Build a topic cluster from reader decisions, not keyword lists

Start with unfinished reader decisions

A useful topic cluster begins with decisions people still need to make, not with a circular diagram drawn around a high-volume phrase. Choose one bounded subject and interview the existing material: what can a visitor decide after reading each page, what evidence supports that decision, and what must be understood first? Put unanswered questions in a separate backlog. This keeps a cluster from becoming a licence to publish near-duplicates. Google’s starter guidance emphasizes logical organization for people and discoverable links for crawlers; it does not prescribe a required hub shape or promise that clustering will improve rank. The working outcome here is narrower: every retained URL owns a recognizable information job and fits a route a newcomer can explain.

Evidence: Google Search Central; GOV.UK Service Manual

Capture the map before consolidation

Before editing, export the current URLs, titles, canonical targets, indexability state, and inbound internal links. That snapshot makes consolidation reversible and exposes pages that were invisible in a brainstorming session.

Evidence: Google Search Central; World Wide Web Consortium

Build a URL ownership ledger

Create an original pillar-to-question ledger with one row per real visitor question. Useful columns are: starting situation, decision sought, minimum evidence, answer owner, prerequisite page, sensible follow-up, and excluded angles. An answer owner is the single canonical page expected to settle that question. If two URLs claim the row, compare their evidence and audience rather than splitting synonyms between them. A page may own a closely related follow-up only when the same reader can complete both jobs without hunting through an overgrown article. The exclusions column matters because future editors can see why a fashionable keyword was not given another URL. Keep unknowns marked as unknown; an empty evidence cell is a research task, not a writing prompt.

The ledger is an editorial contract, not a ranking model. Its value is traceability: a reviewer can challenge who the page serves before time is spent drafting.

Evidence: Google Search Central; GOV.UK Service Manual

Draw prerequisite and continuation edges

Turn the ledger into two different link types. A prerequisite edge points to knowledge required to understand the current answer; a continuation edge offers the next reasonable decision. Label each edge with destination purpose rather than a vague invitation such as ‘learn more.’ W3C’s link-purpose guidance is useful here: readers should be able to determine what a link does from its text together with context. Google likewise documents crawlable links as ordinary anchor elements with destinations. Neither source says every related page must link to every other page. Dense cross-linking can erase hierarchy and distract people. Prefer a small number of defensible edges, link to final canonical URLs, and remove an edge when its destination does not advance the row’s stated reader job.

On the diagram, use solid arrows for prerequisites and dotted arrows for continuations. This visual distinction is the article’s practical artifact; it prevents editorial sequence from being confused with technical discovery.

Evidence: Google Search Central; Google Search Central

Run a destination-choice rehearsal

Select three representative tasks and ask someone unfamiliar with the diagram where they would begin. Do not explain the proposed categories. Record the first label chosen, any competing destination, the reason for the choice, and whether the reached page actually closes the task. GOV.UK’s usability benchmarking guidance recommends realistic tasks, consistent measures, and observation rather than a guided tour. Apply that discipline to the architecture: a successful crawl is not proof that a human understands the route. If participants repeatedly hesitate between two pages, revisit ownership or labels before adding more links. If they reach a page and still need a second page to obtain the promised answer, decide whether the job should be combined or the first page should state its narrower scope.

Keep the rehearsal small. Its purpose is to reveal a structural ambiguity, not to manufacture a statistically precise conversion claim.

Evidence: GOV.UK Service Manual; World Wide Web Consortium

Worked affiliate-research cluster

Imagine a site teaching offer due diligence. The ledger assigns ‘Is this merchant relationship legitimate enough to investigate?’ to a merchant-screening guide. ‘How does the commission get attributed?’ belongs to a tracking explainer, while ‘What evidence supports the product claim?’ belongs to a claims-verification worksheet. A general affiliate-marketing hub introduces the route but does not repeat all three answers. Its prerequisite links send newcomers to disclosure duties and basic attribution; its continuation links branch according to the decision already completed. A proposed page called ‘best affiliate verification tools’ remains in backlog because the title describes products, not an unresolved reader decision. This example contains no observed traffic result. It demonstrates how the ledger can reject a superficially relevant URL until its distinct contribution and evidence are clear.

The bounded change would be one ownership correction plus its affected links, followed by a documented review rather than a site-wide rearrangement.

Evidence: Google Search Central; Google Search Central; GOV.UK Service Manual

Freeze version one and schedule the challenge

Save the dated ledger, diagram, URL snapshot, and three rehearsal notes together. Assign an owner for broken edges and a review date tied to a plausible trigger: a new decision area, a consolidation, or repeated search and support language that the current map cannot serve. The immediate next action is to fill ten ledger rows for one subject and resolve only the clearest ownership conflict. Do not treat impressions, clicks, or crawl frequency as proof that the architecture caused an outcome; search systems change, small samples fluctuate, and navigation comprehension needs direct observation. A compact site may reasonably use a flat library instead of a prominent pillar. Stop the workshop when one coherent route can be reviewed, even if the keyword backlog is large. The artifact should make the reasoning recoverable, while later evidence remains free to overturn the first design.