Google patented topic authority in 2012, shipped “things, not strings” the same year, and now scores whether your page adds information or repeats it. Ranking for a topic stopped being about one great page a decade ago — this module is the architecture: pillars, spokes, coverage matrices, and the discipline to finish what you start.
Google's algorithm has spent a decade moving from keyword matching to topic understanding. The Hummingbird update (2013) introduced entity-based search; RankBrain (2015) added machine-learned query interpretation; BERT (2019) added natural-language understanding; MUM (2021) added multimodal cross-language understanding. Each step further rewards sites that cover topics comprehensively rather than sites that target individual keywords.
Topical authority is the systematic outcome of that shift. A site that covers a topic from many angles — pillar content, subtopic content, related-topic content, internal linking that connects them — signals to Google that it's a knowledgeable source on that topic. The rewards: better rankings on competitive head terms, broader long-tail coverage, easier ranking for new content in the topic area, more featured snippet opportunities.
Most SEO programs under-invest because topical authority compounds slowly. The first 10 pieces in a cluster don't move the needle much. The 50th piece can suddenly unlock cluster-wide ranking lift as Google's topic model recognizes the breadth and depth.
By the numbers Google has been engineering “topics” for over a decade
The paper trail: topics, entities, and information gain are patented machinery
2012
Google’s “determining topic authority” patent (US8458196B1) — author expertise scored per topic, thirteen years ago.
500B
facts about 5 billion entities in the Knowledge Graph (Google, 2020). Topics are entities now, not strings.
2022
“Contextual estimation of link information gain” granted (US11354342B2) — new-information content scored above repeated information.
2023
Google publicly documents “topic authority” as a news ranking system. The concept graduated from patent to product.
For each new document, an information gain score is determined that is indicative of whether the new document includes information that was not contained in documents the user has already viewed.
Google patent US11354342B2, “Contextual estimation of link information gain” — the case against the copycat post, in Google’s own words — Google Patents
From keywords to topics to entities
The old model: target a keyword per page. Build pages keyword by keyword. Each page competes individually.
The modern model: target a topic with a cluster of pages. Each page covers a sub-aspect. The cluster competes collectively.
The leading-edge model: target an entity (a person, place, thing, concept) and its relationships. Build content that maps the entity's ecosystem — what it is, what it relates to, how it works, who's involved, what it's used for. Google's Knowledge Graph and entity-based ranking systems reward this approach because they were built around entities, not keywords.
Interactive timeline From strings to things: how Google learned topics — tap a year
Each step made covering a topic matter more than repeating a keyword
Knowledge Graph · “things, not strings”
Google ships an entity database alongside the keyword index. Pages stop being bags of words and start being statements about entities — the foundation everything below builds on.
Hummingbird · the whole-query rewrite
Google rebuilds its core engine to parse meaning rather than match terms. Long, conversational queries map to topics; exact-match keyword pages begin their long decline.
RankBrain · the unseen-query solver
Machine learning starts interpreting the ~15% of daily queries Google has never seen, by mapping them to known concepts. Coverage of a concept beats ownership of a string.
BERT · context cracks open
Bidirectional language models read queries the way humans do — prepositions and word order finally matter. Google calls it the biggest leap in five years.
MUM · multitask, multimodal
A model Google describes as 1,000× more powerful than BERT, designed to answer complex tasks by stitching together knowledge across pages, languages, and media.
Topic authority, in public
Google documents topic authority as a live system for news: publications known for a beat get surfaced for that beat. The patent-to-product arc completes.
AI Overviews · synthesis era
Answers assembled from many sources reward the sites whose clusters supply the facts. Being THE reference on a narrow topic now earns citations, not just clicks.
We’ve been working on an intelligent model — in geek speak, a “graph” — that understands real-world entities and their relationships to one another: things, not strings.
Amit Singhal, SVP of Google Search, announcing the Knowledge Graph (2012) — Google blog
Pillar-cluster architecture
The most common operational pattern. The structure:
Pillar page. Comprehensive coverage of a broad topic. Often 3,000–10,000 words. Targets a head term ("email marketing").
Cluster pages. Detailed coverage of specific subtopics. 1,500–3,500 words each. Targets long-tail or mid-tail terms ("email subject line best practices," "welcome email sequences," "cart abandonment emails").
Internal links. Pillar links out to cluster pages. Cluster pages link back to pillar. Cluster pages link to other clusters where topically relevant.
The math of pillar-cluster authority
A pillar with 30 cluster pages, each averaging 1,500 words = ~50,000 words on a topic. Hub-and-spoke internal linking gives the pillar PageRank-equivalent authority from 30 internal references. Search engines see this as a serious topic site. Compare to a competitor with 5 disconnected articles totaling 12,000 words — they're reading themselves out of consideration.
Interactive map Anatomy of a working cluster — click any node
The hub ranks because the spokes exist; the spokes rank because the hub binds them
The pillar
Broad, evergreen, targets the head term (“email marketing”). It links DOWN to every spoke with descriptive anchors and earns most of the cluster’s external links. It wins precisely because it does not try to answer every subtopic in depth — the spokes do.
Two link directions do the work: pillar→spoke passes authority and context down; spoke→pillar concentrates relevance up. The dashed spoke-to-spoke links handle adjacent intents. A “cluster” without these links is just a category of posts.
RGM EXPERT TRICK
Map clusters from your own impressions, not a keyword tool
Before planning any cluster, we export every query the site already gets impressions for — including position 30-90 stragglers. Those stragglers are Google saying “I considered you for this.”
Subtopics where impressions already exist get built first: Google has pre-approved the association, so new spokes index faster and rank sooner. Cold topics need the full S-curve.
A keyword tool tells you what is searched. Your GSC export tells you what Google already believes you might be an answer for. We expand from belief, not from volume.
WHY IT’S RARE · Keyword tools are where everyone starts, so everyone builds the same clusters. Impression mining builds the cluster only your site can shortcut.
Topic mapping: planning a cluster
The process:
Define the topic. Pick a broad subject your business serves. Should be commercially valuable and aligned with what you can credibly cover.
Identify the pillar term. The broadest commercial keyword ("CRM software," "running shoes," "tax preparation").
Pull the topic graph. Use Ahrefs, SEMrush, Moz, or AlsoAsked to find related queries. Add SERP feature analysis (People Also Ask). Add Reddit and Quora for human-language variations. Add competitor sitemap mining for what they cover.
Categorize subtopics. Group queries into themes. Each theme becomes a cluster page candidate. Aim for 10–50 cluster pages per pillar.
Prioritize. Score each subtopic by search volume, commercial intent, ranking difficulty, and strategic importance. Start with quick wins (low difficulty + meaningful volume) and high-priority head subtopics.
Map internal linking. Plan the pillar ↔ cluster ↔ cluster link graph before writing. Identify which cluster pages should link to which others.
Plan content depth. What angle, format, depth for each cluster page? What information gain over current SERP?
Build production calendar. Pillars take longer than clusters. Stage releases — pillar + first 5 clusters at launch, then clusters at sustainable cadence.
Planning grid The coverage matrix: where authority gaps hide in plain sight
Map subtopics × intent before writing — the colors are the roadmap
Learn (what/why)
Do (how-to)
Compare (best/vs)
Solve (fix/trouble)
Deliverability
COVERED
COVERED
GAP
THIN
List building
COVERED
THIN
GAP
n/a
Subject lines
COVERED
COVERED
COVERED
n/a
Automation
THIN
GAP
GAP
GAP
Segmentation
COVERED
GAP
n/a
THIN
Read it like a heatmap: a column of gaps (Compare) means an intent you never serve — competitors own your shortlists. A row of gaps (Automation) means a subtopic Google has no reason to trust you on. Fill rows before columns: depth on one subtopic beats a thin layer over all of them.
Internal linking patterns for topical authority
Pillar-out, cluster-in. The pillar links to every cluster page with descriptive anchor text matching the cluster's topic. Each cluster links back to the pillar.
Cluster-to-cluster contextual links. Where two cluster pages relate, link them. "Email subject lines" cluster links to "email open rates" cluster naturally.
Topical anchor text. Anchor text should describe the destination's topic, not generic phrases. "Welcome email sequence examples" not "learn more."
Authority injection. Link from your highest-authority pages (homepage, top blog posts) into the pillar to seed authority.
Sidebar / footer cluster modules. "More from the [Topic] series" modules at the bottom of cluster pages reinforce the topic identity.
Breadcrumb signaling. Breadcrumbs that surface the topic hierarchy ("Home > Email Marketing > Subject Lines") help users and signal structure to search engines.
Case study · HubSpot · the experiment that named the model
2016topic cluster experiment↑ SERP placementwith more interlinkingde factoindustry standard since
HubSpot’s research team ran the experiment that turned pillar-and-cluster from intuition into doctrine: clusters with denser, deliberate interlinking between pillar and spokes earned better placement in search results, and the effect strengthened with cluster completeness. They rebuilt their entire blog architecture around it — thousands of posts reorganized under pillars — and published the model the industry still uses. The often-cited caveat: their numbers are directional, from their own property. The architecture’s logic — concentrated internal relevance beats scattered posts — is what every result since has kept confirming. (HubSpot — topic clusters research)
Breadth vs depth: when to fan out, when to go deep
Two strategic patterns:
Fan out (breadth-first). Cover many subtopics with adequate depth on each. Best when topic is new to your site, you need to establish coverage, or competitors have thin coverage.
Go deep (depth-first). Pick the highest-priority subtopics and write definitive, primary-research-backed content. Best when topic is mature on your site and you're competing against established authorities.
Mature programs do both: maintain breadth through systematic coverage of all major subtopics, then deepen on the highest-value subtopics through original research, primary case studies, and continuous refresh.
RGM EXPERT TRICK
The 70% rule: finish clusters before starting new ones
Our hard rule: no new pillar launches while an existing cluster is under ~70% of its planned spokes. Half-built clusters are the most common topical-authority failure we inherit.
The math is the S-curve above: five clusters at 40% built means five clusters stuck in phase 1 forever — maximum spend, zero compounding. One cluster at 100% reaches phase 3 and starts paying for the next.
When a stakeholder wants a new topic mid-build, we show the coverage matrix and trade explicitly: which planned spokes are we cutting to fund it?
WHY IT’S RARE · Content calendars optimize for novelty — new topics feel like progress. Finishing feels like repetition. The discipline to finish is the entire moat.
Measuring topical authority growth
No single metric captures topical authority. Use a portfolio:
Cluster impressions and clicks. Aggregate Search Console metrics for all pages in the cluster. Total impressions and clicks should grow even when individual page metrics stagnate.
Average position for cluster queries. Track ranking position across the full keyword set for the topic. Cluster-wide position improving is the strongest signal.
Featured snippet wins. Topical authority correlates with snippet wins. Track which queries display snippets and whether you hold them.
Internal link audit metrics. Inbound internal links per cluster page. Pages with growing internal links typically rank better.
External links to cluster pages. Backlinks to cluster pages (especially from authoritative sources in the topic) accelerate topical authority.
Topic-graph completeness. Of all subtopics surfaced in research, what % do you cover? Aim for 70%+ coverage on the priority cluster.
Pattern What topical authority growth actually looks like
The S-curve every cluster follows — and where teams quit
Shape is RGM analysis from client clusters; your slope varies with site authority and competition. The strategic point: phase 1 looks like failure. Teams that judge clusters at month three kill compounding assets the quarter before they pay — budget for the full curve or do not start.
Refresh and consolidation
Refresh top pillar and cluster pages every 6 months. Update statistics, examples, screenshots. Add new sections for emerging subtopics.
Consolidate thin or overlapping pages. If three cluster pages all partially cover the same subtopic, consolidate into one definitive page with 301 redirects from the others. Loses URL count, gains topic depth.
Deprecate dead clusters. If a cluster has consistently underperformed for 12+ months, evaluate whether to invest in revitalization or accept it and stop dilution.
Spin off mature subtopics. When a sub-cluster within a cluster reaches 10+ pages, consider promoting it to its own pillar with its own sub-clusters.
Calculator Refresh, rewrite, consolidate, or leave it — the decay triage
Three numbers from GSC decide what each aging page needs
REFRESH
Thresholds are RGM analysis — the triage we run quarterly across client content libraries. The principle is sourced everywhere in this module: Google rewards fresh, differentiated coverage, and decaying pages drag their whole cluster.
RGM EXPERT TRICK
Consolidation is a growth tactic, not spring cleaning
Some of our biggest “new content” wins shipped zero new pages: six thin posts on one subtopic merged into a single URL, with 301s from the other five.
The merged page inherits every internal link, every backlink, and the combined query surface. We measure queries-per-URL in GSC before and after — the merged page typically outranks the best of its parents within weeks.
Candidates come straight from the coverage matrix: any cell with three pages and no rankings is a merge, not a gap.
WHY IT’S RARE · Publishing feels productive and deleting feels dangerous, so libraries bloat. Teams that treat consolidation as a growth line item — with before/after numbers — are vanishingly rare.
Advanced playbook
Entity-first cluster planning. Identify the central entity (the brand, product, person, technology), list all related entities (suppliers, alternatives, regulators, key people), and build pages covering the entity-to-entity relationships.
Information gain investment per cluster. Reserve budget for primary research within each major cluster — original survey, dataset, expert interviews. Information gain at the cluster level lifts every page in the cluster.
Reciprocal cluster integration with sister topics. If you cover "email marketing" and "marketing automation," build deliberate cross-cluster links. Topical authority on adjacent topics reinforces each other.
Topic-level reporting dashboards. Build dashboards that roll up cluster metrics (impressions, clicks, positions, links) so leadership sees topic-level progress, not URL-level noise.
Topical anchor text diversity. Across all internal links pointing to a target page, vary the anchor text within the topic. Pure exact-match anchors look spammy at scale.
External validation integration. Get external authoritative sources to link to your pillar (industry reports cite you, expert blogs reference your data). Topical authority + external validation is the strongest combination.
Topical author authority. For each major topic, identify 1–3 author bylines. Build their authority externally (conference talks, podcasts, guest posts, Wikipedia/Wikidata entries). Author authority transfers to your site.
Cross-language cluster replication. Once a cluster matures in one language, replicate in priority languages with localization (not translation). International topical authority is largely greenfield for most brands.
Cluster-level competitor analysis. Run quarterly cluster-level comparison vs top competitor. Identify subtopics they cover that you don't. Decide: add, ignore, or differentiate.
Hub-page versions for new audience segments. When you find a new audience segment using the same topic differently (e.g., beginners vs experts), create alternate pillar entry points serving each segment.
Step by step Build a cluster from zero — the RGM sequence
From topic choice to compounding, in the order that works
Choose a topic you can be definitive about.The test: could your team write the best page on the internet for 25 subtopics here? If not, narrow until yes. Authority is binary at the niche level.
Mine GSC before any keyword tool.Export existing impressions; subtopics Google already associates with you are your beachhead. Then layer keyword tools for gaps your data cannot see.
Draw the coverage matrix.Subtopics × intents, colored honestly. The matrix IS the content calendar — rows before columns, money intents flagged.
Write 3-5 spokes BEFORE the pillar.Spokes give the pillar something to bind. A pillar published into a void is a table of contents for a book that does not exist.
Ship the pillar as a true hub.Substantial standalone answer + descriptive-anchor links to every live spoke + visible structure machines can parse. Every spoke links back with consistent anchors.
Add one information-gain element per page.Original data, a tested process, a contrarian-but-earned position. The patent language is literal: new information scores; repetition does not.
Review quarterly against the matrix + decay triage.Fill the next gaps, refresh what decayed, merge what fragmented. The cluster is a product with a roadmap, not a campaign with an end date.
Common mistakes
Building a pillar page with no cluster strategy — the pillar sits alone with no authority signal.
Building cluster pages with no pillar — no central authority hub.
Treating cluster pages as keyword-per-page targets instead of subtopic coverage.
Hub-and-spoke linking only one direction — cluster pages don't link to pillar or to each other.
Stuffing identical anchor text across all internal links.
Fan-out without quality — 30 thin cluster pages do less than 10 deep ones.
No information gain in cluster pages — commodity content that summarizes what already ranks.
Never refreshing cluster pages; topical authority decays as content goes stale.
Skipping internal linking from authoritative existing pages into the new cluster.
Building clusters around keywords search volume alone, ignoring intent fit.
No measurement at cluster level — only URL-level reporting that misses the topical authority story.
Killing or deprioritizing clusters before they mature (12–24 months is typical).
Practitioner references: Bill Slawski’s patent archive (SEO by the Sea) · Kevin Indig on topic-driven SEO · Koray Tuğberk Gübür on semantic SEO · Ahrefs & Moz cluster guides
Ten questions, CASE method (Context · Analysis · Strategy · Execution). Pass at 90% to unlock this module’s completion passcode — retake as many times as you like.