EXTRACTIVE SUMMARY
A content brief in a semantic content network is the document that transfers structural decisions from the topical map to the writer so that no decision is made twice. Three documents are routinely mistaken for briefs: the outline, the keyword sheet, and the competitor summary, and none of them transfers a decision. The rule that separates a brief from those three is provenance: every field in a brief traces to an upstream artifact, and a field the briefer invented is a decision being made in the wrong place. Eleven fields carry the brief, and each has one source, which is either the topical map or the query network. Leaving a field empty has a specific and predictable cost rather than a general one, and the cost is paid by the network rather than by the article. The brief ends where the prose begins, and the writer holds full discretion over everything the brief does not name. A brief is checked before writing starts by verifying provenance on every field.
What a content brief is here
A content brief in a semantic content network is the document that transfers structural decisions from the topical map to the writer, so that no structural decision is made twice.
Structural decisions are the ones that affect other pages: which questions this URL answers, which angle it takes, which terms it uses, which pages it links to, and where it forwards the reader. Each of these is decided once, at map level, with the whole network visible. The brief carries those decisions to the person writing.
This is a narrower definition than the industry’s. A brief here is not a set of instructions about how to write well. It is a transfer mechanism for decisions already made, and every field in it exists because that decision would otherwise be remade at the desk by someone who can see one page.
The three documents mistaken for briefs
Three documents are routinely delivered as content briefs, and none of them transfers a decision.
The outline. A heading structure with notes beneath each heading. An outline specifies what the article says and specifies nothing about how the article relates to the other four hundred. Two outlines produced independently for adjacent attributes will overlap, because neither writer could see the other’s.
The keyword sheet. A primary term, a list of secondary terms, and a target word count. A keyword sheet is an instruction to include strings, which is the practice semantic SEO replaced. It cannot express page scope, link direction, or forward targets, because a list has no relationships in it.
The competitor summary. An analysis of the top three ranking pages with an instruction to cover what they cover and add more. This document guarantees the absence of information gain, because its explicit standard is the contents of the index. An article written to it is a better version of something the index already holds.
All three are useful documents. None is a brief, because a brief’s job is to prevent a decision from being made twice and these three make no decisions at all.
The provenance rule
Every field in a content brief traces to an upstream artifact, and a field the briefer invented is a decision being made in the wrong place.
The rule is checkable. Take any field in a brief and ask which document it came from. If the answer is the topical map or the query network, the field is valid. If the answer is that the briefer decided it while writing the brief, the decision has been made at page level and the network’s consistency depends on that person making the same judgment on every other brief.
Provenance is what makes briefs reproducible across writers and across months. Two briefers working from the same map produce compatible briefs. Two briefers working from their own judgment produce a network whose consistency degrades in proportion to how many people touched it.
The practical consequence is that a brief cannot be written before the map exists. A brief produced without a map is an outline with more fields.
The fields and their sources
Eleven fields carry a content brief, and each has exactly one upstream source.
| Field | Source | What it fixes |
| Source attribute, with its classifications | Topical map | Which map row this URL discharges |
| Contextual vector | Topical map | The single angle, preventing drift and cannibalization |
| Predicate and section | Topical map | Whether the page serves the informational or transactional predicate |
| Source query cluster | Query network, correlative signal | Page scope: which queries share this URL |
| Question inventory, ordered | Query network, annotation layer | Which question types the page answers and in what order |
| Entity vector, ordered | Topical map and query network | Which entities appear and in what sequence |
| N-gram vocabulary, locked | Query network | The exact term forms, with synonym rotation forbidden |
| Sequential forward targets | Query network, sequential signal | Which next queries the page answers or links to |
| Internal link targets | Topical map, link graph | Which pages this URL links to |
| Bridge target and justification | Topical map, path signal | Where the page forwards relevance |
| Information gain element | The publisher’s own source | What this page contains that the index does not |
Ten of the eleven come from the map or the network. The eleventh, the information gain element, is the only field that originates with the publisher, and it is the field most often left blank.
→ How to build a topical map: nine steps, two halves, four failure modes
What an empty field costs
Each empty field has a specific downstream cost, and the cost is paid by the network rather than by the article.
| Field left empty | What happens |
| Contextual vector | The writer picks an angle. Two URLs eventually pick the same one and compete for the same query |
| Source query cluster | The writer decides page scope. Scope creeps to fill the word count, absorbing tuples belonging to other URLs |
| Question inventory | The article answers questions in the order they occurred to the writer, and the extractive summary contract has nothing to inherit from |
| N-gram vocabulary | Synonym rotation appears, because varying vocabulary is what good writing training teaches. One strong lexical signal becomes three weak ones |
| Sequential forward targets | Links get placed by topical resemblance, and the reader returns to the search engine for the next question |
| Bridge target | The page terminates. Earned relevance stops where it landed |
| Information gain element | The article restates the index. This is the most expensive empty field and the least visible, because the resulting page reads perfectly well |
The pattern is consistent: an empty field is not an omission, it is a delegation. The decision still gets made, by someone with less information, at a moment when the network is not visible.
Where the brief ends and the writer begins
A brief ends where the prose begins, and the writer holds full discretion over everything the brief does not name.
The brief fixes what the page covers, in what order, with which terms, linking where. It does not fix sentence construction, paragraph length, examples, analogies, tone, or how a section is opened. Those are the writer’s, and a brief that specifies them has confused structural decisions with editorial ones.
This boundary matters commercially. Writers refuse rigid briefs when rigidity extends into prose, and they accept rigid briefs when the rigidity is confined to structure and the reason for each constraint is stated. A brief naming the n-gram lock and explaining that synonym rotation splits a lexical signal is followed. A brief listing forbidden words is resented and quietly violated.
One constraint sits on the boundary and belongs to the brief: the first sentence beneath every heading answers that heading with an extractable triple carrying an exact value. That is a retrieval requirement rather than a style preference, and it is the single instruction most likely to be overridden by a writer’s training.
→ How to lower cost of retrieval: the six stages and the five layers
How to check a brief before writing starts
Check a brief in four passes before it reaches a writer.
Pass one: provenance. Name the source artifact for every populated field. Any field whose source is the briefer’s judgment is either corrected upstream or accepted knowingly as a local decision.
Pass two: completeness. Confirm no field is empty. An intentionally empty field is recorded as intentionally empty with a reason, so that its cost is chosen rather than incurred.
Pass three: the vector. Confirm exactly one contextual vector, and confirm the question inventory is consistent with it. A definitional vector with a question inventory full of procedural questions means the vector is wrong or the cluster is.
Pass four: uniqueness. Confirm the source query cluster does not overlap another brief in the same seed cluster. Overlap at brief stage becomes cannibalization at publication and is far cheaper to resolve here.
Four passes take a few minutes per brief and prevent the two failures that are most expensive to fix after publication: cannibalization and missing information gain.
BRIDGE
Everything above describes one brief. The difficulty is not writing one, and any competent strategist produces a good brief given a good map and an afternoon.
The difficulty is producing four hundred with identical provenance, across multiple briefers, over eighteen months, while the map is being extended and people are joining and leaving. At that scale the four checking passes stop being a habit and have to become a process, the n-gram vocabulary has to live in one place rather than in each brief, and the source query clusters have to be checked against every other brief rather than against the ones a briefer remembers.
That is a content brief system rather than a brief template. It holds the vocabulary centrally, derives every field from the map by rule rather than by recall, flags cluster overlap automatically, and refuses to emit a brief with an empty information gain field, so that building topical authority to rank survives the point where more than one person is producing the content.
→ Content brief systems: provenance enforced at scale
SUPPLEMENTARY CONTENT
How long a content brief should be
Length is the wrong measure, because a brief is complete when its eleven fields are populated from their sources and not before. A brief with eleven populated fields is often shorter than a detailed outline and carries considerably more decision.
Prose instructions are the part to cut. A brief that runs long is usually long because it contains writing advice, which belongs in an onboarding document read once rather than in every brief.
Whether AI can write content briefs
AI populates the derivable fields well: source query cluster from a correlation matrix, question inventory from annotated query types, entity vector from extraction, n-gram vocabulary from frequency analysis. These are transformations of existing data and automation is more consistent at them than people are.
AI cannot populate the information gain field, because that field asks what your organization knows that the index does not, and the answer exists inside the business rather than inside the corpus. A fully generated brief is systematically empty in the one field that determines whether the article was worth writing.
Whether writers should see the topical map
Writers should see their own row and its neighbours, not the whole map. The full map invites scope negotiation on every brief, and scope was decided upstream for reasons the writer cannot evaluate from one row.
The neighbouring rows matter, because they show the writer what the adjacent pages cover, which is the most reliable way to prevent overlap that survives the checking passes.
What changes in a brief for an outer section page
Outer section briefs carry the same eleven fields with two differences. The central search intent is not required in main content, because outer pages serve the informational predicate. The bridge target is required rather than optional, because an outer page without a bridge has no reason to exist in a commercial network.
→ What is a query network: the three signals a keyword list cannot give you