Topic Clusters#

1. What is this page#

The Topic Clusters page is your content production blueprint — the page where you convert raw keyword research into a structured 3-tier hierarchy (Pillar → Sub-Pillar → Spoke) that drives every article you write.

It is not a writing surface. It is a planner. You upload a keyword export, the engine screens for quality and overlap, an AI agent organizes the survivors into a logical hierarchy, and you get back a page tree, a 12-week build calendar, and an internal link plan. Each cluster covers one topic domain on your site. Each page in the hierarchy becomes one article in your production queue.

Screenshot — Topic Clusters list with LANI SILK brand selected and two clusters shown (6 numbered callouts)

1. Brand selector — each brand has its own clusters
2. • New Cluster — opens the create modal with root keyword field
3. Cluster name + root keyword — display label and the broadest commercial term
4. Status pill — Planning / In Progress / Complete based on coverage percentage
5. Pages count — number of articles planned for this cluster after generation
6. Created date — when the cluster was first imported

The Topic Clusters page — each numbered area is explained below

2. Why it matters#

• Topical depth, organized for editors and machines — Pages that demonstrate consistent depth on one topic tend to be easier for both editors and search systems to navigate than scattered posts on unrelated topics. A clean Pillar → Sub-Pillar → Spoke hierarchy supports your internal editorial process and presents a clearer signal of topical coverage to readers. Whether AI search engines specifically reward hierarchical structure is not publicly confirmed, but a well-organized cluster is generally easier to summarize and cite than fragmented content.

• Anti-cannibalization by design — Every screened keyword is assigned to exactly one page as primary_keyword or supporting_keyword. The engine catches duplicate intent before you write. You stop accidentally publishing two pages that compete for the same SERP slot.

• Clearer topic relationships for downstream systems — A well-structured cluster makes the relationships between pages explicit: which page is the topic anchor, which pages explain sub-themes, which pages cover specific sub-questions. This helps editors maintain consistency over time and may help downstream systems — search engines, LLMs, internal site search — interpret the topic structure more accurately. It is not a guarantee of placement in AI Overviews, AI Mode, or any specific feature, but it is good editorial hygiene that supports both human readers and machine systems.

• Intelligent build order — The engine assigns P1 (pillar + hubs), P2 (high-volume spokes), P3 (long-tail spokes) priorities and lays them out across a 12-week calendar. Internal linking only works if you build the link target before the linker — P1-first ordering preserves dependency direction so no spoke is published with broken internal links.

• Brief pipeline integration — Click "Generate Brief" on any Topic Page and the page metadata (primary keyword, supporting keywords, intent, target word count, internal link targets) pre-fills the Brief module. You skip most of the brief setup work and stay focused on the strategic decisions.

3. Where it fits in the workflow#

Brand → Discover → [Topic Planner] → Brief → Outline → Draft → Review → Validate → Publish → Monitor → Refresh

Topic Planner sits between Discovery (raw keyword research) and Brief (single-article spec). Discovery answers "what topics could we cover?" Topic Planner answers "which topics, in what order, with what hierarchy?" Brief answers "for this one topic, what does the article say?"

Without Topic Planner, you go from 500 raw keywords directly to writing — and end up with disconnected articles, accidental duplicates, and no internal link strategy. With Topic Planner, those 500 keywords become a 50-page hierarchy with a 12-week build calendar where every page is traceable end-to-end into a Brief → Outline → Draft → Published Asset.

4. How to use it#

Vocabulary — terms used in this guide#

A few terms appear with different labels in different parts of the UI. Standardize them in your head before reading the workflow:

One cluster = one Pillar + N Groups, where each Group = one Sub-Pillar + multiple Spokes.

The full workflow is 9 steps. Steps 1–6 build the cluster; steps 7–9 execute it.

Build the cluster#

1. Click + New Cluster and enter a root keyword. Pick the broadest commercially relevant term for this topic domain (e.g. silk nightgown, not silk nightgown for hot sleepers). One root keyword = one cluster.

2. Upload your keyword CSV. Export from Ahrefs Keywords Explorer, Semrush Keyword Magic Tool, or any custom CSV with at minimum keyword and volume columns. The engine auto-detects format. A green confirmation appears: "Imported N keywords · M duplicates skipped."

3. Screen Keywords. Click the button. A deterministic pass removes obvious junk (zero-volume entries with no SERP features, exact duplicates, near-zero one-word phrases). An AI pass then classifies the ambiguous middle into Keep / Remove / Merge. Each decision shows the volume, AI reason, and confidence score. The stats panel updates: Total / Kept / Removed / Merged / Review / Pending.

4. (Optional) Import Existing URLs. If you already have published content on this topic, paste URLs (manual list or sitemap crawl). The engine fetches each page's title and matches it to a keyword via semantic similarity. Mapped URLs become published Topic Pages in the cluster — preventing the AI from planning duplicate articles for content you already rank for.

5. Generate Cluster. Click the button. The engine assembles brand context + kept keywords + any existing URLs and asks the model to build a full 3-tier hierarchy. Output: 1 Pillar + N Groups (each Group = 1 Sub-Pillar + multiple Spokes) plus suggested internal links and a 12-week build calendar. Generation typically takes 3–6 minutes for an 80–150 keyword cluster based on current generation speed; larger clusters or slower model-response periods may take longer. The hard timeout is 10 minutes.

6. Review and edit. The hierarchy renders as a tree under the Page Hierarchy tab. Inline edits available per page: rename meta title, change role (promote Spoke → Sub-Pillar, demote Sub-Pillar → Spoke), move to a different Group, mark duplicate (with reason), mark skipped, link to an existing URL, reorder build sequence. Per Group: add manual Spokes, reorder pages. Per cluster: regenerate the AI hierarchy, add a manual Group, export the full plan to Excel.

Execute the cluster#

7. Generate Brief from a Topic Page. Open any page in the tree and click "Generate Brief". The page metadata pre-fills a new Brief in the Brief module, and you continue normal article production from there. A dependency guard runs at this step: you cannot brief a Sub-Pillar before its Pillar is at least briefed, and you cannot brief a Spoke before its Sub-Pillar is briefed. This enforces correct build order at the system level — no orphan articles.

8. Track progress via Dashboard and Calendar. The Dashboard surfaces per-cluster coverage (published pages / planned pages, as a percentage and an authority level). The Calendar lays the rolling 12-week build plan out visually, with each page in its scheduled week.

9. Manage Internal Links via the Internal Links tab. The engine pre-generates the link plan: every Spoke → its Sub-Pillar + Pillar, every Sub-Pillar → Pillar, plus cross-Group links where entities overlap. The tab shows link source, target, link type, suggested anchor text, and implementation status. Mark each link as Implemented once your published article actually contains it.

Hierarchy roles — what each page does#

Priority levels — what to build first#

Within the same priority bucket, pages are ordered by search volume DESC. The engine handles this automatically when you click Generate Cluster — you only override priority manually if you have a business reason (e.g. seasonal urgency on a specific Spoke).

5. Field reference#

A Topic Cluster has three nested entities: Cluster (level 1), Group (level 2), Page (level 3). You enter only a handful of fields directly — the rest come from the AI generation step and you edit them inline after review.

Fields you enter#

Engine-managed fields (read-only)#

The engine recomputes these read-only fields whenever you add or remove pages, mark duplicates, link to existing URLs, or transition a page through brief → draft → published states. You never write these directly.

6. Examples for your category#

A well-built cluster delivers one topic domain end-to-end. Curate foundation clusters first — the 5-7 highest-value topic areas for the brand — then add product-specific clusters for SKU-level depth as the catalog expands.

Starter clusters — foundation coverage#

For a silk apparel brand like LANI SILK, these seven cluster themes cover most commercial intent traffic at the category level (coverage varies by market, season, and competitive density). Each cluster typically generates 1 Pillar + 5–8 Groups + 30–60 total pages.

1. Core product clusters — one cluster per primary product line: silk nightgowns, silk pajamas, silk pillowcases, silk eye masks, silk robes. Each becomes a separate Topic Cluster with its own root keyword and Pillar page. This is your commercial-intent backbone.

2. Material education — root keywords like mulberry silk, silk grades, momme weight. TOFU-MOFU informational intent. Spokes cover specific sub-questions (what is 22 momme silk, grade 6A vs 5A silk); the Pillar synthesizes. This cluster builds the topical authority that pays off when commercial pages link back here.

3. Care and maintenance — how to wash silk, silk drying, silk storage, removing stains from silk. Informational with embedded commercial intent (readers buying silk care products). High evergreen value — these pages keep ranking for years.

4. Comparison and decision content — silk vs satin, silk vs cotton sleepwear, silk vs bamboo bedding. Commercial intent, bottom of funnel. Drives conversions for visitors comparing options before purchase.

5. Seasonal and use-case — silk for hot sleepers, silk for menopause sleep, silk for sensitive skin, silk for hair. Audience-specific clusters that convert well in both paid and organic. Spoke-heavy because each audience segment branches further.

6. Sustainability and ethics — ethical silk production, OEKO-TEX silk, peace silk vs traditional, sericulture environmental impact. Mid-funnel trust content that reinforces brand positioning. Pairs naturally with Evidence Library entries about certifications and sourcing.

7. Gift and occasion — silk gifts for women, wedding silk gifts, anniversary silk, Mother's Day silk. Seasonal traffic spikes; Spoke-heavy clusters with strong commercial intent in the lead-up to each occasion.

Product-specific clusters — drilldown#

When a foundation cluster shows traction (50%+ of pages published, coverage_pct ≥ 50), spin up product-specific clusters for SKU depth. Example: the Silk Nightgowns foundation cluster covers the category at the Pillar/Sub-Pillar level; a follow-on Long Silk Nightgowns cluster goes deep on length variants (long sleeve, knee-length, midi, ankle-length, with/without lace, with/without slit). Each follow-on inherits the parent Pillar as a link target, building deeper hierarchy without competing.

Recommended order#

• Build one foundation cluster fully (Pillar + all Sub-Pillars + half the Spokes published) before starting a second cluster. Coverage on one topic outperforms partial coverage on many.
• Inside a cluster, follow the engine's P1 → P2 → P3 priority. Skipping ahead breaks internal linking — the Spoke wants to link to a Sub-Pillar that does not exist yet.
• Re-run Generate Cluster only when keyword research changes substantially. Manual edits are preserved per page, but regenerating overwrites the AI-suggested hierarchy and you lose unsaved manual moves.

7. Common mistakes & FAQ#

Anti-patterns#

✓ Good clusters

Clear differentiating dimension per Group — By Color, By Length, By Material Weight each define one axis. Spokes inside one Group all share the same axis variation. Tree is scannable; users (and the AI) can predict where to look for a topic.

One keyword = one page — every screened keyword is the primary or supporting keyword of exactly one page. Cannibalization caught at planning time, before you write a word.

Pillar built before Sub-Pillars, Sub-Pillars built before Spokes — internal link targets exist when the linker is published. P1 → P2 → P3 build order respected; no orphan pages with broken internal links.

✗ Avoid these

Flat keyword list, no Groups — 50 Spokes hanging off the Pillar with no Group structure. Internal linking becomes spaghetti. Topic structure becomes hard to interpret for both editors and downstream systems. Page tree is unscannable for editors.

Skipping the Screen Keywords step — generating a cluster directly from raw imported keywords feeds the AI low-quality terms (typos, near-duplicates, off-niche entries). Output cluster has wrong Sub-Pillars and irrelevant Spokes you'll have to manually delete.

Marking too many pages as duplicate — duplicates lose internal link value AND brief generation eligibility. Only mark as duplicate when two pages genuinely target the same intent + same primary entity + same funnel stage. Different angles on the same topic are not duplicates.

Common mistakes#

• Uploading the CSV before refining the keyword export in Ahrefs/Semrush. Junk in, junk out. The Screen step catches obvious issues but cannot fix a fundamentally off-topic keyword list. Filter your export in the source tool first — by KD range, parent keyword, intent filter, country — before exporting.

• Generating Cluster on a very large keyword set. The engine works best in the recommended 50–300 keyword range. Above the current screening ceiling (~327 keywords at time of writing — this number may shift with model and configuration updates), the screening pass returns partial results and the cluster generator may miss Sub-Pillars. Split very large exports into two clusters with different root keywords.

• Skipping Import Existing URLs when you have published content. Without this step, the cluster generator plans new pages for keywords you already rank for, creating duplicates. Always import existing URLs first if your site has any topic-relevant content already live.

• Editing the hierarchy heavily, then clicking Regenerate. Regenerate overwrites all AI-suggested structure. Manually added pages persist (flagged separately), but role changes, renames, and reorders within AI-suggested pages get reverted. If you have done significant manual work, do not regenerate — keep editing instead.

• Generating Briefs out of priority order. Clicking Generate Brief on a P3 Spoke before its P1 Pillar exists creates an orphan in the Brief queue. The dependency guard surfaces a warning at the system level; respect it. P1 first, always.

FAQ#

How many keywords should I import per cluster? Recommended range is 80–250 keywords post-screening. Below 50, the cluster lacks depth. The current upper ceiling sits around 327 keywords per screening pass (derived from current model and configuration — this may shift as the system evolves). Above the ceiling, split the export into two clusters with distinct root keywords or re-screen the remainder.

How long does Generate Cluster take? Typically 3–6 minutes for an 80–150 keyword cluster based on current generation speed. Larger clusters, slower model-response days, or future model changes may shift this range. The hard timeout is 10 minutes — past that the worker returns an error rather than hanging.

Screen Keywords completed but the stats show many "Pending" — what does that mean? Pending means the AI was not confident enough to classify those keywords as Keep, Remove, or Merge. Open the Pending list and decide each one manually, or re-run Screen Keywords on the Pending subset only. Common causes: ambiguous one-word phrases, off-niche keywords that bled into the original export, or branded terms the model could not place in context.

Cluster generation is stuck on "Generating" for more than 10 minutes — what should I do? Refresh the page first — sometimes the worker finished but the UI did not update. If the status still shows Generating after refresh, the worker has likely surfaced an error that the UI did not catch. Open the cluster detail; the error message should appear in the page header or stats panel. If neither shows an error and the status remains stuck, the safest action is to delete the cluster and re-create it with the same root keyword — your screened keyword list is preserved on the export side.

When should I split one cluster into two? Three signals warrant a split. (1) Keyword count exceeds the screening ceiling — the screening pass returns partial results above the current ~327-keyword limit. (2) Root keyword covers two genuinely different intents — e.g. silk gifts meaning both "silk products to give as gifts" and "gift cards for silk brands" are different commercial intents that each deserve their own Pillar. (3) The AI hierarchy comes back with two equal-sized top-level groupings that do not share a natural Pillar — that suggests the root keyword was too broad and two distinct topics were merged. In all three cases, split with two distinct, narrower root keywords for cleaner topical authority.

Can a Topic Page link to pages in a different cluster? Yes. The Internal Links tab supports cross-cluster links where entities overlap. Example: a silk pillowcase Spoke in the Silk Pillowcases cluster can link to a mulberry silk Sub-Pillar in the Material Education cluster. The engine suggests these automatically when it detects entity overlap; you can also add manual cross-links.

What happens if I delete a Topic Page that already has a published Brief? The Brief itself stays — it lives in the Brief module independently. The link from the Topic Page to the Brief breaks, but the Brief content is preserved. You can re-link from a different Topic Page later if needed.

Should I run Generate Cluster again after editing keywords? Only if the keyword changes are substantial (roughly ≥30% of the keyword set added or removed). Small edits do not warrant regenerating — you lose all manual hierarchy edits. For minor adjustments, edit the cluster tree manually instead.

Can I have a Topic Cluster without a Pillar page? No. Every cluster must have exactly one Pillar. The Pillar serves as the topic anchor and the internal link target for every other page in the cluster. If a topic feels too narrow for a Pillar of its own, you might be working at the wrong level — it likely belongs as a Sub-Pillar inside a broader cluster.

Need help with topic planning? See also: Hierarchy roles explained · Priority and build order · Internal linking strategy.