief and the most common.

How long should a content brief be?

Long enough to remove ambiguity about what the piece is for, which is usually one to two pages. A brief that is longer than the article is usually a sign that someone is writing the article inside the brief. The test is whether two competent writers given the same brief would produce recognisably the same piece — not identical prose, but the same argument and structure.

Should a content brief specify word count?

Only as a rough signal of expected depth, and it is better replaced by the list of questions the piece must answer. A word count target invites padding, which is the single most common way a brief produces worse content. Specify what has to be covered and let the length be whatever covering it properly takes.

Who should write the content brief?

Whoever did the research and read the SERP, because the most valuable parts of a brief are the observations that get lost in a handover — what the ranking pages miss, what the searcher has already decided, what the sales team hears. A brief written by someone who has not looked at the results can only ever specify a generic article.

How do you brief for AI-assisted writing?

The same way, and with more discipline, not less. The parts a model cannot supply are exactly the parts that make a piece worth publishing: your first-hand experience, your proprietary data, your opinion on a contested trade-off, and the specific gap you found in what already ranks. A brief that only contains keywords and structure will produce competent, unremarkable output from a human or a machine alike.

Does a content brief need to be written differently for content aimed at being cited by ChatGPT or AI Overviews?

Not structurally. Google's own documentation states there are no special schema requirements and no need to chunk content into small pieces for AI to understand it. What actually changes is emphasis: each section the piece answers should be able to stand alone if it's the only sentence an assistant lifts, and the angle and evidence fields already central to a good brief matter more, not less, because in Whitespark's 2026 AI-visibility scoring three of the top five factors are mentions on other people's sites rather than anything added to your own markup. On-page work still matters — a dedicated page for each service scores second overall — but what surrounds the page outweighs what is embedded in it.

Should a content brief specify the structure of the page currently ranking first?

No. Copying a competitor's section order copies whatever that page decided to leave out along with everything it included, and hands the writer someone else's blind spot as a template. A brief should specify what the ranking pages get wrong and let the structure follow that gap, not the reverse.

How does briefing a subject-matter expert differ from briefing a writer new to the topic?

An expert usually needs help filling in the angle, because what is obvious to them stopped being obvious to a reader years ago — draft candidate angles from things they've said informally and hand those back for confirmation rather than leaving the field blank. A generalist writer needs the evidence field to carry more weight, since they can't reliably judge mid-draft when a claim needs a source, and the brief should name a reviewer for accuracy before the piece goes further.

What does good feedback on a draft look like, compared with vague notes?

Good feedback traces back to a specific line in the brief — a question it asked that the draft didn't answer, or an angle it specified that the draft argued against. 'Make this punchier' isn't actionable; 'paragraph two restates the intro instead of answering question three' is. A note that can't be traced to the brief is usually the reviewer's preference on the day, not a fault in the draft.

Who should review a finished draft against the brief?

Whoever wrote the brief, wherever possible, because the observations that justified the piece existing — the gap in the SERP, the reader's starting knowledge, the angle — live in their head rather than fully in the document. A reviewer meeting the topic for the first time can check prose and flow but not reliably whether the piece did the job the brief set out.