Entities, experts and available evidence set the cluster boundary. Keyword similarity can reveal demand, but it can't decide what the brand is qualified to publish.

Each sourced audience question receives one page job. A named expert or credible source must be ready before that page enters the plan, and a set of questions held back from the design gets used to test whether the architecture holds. Content leads and SEO managers deciding what to consolidate, retire or greenlight in a cluster that has grown beyond its approved scope.

Overlapping pages consolidate around one entity-backed cluster, and every planned page has a named expert or source lined up before it's written.

Figure building a tower of topic blocks around a central entity badge representing topical authority

Some of the 500+ brands we've worked with

See all references
  • BMW
  • Findeks
  • Yves Rocher
  • Memorial
  • İstanbul Gedik Üniversitesi
  • Country Floors
  • Bundle

Strategy and editorial owners set scope, page jobs and evidence requirements. Drafting the map and testing it is our work. Approving the architecture is theirs.

How we hold ourselves to it

  • Entities set the boundary
  • One job per page
  • Expert before expansion
  • Tested on unseen questions
  1. Set entity-led cluster boundaries

    Strategy and domain owners use the approved entity graph to define defensible clusters and name the adjacent topics that remain out of scope.

    An entity-led topic graph with explicit scope and named exclusions.

    AI assist
    Working from approved entities, products, expertise and audience needs, the cluster-mapping agent proposes candidate boundaries with explicit non-claims.
    Human gate
    Domain and strategy owners set the cluster vocabulary and excluded adjacent topics.
  2. Map questions to distinct page jobs

    Before format or target terms enter the discussion, the content strategist assigns each sourced audience question and decision stage to one primary page job.

    A question-to-page matrix, with one accountable owner per question.

    AI assist
    For each sourced audience question, the cluster-mapping agent assigns one primary page job and flags competing or missing destinations.
    Human gate
    The content strategist and audience owner confirm question provenance, intent and the accountable destination.
  3. Audit overlap, gaps, and evidence depth

    Reviewers compare intent, evidence depth and orphaned questions across the existing pages. Consolidation is usually the first remedy for genuine overlap.

    An overlap and evidence-gap register with consolidate, retain, create, or retire decisions.

    AI assist
    To expose unsupported breadth, overlap and gaps, the cluster-mapping agent compares claims, evidence depth and page roles.
    Human gate
    Domain and editorial reviewers judge the findings and separate intentional differentiation from genuine duplication.
  4. Assign experts and source requirements

    Editorial owners attach a named specialist, evidence requirement and claim boundary to every planned page before setting production priority.

    An expert and source assignment map, with readiness status per planned page.

    AI assist
    For each claim family, the cluster-mapping agent proposes named experts, source standards and escalation boundaries.
    Human gate
    Subject-matter, editorial and legal owners take responsibility for the publication limits.
  5. Validate with questions nobody wrote for

    An independent reviewer maps audience questions that weren't used in the design. Ambiguous destinations and broken paths go back into the architecture before it ships.

    A validation report from the questions we held back, showing where the architecture holds and where it doesn't.

    AI assist
    The coverage-validation agent maps representative and holdout questions to the proposed architecture and flags ambiguity or broken paths.
    Human gate
    An independent reviewer applies the original one-owner-page rule and reopens anything that fails it.

The architecture gives writers a sequence they can use. Its coverage measure shows which audience questions already have named expertise or adequate source evidence.

  • Topical cluster architecture

    A defensible hub-and-supporting-page model grounded in entities, audience decisions, and named experts.

    Accepted when

    Which clusters the brand can credibly own, and which adjacent demand stays explicitly out of scope.

    Cadence: Experts named

  • Consolidation and creation roadmap

    Sequences consolidation, retirement, evidence development, and new page creation by documented user need and readiness.

    Accepted when

    What gets merged or retired before anything new gets written.

    Cadence: Readiness sequenced

  • Internal-link and hierarchy specification

    Hub relationships, contextual links, and next-question paths defined for each page role.

    Accepted when

    A linking plan built around what the reader needs to see next on each page.

    Cadence: User paths mapped

  • Supported cluster coverage

    The count runs against the agreed set of cluster questions. A question only counts as covered once adequate evidence or a named expert supports its page.

    Accepted when

    The share of cluster questions actually backed by adequate expert or source evidence (a published page without support doesn't count as covered).

    Cadence: Evidence counted

  • Cluster recheck schedule

    A cluster that was defensible a year ago doesn't stay assumed defensible without a recheck, and the schedule fixes when that recheck happens.

    Accepted when

    A written schedule naming what forces a recheck, from a material product, audience-question or evidence change down to the lighter overlap checks that follow a content release.

    Cadence: Change-triggered

Fifty related pages exist. Three compete for the same question, while none traces back to an entity or expert the brand can defend.

A good fit when

  • Pages overlap without anyone noticing — Two articles answer the same buyer question and split authority, yet both remain because each still gets traffic.
  • The cluster has no approved entity anchor — The pages have no approved product, service or documented expertise anchoring their shared scope.
  • Nobody owns the claims being made — A comparison page makes a technical claim with no named author, no cited source, and no specialist who'd defend it if challenged.
  • An approved entity graph already exists — The method needs agreed entities and relationships. If the graph is still open, Entity Relationship Mapping comes first.
  • Approved entities draw the cluster boundary — Products, services and relationships set scope, and the plan names adjacent demand that falls outside it.
  • Each page gets one distinct job — One page owns each question and journey stage. Competing pages merge so authority stops splitting across them.
  • Expertise is ready before publication — A named specialist, first-party data or credible source is assigned before production, so claims publish with support.

Better handled as other work when

  • You want authority from volume alone — A cluster is an editorial model, not a shortcut. More pages on topics the brand can't support only add liability.
  • Page volume is not evidence of authority — A large cluster without distinct expertise is not success. It is the failure this method is meant to prevent.
  • Rankings and citations stay outside the guarantee — The architecture consolidates supported expertise, but an external system decides whether to grant authority.
  • Screaming Frog

    surfaces the existing pages already competing for the same topic space

  • AlsoAsked

    maps the sourced audience-question tree the cluster's page jobs get assigned from

  • Airtable

    the cluster map itself, showing editors what to merge, write, and hold back

Choose one disputed question from the cluster. We map the entity boundary, decide the primary page job with your owners and test the result against questions we held back from the design.
Map the disputed topic

No. It makes the brand's supported expertise easier to follow, but external systems still decide what to rank or cite.