Site Architecture

---
name: site-architecture
description: Plan a site's content structure as pillars and clusters, and wire the internal link graph so it earns topical authority. Use when asked to structure a website, plan site architecture, build a content plan or pillar-cluster strategy, fix internal linking, plan topic clusters, decide what pages a site needs, work out why a blog is not ranking, or someone says 'forty posts and no traffic'. For writing the individual pages, see landing-page-copy. For auditing an existing page's build quality, see design-review.
metadata:
  version: 1.0.0
---

# Site architecture

Search stopped rewarding one clever article. Google, ChatGPT, Perplexity and AI
Overviews reward **topical authority**: a site that covers one subject completely,
links it together tightly, and keeps it fresh.

That is a volume-plus-structure game. Most sites lose it on structure — forty
unconnected posts is not authority, it is forty posts. This plans the structure
first, so every page published afterwards has somewhere to attach.

## Before you start

| Input | Why |
| --- | --- |
| The one subject the site should own | Topical authority is per subject. A site about everything owns nothing |
| Existing published URLs, with titles | Cannot plan around what you already have without seeing it |
| Keyword volume and difficulty, if available | Validates a pillar is worth building. Free stack works — see below |
| Whether the site is new or established | New sites plan forward; established ones fix the graph first |

No keyword tool is a valid state. **Search Console** (what you already rank for),
**Keyword Planner** (volume), **AlsoAsked** (question mining) and **Google Trends**
cover it. Swap the data source, keep the system.

## Step 1 — Name the subject the site will own

One subject, stated in a sentence. Everything else in the plan is judged against it.

A topic that does not strengthen an existing pillar **dilutes the site**. Saying no
to a topic is the highest-value decision in this whole process, and it only works if
the subject is written down.

## Step 2 — Define the pillars

A pillar is a hub page covering one major area of the subject completely.

| Property | Value |
| --- | --- |
| Length | 2,500–3,500 words |
| Structure | 8+ H2 sections, each answering one question on its own |
| Contains | A named framework — the thing people cite and link to |
| Role | Everything in its cluster links up to it |

Three to six pillars is typical for a focused site. If you find yourself planning
twelve, the subject in step 1 is too broad — narrow it or split the site.

## Step 3 — Plan the clusters

For each pillar, plan **8–15 cluster posts**. Each takes one subtopic deep.

| Property | Value |
| --- | --- |
| Length | 1,500–2,200 words |
| Scope | One subtopic, covered properly |
| Links up | To its pillar, twice |
| Links across | To two sibling clusters |

For each planned post record: the exact title, the angle, the search volume and
difficulty if you have them, and its position in the link map. That turns one pillar
into a publishing roadmap.

## Step 4 — Wire the link graph

Build a directed graph of every internal link, then find the missing edges. **Three
patterns humans always miss:**

1. **Old posts never link to new ones.** The link only ever gets added in one
   direction, at publish time.
2. **Pillars do not link down to every cluster.** The pillar was written before most
   of the cluster existed.
3. **Siblings never link across.** Nobody thinks to connect two cluster posts to
   each other.

Target **8–15 internal links per post**. Check the sitemap before publishing and
never invent a URL — a link to a page that does not exist is worse than no link.

## Step 5 — Keep a registry

Maintain a running log of every published page: ID, URL, slug, pillar or cluster,
publish date.

Without a registry, every run has to re-crawl the sitemap and will still miss the
newest pages. **Treat the registry as the site's memory.** It is what makes internal
linking possible at all.

## Step 6 — Set the refresh ratio

**For every 2 new posts, refresh 4 old ones.** The 1:2 rule.

A refreshed post keeps its URL, its link equity and its rankings, and gains a
freshness signal. A republished post starts from zero. Refreshing is usually worth
more than the next new post, and almost nobody does it.

## Step 7 — Plan distribution

Without distribution, even a perfect pillar takes **6 to 12 months** to rank. Plan
the channel kit at the same time as the content: a post, a carousel brief, a short
video script, an email.

## Output

```
# Site architecture: <subject>

## Subject
<One sentence. Everything is judged against this.>

## Pillars
| Pillar | Target words | Named framework | Status |

## Clusters
### Pillar 1: <name>
| # | Title | Angle | Volume | Difficulty | Links up | Links across |
[8-15 rows, repeated per pillar]

## Link graph gaps
| From | To | Pattern | Anchor text |

## Registry
<Path to the running log, and its current row count.>

## Cadence
New posts: <n>/month   Refreshes: <2x that>   Distribution: <channels>

## Rejected topics
| Topic | Why it dilutes rather than strengthens |
```

## When it breaks

| What you see | What it means | The fix |
| --- | --- | --- |
| Forty posts, no rankings | Volume without structure — nothing links together | Assign every existing post to a pillar; wire the graph before writing more |
| Two posts competing for one term | Keyword cannibalisation — the cluster was not checked against existing rankings | Merge them, or re-angle one. Check existing rankings before planning |
| Planning twelve pillars | The subject in step 1 is too broad | Narrow the subject, or split into separate sites |
| Pillar has no inbound links | It was published before its cluster existed | Pattern 2 — link down from the pillar to every cluster, and back up |
| New posts get no traffic | Old posts never link to them | Pattern 1 — patch links from existing posts to the new one |
| Cluster posts feel repetitive | Subtopics overlap; they are variants, not distinct posts | Merge the overlapping ones; a cluster of near-duplicates dilutes |
| Rankings decay across the site | Nothing is being refreshed | Apply the 1:2 rule |
| A link 404s | A URL was invented rather than taken from the sitemap | Check the sitemap before publishing. Never guess a URL |
| Ranked for months, then dropped | Content went stale; competitors refreshed | Refresh keeps the URL and equity. Do not republish |

Never fix a structural problem by publishing more. Diagnose the graph first — more
unconnected posts make the problem larger, not smaller.

## Rules

- **One subject per site.** Topical authority is per subject; a site about everything
  owns nothing.
- **Say no to diluting topics.** A topic that does not strengthen a pillar weakens
  the site. Record what you rejected and why.
- **Never invent a URL.** Check the sitemap. A broken internal link is worse than a
  missing one.
- **Refresh, never republish.** Republishing throws away the URL, the link equity and
  the rankings you already earned.
- **Structure before volume.** Publishing into a site with no pillars produces posts
  with nowhere to attach.
- Quote word counts and link counts as the ranges given, not as hard rules —
  completeness beats the counter. Shorter than the floor usually means you lost the
  plot.
- This skill plans **content structure**. It does not cover URL/slug conventions,
  redirects, canonical tags, sitemap XML, or site speed — see `SOURCES.md`.

## Related skills

- **landing-page-copy** — writing the individual pages this plan calls for.
- **design-review** — auditing the build quality of pages that already exist.
- **bulk-ad-variations** — creative volume for distributing a pillar, not planning it.