SEO topic clusters: how to build one, link it, and keep the pages from competing

An SEO topic cluster is a main page on a broad subject, called the pillar, surrounded by supporting pages that each answer one narrower question about it, with links running both ways. The pillar gives the overview and points to the detail. Each supporting page goes deep on its question and links back. You build a cluster so that search engines and readers see that your site covers the whole subject, and so that no two of your pages fight for the same query.

This page shows how to choose the topic, list the supporting pages, link them, and check that each page has a different job. It ends with the maintenance routine that keeps a cluster healthy.

What a cluster is made of

Part	Job	Typical query it answers
Pillar page	Overview of the subject and a map to the detail	The broad term: "bookkeeping for freelancers"
Supporting page	One question answered completely	A narrower term: "how to track receipts for tax"
Comparison or choice page	Help the reader pick between options	"Spreadsheet or software for freelance books"
Process page	Show the steps to do one thing	"How to invoice a client in another country"
Link set	Connect the pages with descriptive anchor text	Not a query, but it tells readers where to go next

The example above is an illustration of the shape, not data from a study. Swap in your own subject.

Why the structure works

People rarely ask one question and stop. A buyer asks the broad question, then three or four narrower ones, then a comparison. A site with one thin page on the broad term and nothing else forces the reader to leave. A cluster keeps them reading your pages for the whole journey.

Search engines read links as a map of what a page is about and how pages relate. A clear pillar with supporting pages that link to it shows which page is the main one for the broad query. Google says in its SEO starter guide that links help it find pages and understand them, so how you link is part of the work, not decoration.

Step 1: choose a topic you can cover completely

Pick a subject where you can write twelve useful pages without repeating yourself. Test three things:

• You sell something related, or you know the subject first hand.
• Buyers ask more than five different questions about it.
• The broad term has some demand but the narrow questions are where you can win.

A topic too wide, such as "marketing", needs a hundred pages. A topic too narrow, such as one product's setup screen, is a single page. The right size has a clear edge.

Step 2: list every question, then group by intent

Collect the questions from five places you already own: customer emails, sales calls, support tickets, your site search, and Search Console queries. Add the related questions that appear in search results for the broad term. The method for finding them is in how to find SEO keywords.

Now sort the list by what the person wants to do. Write four headings on a sheet: learn, compare, do, fix. Put each question under one of them. Questions that mean the same thing go on one line, because they will be answered by one page.

Step 3: give each page one query and one job

This step is the one most clusters skip, and it is why clusters fail. Assign each page a single main query and write it in a column next to the URL. If two rows have the same main query, merge them or change one. Two pages for one query split the signal, and neither ranks as well as one strong page would. The method of choosing one query per page is in focus keyword examples.

Your map should look like this:

URL Main query Job
/guide/subject broad term overview, links to all
/guide/subject/question-a narrow question A answer A fully
/guide/subject/question-b narrow question B answer B fully
/guide/subject/compare option one vs option two help the reader choose
/guide/subject/how-to-c how to do C steps with a checklist

Step 4: link in a pattern, not at random

Use three rules.

1. Every supporting page links up to the pillar with anchor text that names the subject.
2. The pillar links down to every supporting page in a list, with one line saying what each answers.
3. Supporting pages link sideways only where the reader would want the next step, such as from a comparison to the how to page for the option they picked.

Write anchors that describe the destination, such as "how to track receipts for tax", not "click here". Put the link inside the sentence where the reader needs it. The audit that finds pages with no links in is in internal linking and the audit that finds the gaps.

Step 5: write the pillar last, or rewrite it last

The pillar has to describe what the other pages say. Write the supporting pages first, or rewrite the pillar once they exist. A good pillar has a direct answer in its first paragraph, a short section on each subtopic with a link to the full page, and a closing section on what to read next that names real pages.

Do not make the pillar a table of contents with no content. A page of links alone gives a reader and an assistant nothing to quote.

Step 6: publish in an order that helps each page

Publish the supporting pages with the most demand first, add the pillar when four or five supporting pages are live, then fill in the rest. Update the pillar's list every time you add a page. A cluster published all at once, with no sign of care in the links, reads as mass production. A cluster that grows over weeks reads as a site that knows its subject.

How to tell a cluster is working

• In Search Console, the pillar appears for the broad query and each supporting page appears for its own query, not for the pillar's.
• No two pages in the cluster show the same main query in the performance report.
• Readers move from page to page: your analytics show supporting pages leading to the pillar and to each other.
• The URL Inspection tool in Search Console shows the pillar as a page that refers to each new supporting page.

Check monthly. When two pages appear for one query, pick the stronger, fold the weaker into it and redirect.

Mistakes that break a cluster

• Writing eight pages that all answer "what is X" in different words.
• Linking only from supporting pages to the pillar and never back.
• Using the same anchor text on every link, which tells the engine nothing.
• Stuffing variants of a keyword into every page instead of answering a distinct question. Latent semantic indexing is often cited to justify this, and the real technique and the claim around it are different things.
• Leaving the cluster untouched for two years while the subject moves.

Questions about SEO topic clusters

How many pages does a topic cluster need?

As many as there are distinct questions on the subject, usually between six and fifteen supporting pages for a focused topic. Count questions, not words. Add a page only when it answers something the others do not.

Is a pillar page the same as a category page?

No. A category page lists products or posts. A pillar page is an article that explains the subject and routes the reader to the detail. A pillar can sit inside a category structure, but it carries its own content.

Should the pillar be long?

It should be as long as the overview needs. A pillar that answers the broad question in the first paragraph and then maps the subtopics is useful at any length. Padding it to reach a word count adds nothing.

Can I turn existing blog posts into a cluster?

Yes, and it is the fastest route. List your posts, assign each one main query, merge duplicates, write or choose a pillar, and add the links. Many sites find that half the work is merging posts that were competing.

How often should I update a cluster?

Review it every quarter. Check that links work, facts are current and the pillar lists every page. Update the oldest page first.

What an assistant does with a cluster on your subject

A buyer who asks ChatGPT or Perplexity a question inside your topic gets an answer assembled from a handful of pages, and whether yours is one of them depends on whether the page that answers that exact question is clear and reachable. A cluster gives you more distinct answers for assistants to pick from, but only a test shows which ones they choose. Ask each assistant five of the narrow questions from your map and record which sites it names and which pages it cites. AI Knows Us (aiknowsus.com) runs this check across ChatGPT, Gemini, Perplexity, Claude and Google AI Overviews, so you can see which supporting pages earn a citation and which need a rewrite. The planning side is covered in AI content strategy for a reader who might be an assistant.