Topic clusters: the hub, the spokes, and the links that make them one thing

Three things: a hub page about the whole subject, several pages about one part of it each, and links running both ways. The pattern is also called a pillar page with its cluster, or hub and spokes. The names do not matter; the links do.

• The hub answers the broad query ("search console") and says what the subject is, what the parts are, and in what order to work through them.
• Each spoke answers one specific query ("crawled currently not indexed") completely, so it never needs the hub to make sense.
• Every spoke links up to the hub; the hub links down to every spoke; spokes link across where one genuinely depends on another.
• One page per query cluster, so no two spokes compete for the same search.

Google does not have a "cluster" feature. What it has is a system that reads internal links to understand structure, and pages that rank on specificity. A cluster is a way of building for both at once.

Most hubs fail the same way: a heading, two sentences and twenty links. That page has no reason to rank for the broad query, because it answers nothing, and nobody links to it, because there is nothing to cite.

A hub is the page you would write if someone asked you to explain the subject in ten minutes. The links are how they continue afterwards, not the content.

Below four or five spokes, the hub and the spokes compete: the hub covers everything thinly and each spoke repeats it. Build the spokes first, and add the hub when the subject has enough parts that a reader would get lost without a map.

1. 1

   Write two or three spokes

   Pick the specific queries you can answer best. They rank on their own, because specificity is what a small site wins on.

2. 2

   Watch the queries they pick up

   In Search Console, the Performance report shows the searches those pages appear for. The broad ones you are shown on and rank badly for are the hub's query.

3. 3

   Write the hub when there are five or more

   Now the hub has something to organise, and something to link to that already ranks.

4. 4

   Add spokes as the questions arrive

   Every new spoke gets linked from the hub the day it is published, not in a later tidying pass that never comes.

There is no upper limit that matters. A subject with thirty parts is a subject with thirty spokes, as long as each part is a real question somebody asks.

The links are the cluster. Get them wrong and you have a folder of unrelated pages.

1. Every spoke links to the hub, once, high on the page, with the subject as the anchor text: "page speed" rather than "click here" or "our pillar page".
2. The hub links to every spoke, grouped, with one line of context each.
3. Spokes link across only where one needs the other: the guide on layout shift links to the guide on image dimensions because the fix lives there.
4. The breadcrumb on each spoke names the hub, which gives a second path and marks the hierarchy.
5. Nothing links to a page that does not exist yet, however good the plan.

The anchor text is the part people get wrong. Internal anchors are fully under your control, so write them as the subject of the page you are linking to. That is the description search engines read, and it is what tells a reader whether to follow.

Take a fictional product, yourproduct.com, that helps people import data. The subject is data import, and the cluster looks like this.

The hub ranks for the broad query because it explains importing data properly. The spokes rank for the specific ones because each answers one completely. The product page gets linked from the spoke where limits are the reason someone pays.

• Two spokes on the same query. They split the impressions and neither wins. Merge them and redirect one.
• A hub that repeats the spokes. The hub says what the subject is and how the parts fit; it does not explain each part.
• Links added only at the hub. A spoke with no link back is an orphan in everything but name.
• A cluster built from a keyword tool rather than from questions. If you cannot say who asks the question and why, the page has no reader.

Judge the cluster, not the pages. A spoke that loses a little while the hub gains and the group gains overall is a cluster doing its job.

1. 1

   Filter the Performance report to the folder

   Set a page filter that contains the cluster's path. That is the cluster's total impressions and clicks.

2. 2

   Compare periods

   The last 28 days against the 28 before, then against the same period last year once you have the history.

3. 3

   Read the queries, not just the totals

   New queries appearing is the cluster earning breadth. The same queries with better positions is it earning depth.

4. 4

   Check the hub separately

   If the hub has impressions on the broad query and no clicks, the title is the problem, not the cluster.