EXTRACTIVE SUMMARY
A query network is the complete set of queries a subject generates, annotated with the behavioural relationships between them. A keyword list cannot tell you three things a query network can: which queries belong on one page, which page should link to which, and where a reader converts. Correlative queries solve the first problem by revealing which queries co-occur, which determines page scope. Sequential queries solve the second by revealing which query follows which, which determines internal link direction and supplementary content. Query paths solve the third by revealing the full route from first query to final action, which determines where contextual bridges are placed. Each signal governs one structural decision and populates one field in a content brief. The data for all three is partly unobtainable, and the approximations practitioners use for each should be stated rather than concealed.
What a query network is
A query network is the complete set of queries a subject generates, annotated with the behavioural relationships between those queries.
A query network contains two layers. The first layer is the queries themselves, harvested from every available demand source and annotated with entities, attributes, predicates, question types, and n-grams. The second layer is the relationships: which queries occur together, which queries follow which, and which routes users take from first search to final action.
The second layer is what makes it a network rather than a list. A set of 8,000 queries with no relationships recorded is a large keyword export. The same 8,000 queries with co-occurrence and sequence recorded is a structure that dictates how a website should be built.
The problem: what a keyword list cannot tell you
A keyword list cannot answer three structural questions, and all three must be answered before any content is written.
A keyword list cannot tell you page scope. Given 40 related queries, a list offers no basis for deciding whether they become one page, four pages, or forty. Practitioners resolve this by intuition, and the two failure modes are symmetrical: too many pages produces near-duplicates that compete with each other, and too few produces pages that answer several questions shallowly.
A keyword list cannot tell you link direction. A list is unordered by construction, so it contains no information about which page should link to which. Internal linking then gets built from site hierarchy or from whatever seems related, neither of which reflects how users actually move.
A keyword list cannot tell you where the money is. Volume identifies which queries are searched most. It does not identify which queries sit on a route that terminates in a purchase, which is a different property entirely and frequently attaches to low-volume queries.
Three structural decisions, none of them answerable from the artifact most content plans are built on. The query network exists to supply the missing data.
Correlative queries determine page scope
Correlative queries are queries that co-occur, appearing within the same search session or within the same People Also Ask cluster, and their co-occurrence determines which queries belong on one page.
The rule is direct. Queries that reliably co-occur belong on the same page or in the same tight cluster, because a user asking one is likely to want the other in the same visit. Queries that do not co-occur belong on separate pages, regardless of how similar their wording is.
Wording similarity and behavioural correlation frequently disagree, which is the whole value of the signal. “Topical map template” and “topical map cost” share most of their words and satisfy different sessions. “Topical map cost” and “how long does a topical map take” share almost no words and satisfy the same session. A tool grouping by string similarity gets both cases wrong.
The output is a co-occurrence matrix, clustered to produce the page scope of every URL in the topical map.
Sequential queries determine link direction and supplementary content
Sequential queries record the order in which users move from one query to the next, and that order determines internal link direction and what a page’s supplementary content must contain.
The rule is direct. If query B reliably follows query A, then the page serving A must either answer B or link to B before the reader returns to the search engine. A return to the search engine is a lost session, and the sequence data is what tells you which return is about to happen.
This converts internal linking from a design decision into a data-driven one. Links are placed where the sequence says the reader is about to go, at the point in the page where they are about to go there, rather than collected into a related-articles strip at the bottom where the sequence has already been abandoned.
Sequence also determines supplementary content. The questions that follow a page’s primary question are the questions the supplementary section answers, which is why supplementary content built from sequence data outperforms supplementary content built from whatever else seemed worth adding.
Query paths determine where bridges are placed
A query path is the full multi-step route from a user’s first query to their final action, and the path determines where contextual bridges are placed.
A path such as “what is semantic SEO” then “how to build a topical map” then “topical map template” then “semantic SEO consultant” is a funnel that already exists in user behaviour. The path was not designed by the publisher. It was observed, and the publisher’s job is to pave it.
Paths sort into two kinds. A path terminating in an informational query ends in understanding and is worth serving for coverage and for the historical data it accumulates. A path terminating in a transactional query is a revenue path, and every step on it must hold a page and every page must carry a bridge toward the next step.
The most common structural error this signal catches is a revenue path with a missing step. The publisher holds the entry page and the service page and nothing in between, and readers arrive, understand, and leave, because the third and fourth steps of their path were served by somebody else.
What each signal determines
Each behavioural signal governs one structural decision and populates one field in a content brief.
| Signal | What it records | Structural decision it governs | Brief field it populates |
| Correlative queries | Which queries co-occur in a session or PAA cluster | Page scope: which queries share a URL | Source query cluster |
| Sequential queries | Which query follows which | Internal link direction and supplementary content | Sequential forward targets |
| Query paths | The full route from entry query to final action | Contextual bridge placement and funnel structure | Bridge target |
The table is the reason query network research precedes topical map production rather than accompanying it. Page scope, link direction, and bridge placement are all decided by these three signals, and all three decisions are made when the map is built, not when the article is written.
Where the data comes from, and what has to be approximated
Query network data is partly unobtainable, and the honest position is to name which parts.
Query harvesting is fully obtainable. Keyword tools with full-match and related reports, autocomplete expansion across alphabet and question modifiers, People Also Ask trees expanded several levels, related searches, Search Console query reports, forum and community mining, and competitor ranking exports together produce the raw pool. Nothing here is estimated.
Correlation is approximated. True session co-occurrence data belongs to the search engine and is not published. Practitioners approximate it with People Also Ask cluster membership, SERP overlap between two queries, and Search Console pages that already earn impressions for both. These proxies are reasonable and they are proxies, and a practitioner who presents a co-occurrence matrix as observed session data is overstating what they have.
Sequence is approximated more heavily. Public data on which query follows which is thinner still. Practitioners infer sequence from People Also Ask ordering, from the logical dependency between questions, and from Search Console query sets on individual pages over time. Inference from logical dependency is the weakest of the three and the most used.
Paths are constructed rather than observed. A query path is assembled from the sequence inferences above, so it inherits their uncertainty and compounds it across four or five steps.
The practical consequence is that a query network is a well-evidenced model of user behaviour, not a recording of it. That distinction does not weaken the method, because a modelled sequence still outperforms no sequence. It does mean the sequence data should be revisited against Search Console once a property has enough history to check the model against what actually happened.
BRIDGE
Look at the determinism table again and notice what it is. It is three columns of a content brief, sourced from three datasets that most briefs are written without.
A brief lacking them is a title, a word count, and a list of keywords to include, which leaves the writer to decide page scope, link placement, and forward direction on instinct at the moment of writing. Those three decisions are structural, they are made hundreds of times across a network, and made inconsistently they produce exactly the near-duplication and orphaning that raise cost of retrieval.
A content brief system fixes them upstream. Page scope arrives from the correlation matrix, forward targets arrive from the sequence data, and the bridge target arrives from the path, so that every brief in the network makes the same structural decisions the same way and the writer is left with the writing.
→ Content brief systems: structural decisions made once, applied every time
SUPPLEMENTARY CONTENT
How many queries a query network should contain
A commercial query network starts from a raw pool in the thousands rather than the hundreds, because the behavioural signals only emerge at volume. A pool of 300 queries produces a co-occurrence matrix too sparse to cluster.
Do not deduplicate the raw pool before annotation. The frequency of near-duplicate phrasings is itself signal about which attribute users care about, and deduplication destroys it.
Query network research compared to keyword research
Keyword research produces a ranked list of phrases with volumes attached. Query network research produces the same phrases plus the relationships between them, and the relationships are what determine site structure.
The two are not alternatives, because query network research contains keyword research as its first stage. The difference is that keyword research stops where the query network’s second layer begins.
Whether Search Console alone is enough
Search Console is not sufficient on its own, because it reports only queries a property already ranks for. A property planning coverage needs the queries it does not yet rank for, which are precisely the ones Search Console cannot show.
Search Console is the strongest available source for one specific job: validating inferred sequence against observed behaviour on pages that already have history. Use it to correct the model rather than to build it.
How often a query network should be re-harvested
Re-harvest annually, or when a subject acquires new entities. The query pool changes faster than the topical map’s classifications do, because new phrasings appear continuously while prominence remains a property of the subject.
→ What is a topical map: the data structure, the schema, and the five conditions
FEATURE FIGURE
File: query-network-correlative-sequential-paths.svg viewBox: 320 × 130, matching the site’s built-in figures Animation: inherits the site animation through the lead-line, ghost, dot, bd and ring classes. No animation is defined inside the file, so the site stylesheet remains the single source of motion.
Alt text: A query network shown as connected nodes, with one highlighted path running from an entry query through three intermediate queries to a terminal query.
Caption: A query path through the network. The unhighlighted nodes are the pool; the highlighted route is the path that terminates in an action.
Cost of retrieval notes. The figure states no fact the article body does not also state in text, which keeps it consistent with the verification requirement. Colours use currentColor so the figure inherits the site theme rather than hardcoding a palette that will drift from the brand.