Skip to content

Decides

Cluster fit

Places a new page in your topic clusters as a pillar or supporting page, names the gap it fills and lists the pages it should link to and from.

Other tools

Add this skill to Claude

Claude takes skills as a ZIP file. Your download has started; if it didn't, use the button below.

  1. 1In Claude, open Customize, then Skills.
  2. 2Choose +, then Upload a skill, and pick cluster-fit.zip.
  3. 3Ask for it by name, or let Claude use it when a task matches.

Skills need code execution turned on in Claude's settings. On a Team or Enterprise plan, an admin can share it with everyone.

Download the folder

Try it now, without installing

See the skill work on a made-up input before you add it. Nothing is installed, and you can swap in your own input.

Open in the Claude app

Opens a new chat in the Claude desktop app with the prompt filled in. Nothing is sent until you press enter.

Includes the whole SKILL.md and the sample input. Paste it into Claude, ChatGPT or any chat.

Influenced by
Link best practices
Needs
The keyword, its intent and its demand · Your site's pages, with URL and title
Works with
Claude, ChatGPT, Codex, Cursor and any agent that reads SKILL.md
Version
1.0.0 · Last checked · License MIT
Maintained by
Sabih AhmedOrganic visibility
SkillSEO and AI search

Cluster fit

Example output

Place a new page
PageClusterRole
Attribution softwareAttributionSupporting
Pipeline forecastPipelinePillar first
Sales comp plansNoneStandalone

Places a new page in your topic clusters as a pillar or supporting page, names the gap it fills and lists the pages it should link to and from.

The skill

The whole skill, in plain text.

Nothing is hidden. Read it, change it to fit your team, or take the whole folder.

Scripts
Optional Python scripts for the counting. It works without them and says which way it ran.
Network
Opens web pages or your tools if your agent can. Works from pasted data if not.

3 test cases

cluster-fit/SKILL.md

In the folder

  • cluster-fit/SKILL.md
  • cluster-fit/scripts/cluster_place.py
---name: cluster-fitdescription: >-  Places a planned B2B SaaS page in the site's topic clusters: decides whether it is a pillar or a supporting page,  names the gap it fills in its cluster, and lists the pages it should link to and get links from. Returns  SUPPORTING, PILLAR, PILLAR FIRST, NEW CLUSTER, STANDALONE, NEEDS DECISION or BLOCKED. Use it after  keyword-selection and cannibalization-audit clear a page, before the outline. Not for writing anchor text or  placing links in a draft (use internal-linking) or for checking clashes (use cannibalization-audit).license: MITcompatibility: >-  Works in Claude, ChatGPT, Codex, Cursor and any agent that reads SKILL.md. Reads a crawl or sitemap if the agent  can browse, otherwise a pasted page list. Python 3 is optional, for the placement score.metadata:  author: SHAKEUP  version: "1.0.0"  stage: decide  plugin: seo-ai-search  homepage: https://www.shakeuphq.com/skills/cluster-fit--- # Cluster fit Gives a new page a place in the site before it is written, so it doesn't ship as an orphan with no pillar above it and no links in. ## Purpose - One job: decide the role of one planned page in the site's clusters, the gap it fills and its link targets.- Stage: decide. It ends in a verdict with the rule that fired.- It picks pages to link, never anchor text or placement (that is `internal-linking`).- It never invents a pillar URL or a page that doesn't exist; a missing page is listed as a missing target. ## Use when / not | Situation | Use || --- | --- || A page has a keyword and passed the clash check | this skill || Planning a new cluster and its first page | this skill (it returns NEW CLUSTER or STANDALONE) || The keyword isn't chosen yet | `keyword-selection` first || You don't know if the page repeats one you have | `cannibalization-audit` first || You need anchors and positions for links in a draft | `internal-linking`, after this skill | ## Inputs | Input | Required | Ways to provide, in order | If missing || --- | --- | --- | --- || Primary keyword, variants, intent, demand tier | yes | the `keyword-selection` handoff; the reader's request | ask once || Clash result | no | the `cannibalization-audit` handoff | say "clash not checked" in open items || Cluster map: each cluster's name, pillar URL, pillar keyword | no | `## Content` topic clusters in the profile; a clusters file the reader keeps; inferred from the page list | infer from the page list, coverage `partial` || The site's pages: URL and title | yes | a crawl or sitemap the agent can read; a crawl export; a pasted list (URLs alone work, weaker) | `BLOCKED`, ask for it || Planned topics | no | the reader's content plan or calendar | adjacent gaps can't be checked against the plan | ## Profile read From `company-profile.md`: `## Content` (topic clusters with pillar URLs, site sections), `## Product` (product pages a page can link to), `## Buying moments` (to judge whether an unmatched topic deserves a new cluster), `## Buyers`, and every `## Learnings` entry for `cluster-fit` with status `noted`, `fact` or `accepted`. No profile: ask for the website and, if they keep one, their list of topic clusters with pillar URLs, in one message. Say that `company-profile` saves this next time. ## Steps 1. **Apply learnings.** Read them first; an accepted learning overrides the default rule it names. 2. **Read the upstream verdicts.** If `cannibalization-audit` returned `REFRESH` or `BLOCK`, this is not a new page: stop with `BLOCKED` and point to `refresh-audit` or `internal-linking` for the existing page. If it returned `DIFFERENTIATE`, carry its angle and sibling URL into every step below. 3. **Build the cluster map.** Start from the profile's clusters. Add the page list: group pages by URL folder and shared title words. For each cluster record:    | Field | How |   | --- | --- |   | Pillar | the profile's pillar URL; else the broadest live page in the group; else `missing` |   | Pillar keyword | the pillar's main topic in 1 to 3 words |   | Supporting pages | URL, title, the sub-intent each serves |   | Stage coverage | which of learn, compare and buy the cluster's pages serve |    From URLs alone (no titles), read the slug as the title and mark the map `inferred`. 4. **Score the match.** With Python, write the keyword, variants and map to JSON and run:    ```   python3 scripts/cluster_place.py --json place.json   ```    Without Python: for each cluster, count the keyword's words that also appear in the pillar keyword or a supporting title, divided by the words in both. Best at 0.05 or under is no match (the words barely meet). A second cluster within 0.10 of the best is ambiguous. Say which way you ran it. 5. **Confirm on meaning.** A word match is a candidate, not a placement. Confirm that a buyer reading the cluster's pillar would expect this page as a next step. When the scores are ambiguous, pick the cluster whose pillar answers the question this page narrows; if both do, that is `NEEDS DECISION`. 6. **Decide the role.**    | Condition (first match wins) | Role | Why |   | --- | --- | --- |   | The keyword is the cluster's head term (it holds 80% or more of the pillar keyword's words and adds at most 1) and a live pillar exists | conflict: `NEEDS DECISION` | a pillar already owns the head term; a second one splits it |   | Head term, no live pillar, and the page will link out to 3 or more supporting pages | `PILLAR` | a pillar is the broad page the rest point up to |   | Narrower keyword (a modifier: a team, an industry, "vs", pricing, a use case, a how-to) and a live pillar | `SUPPORTING` | one sub-intent under an anchor |   | Narrower keyword, matched cluster, no live pillar | `PILLAR FIRST` | a supporting page with no pillar has nothing to pass authority to |   | No match, the topic sits on a buying moment or the profile's category, and 3 or more supporting topics can be named now | `NEW CLUSTER` (this page is its pillar) | a cluster needs a pillar and at least 3 pages under it |   | No match otherwise | `STANDALONE` | a page with no cluster gets no links in and rarely earns its cost |    With a demand tier from `keyword-selection`, a head term is usually high demand (1,000 or more searches a month). Without one, decide on breadth and say so. 7. **Name the gap.** Write it in this form: "This page covers <sub-intent> for <buyer>, which no page in <cluster> covers." Check the supporting pages: if one already serves the sub-intent, go back to `cannibalization-audit`. Also note:   - a stage gap it fills (the cluster had no compare or buy page, say);   - up to 3 adjacent gaps it surfaces but doesn't fill, as future topics. 8. **List link targets** (pages and reasons, no anchors):    | Direction | Role | Targets |   | --- | --- | --- |   | Out | SUPPORTING | the pillar (always), 2 to 4 siblings with related but different sub-intents, at most 1 product page from `## Product` when the topic maps to it |   | Out | PILLAR, NEW CLUSTER | every live supporting page in the cluster, and the main product page |   | In | any | up to 5 live pages: the pillar, siblings that mention this sub-intent, pages that now link to a weaker page on it |   | Missing | any | pages it should link to that don't exist yet, listed as future topics |    Never list a page with a 4xx or 5xx status, a redirect, or a legal, login or ad landing page. ## Check and verdict | Verdict | Condition (first match wins) || --- | --- || `BLOCKED` | No keyword; no page list after asking; or the clash check returned REFRESH or BLOCK || `NEEDS DECISION` | The head term of a cluster whose pillar is live; or 2 clusters fit equally on meaning; or `keyword-selection` named a pillar this map contradicts || `PILLAR FIRST` | Narrower page for a cluster with no live pillar || `PILLAR` | Head term, no live pillar, 3 or more supporting pages to link out to || `SUPPORTING` | Narrower page under a live pillar || `NEW CLUSTER` | No match; the topic sits on a buying moment or the category; 3 or more supporting topics named || `STANDALONE` | No match otherwise | Named stops: `NEEDS DECISION` shows the 2 options (which cluster, or which page is the pillar) and what each costs. `BLOCKED` asks for the missing input or routes to the skill that fits. No other stops: `PILLAR FIRST` and `STANDALONE` are calls with reasons, not questions. ## Learn Append a learning to the profile when: - the reader overrides the role or the cluster (a `proposal` with their reason);- the reader confirms or adds a cluster and its pillar URL: update `## Content` topic clusters and record a `fact`;- a result comes back, such as a supporting page that ranked only after its pillar shipped (a `fact` with its source). Example: `- 2026-10-24 | cluster-fit | fact | Pipeline reporting cluster has a pillar at /guides/pipeline-reporting | Why: reader confirmed, page published 2026-10-20 | Status: noted` ## Output ```Verdict: <SUPPORTING | PILLAR | PILLAR FIRST | NEW CLUSTER | STANDALONE | NEEDS DECISION | BLOCKED> · Coverage: <complete | partial (what's missing) | blocked> · Score: <script | by hand>Rule fired: <the condition, in words> Cluster: <name> · Pillar: <URL | missing> · Match score: <n> (second: <cluster, n>)Role: <role>, because <reason>Gap filled: <one sentence in the form above>Stage gap: <stage it fills, or none>Adjacent gaps (future topics): <up to 3> | Direction | Page | Why |(out, then in, then missing) Flags: <pillar missing, possible clash, inferred map, or none>Learnings applied: <list, or none>``` Then the handoff block. ## Hand off ```handoffskill: cluster-fitversion: 1.0.0verdict: SUPPORTINGcoverage: partial (cluster map inferred from URLs)for: <primary keyword>outputs:  cluster: attribution  pillar: https://example.com/guides/b2b-attribution  role: supporting  gap: how each tool reads CRM lifecycle stages, for RevOps  links_out: [https://example.com/guides/b2b-attribution, https://example.com/blog/b2b-attribution-models]  links_in: [https://example.com/guides/b2b-attribution]next: [article-research, internal-linking]open: [confirm no planned post covers the same sub-intent]``` For `PILLAR FIRST`, `next` is `[keyword-selection]` for the pillar's keyword. For `NEEDS DECISION`, `next` waits on the reader. ## Failure modes | Problem | Do this || --- | --- || The profile's pillar URL returns a 404 or redirects | Treat the pillar as missing and flag the broken URL || Every page sits in one folder (/blog/) so folders say nothing | Group by title words and the sub-intents you read, and mark the map inferred || The site has no clusters at all | Run the NEW CLUSTER or STANDALONE rule, and list the 3 clusters the profile's buying moments suggest || The keyword fits a cluster but the buyer differs (a CFO page in a RevOps cluster) | Place it by need, note the buyer shift for the writer || A product page is the best pillar | Allowed: a product page can anchor a buy-stage cluster. Say so | ## Influenced by - SHAKEUP's content engine: placement scored on word overlap first, with ambiguity flagged for a judgment on meaning, and a rule that a supporting page never ships without a pillar to point to.- Google's link guidance, which explains how crawlers find pages through internal links and read anchor context: https://developers.google.com/search/docs/crawling-indexing/links-crawlable

How it thinks

Inputs in, a verdict out, and a stop only where you decide.

What we built

A supporting page with no pillar above it has nothing to pass authority to, so we flag the missing pillar before the page is written. The match is scored on words and confirmed on meaning, so a shared word never puts a page in the wrong cluster.

It reads your company profile if you have one, or asks for only what it needs. Then it works through its steps, checks its own work and ends in a verdict.

The steps

  1. 01

    Read your learnings and the clash check

    If the clash check found a page that already owns the need, it stops here. A new page has no place to take.

  2. 02

    Map your clusters

    Each cluster gets its pillar, its supporting pages and the stages they serve. A cluster with no live pillar is flagged.

  3. 03

    Score the match on words, then confirm on meaning

    Its script scores word overlap with every cluster. A close second is flagged, and the call is made on what the buyer needs next.

  4. 04

    Decide pillar or supporting

    A broad head term with no pillar above it becomes the pillar. Narrower pages sit under a live pillar, or wait for one.

  5. 05

    Name the gap and the links

    One sentence on the sub-intent it fills, then the pages it links to and the pages that should link back. Anchors are left to internal linking.

What you need

  • Required

    The keyword, its intent and its demand

    The handoff from keyword selection → Your request

  • Optional

    The clash check result

    The handoff from the cannibalization audit

  • Optional

    Your clusters and their pillar pages

    The topic clusters in your company profile → A clusters file you keep → Inferred from your page list

  • Required

    Your site's pages, with URL and title

    It reads your sitemap or crawl if your agent can browse → A crawl export you upload → A list you paste, even URLs alone

What it can return

  • SUPPORTING

    The page narrows a topic whose pillar is live.

  • PILLAR

    The page is the cluster's head term and no pillar exists yet.

  • PILLAR FIRST

    The page is supporting, but its cluster has no pillar. Plan the pillar first or alongside.

  • NEW CLUSTER

    No cluster fits, the topic sits on a buying moment and at least 3 supporting topics are named.

  • STANDALONE

    No cluster fits and there's no plan to build one around it.

  • NEEDS DECISION

    A live pillar already owns the head term, or 2 clusters fit equally.

  • BLOCKED

    There's no keyword or page list, or the clash check said the page shouldn't be new.

Where it stops for you

  • When a live pillar already owns the head term
  • When 2 clusters fit equally well

Example output

What you get back.

A run for a made-up company. Your output depends on your inputs.

Here's a run for a made-up B2B SaaS company that sells attribution software. The page is a comparison page for buyers choosing between tools.

Verdict

SUPPORTING

Coverage: complete

Score: script

Cluster

attribution, under the pillar /guides/b2b-attribution. The second cluster, pipeline reporting, scored 0.

Role

supporting, because "attribution software" narrows the pillar's topic to buyers comparing tools.

Gap filled

this page covers how attribution tools compare for RevOps teams on a sales-led cycle, which no page in the cluster covers.

Stage gap

every page in the cluster is a how-to, so this is its first page for buyers comparing tools.

DirectionPageWhy
Out/guides/b2b-attributionThe cluster's pillar
Out/blog/b2b-attribution-modelsSibling: which model fits a long cycle
Out/blog/self-reported-attributionA sibling on the field buyers ask about in demos
Out/product/pipeline-attributionThe product the comparison leads to
In/guides/b2b-attributionPillars link to every supporting page
In/blog/b2b-attribution-modelsReaders comparing models are close to comparing tools
The handoff block the next skill reads
skill: cluster-fit
version: 1.0.0
verdict: SUPPORTING
coverage: complete
for: attribution software
outputs:
  cluster: attribution
  pillar: /guides/b2b-attribution
  role: supporting
next: [article-research, internal-linking]
open: []

What good looks like

  • One cluster and one role, with the rule that decided it
  • A gap sentence a writer could build an outline around
  • The pillar listed first among the outbound links of a supporting page
  • A missing pillar flagged before the page is written
  • No page or URL that isn't on your site

Install

Use it in the tool you already have.

Pick your tool. Each tab has the shortest path we know of.

  1. 01
    Download cluster-fit.zip.
  2. 02
    In Claude, open Customize, then Skills.
  3. 03
    Choose +, then Upload a skill, and pick the ZIP. Code execution must be on in your settings.
  4. 04
    Ask for it by name, or let Claude use it when a task matches.

Questions

Before you add it.

Do I need a list of my topic clusters?

No. It reads them from your profile if they're there. Otherwise it groups your pages by folder and title, and says the map was inferred.

Will it write the internal links?

It names the pages to link and why. The internal linking skill writes the anchors and places them in the draft.

What if the page fits 2 clusters?

It scores both. When neither wins on meaning, it stops and shows you both options and what each would cost.

What does the script do?

It scores word overlap between the keyword and each cluster, flags a close second, and suggests a role and link targets. Without Python, the skill makes the same count itself and says so.