Skip to main content
Skip to content
Back to the journal
The Unveil SEO journal / Guide

How to Build Topic Clusters, Based on a Study of 42 Sites

Articles in the topics whose pages link to each other the most are 21% more likely to rank in Google’s top 3. The method takes four steps.

November 19, 202614 min read
Groups of pages linked to each other within their topic, with a few links crossing from one group to the next

I set out to understand internal linking by testing its practices against data, at a scale I have not seen published. This study measures how 42 sites link their 60,356 articles to each other and sets those links against nearly 905,000 Google positions. Topic clusters are the first method put to that test. They are taught everywhere in SEO and, as far as I know, nobody has ever checked them with figures.

The idea behind them is to link the pages of a topic to each other. It holds up: articles in the topics whose pages link to each other the most are 21% more likely to rank in Google's top 3 than those in the topics that do so the least. However, their stricter rules fare worse, since neither walling a topic off nor building it around a pillar page goes with better rankings. The topic cluster that works best in this data is simpler than the one the guides describe.

I tested the three rules of topic clusters on 42 sites

That guidebook version, also called a content silo or semantic cocoon, adds three rules to the idea of linking a topic's pages. No link crosses from one topic to another. Each page belongs to a single topic. And all of the topic's links converge on a central pillar page. The first rule splits the guides. BruceClay.com's guide to siloing applies it to the letter, with links that "should not cross over to pages in different silos".

Ahrefs argues the opposite, that banning those links "doesn't help SEO or users". So who is right? Neither side has put its case to a test, and that test is what this study runs.

It covers 42 sites partly drawn at random, 20 in France and 22 in the United States. Our crawler collected their 60,356 articles and all the contextual links between them. These are links set in a sentence, as opposed to those in menus, footers or breadcrumbs. Topic clusters are built with them, and Ahrefs advises placing your most important links in the main content. The roughly 905,000 Google positions of these articles were collected from July to October 2026 through DataForSEO.

That left the question of which topic each article belongs to. The study does not reuse the sites' own categories. It rebuilds topics from the meaning of the texts, in groups of about 25 closely related articles, and lets one article belong to several groups. Each comparison then sets similar articles from the same site that receive as many links against each other. A small site weighs as much as a large one.

Three common beliefs about topic clusters fail that test. Walling topics off, filing each page under one topic or building around a pillar page does not go with better rankings. The gain comes from the links the pages of a topic exchange with each other.

PracticeWhat the data shows
Never link across topicsA link from another topic is worth about as much
One page, one topicOnly 2% of articles fit, and the most shared rank no worse
A pillar at the centreNo demonstrated gain for the topic or the pillar
Link the topic's pages to each other21% more chances of top 3, 15%* of first place, 49% more queries

How to read these figures. Each gap compares similar articles from the same site at the same moment. A gap is demonstrated when chance alone rarely produces one that large. An asterisk marks a gap that too few first places leave unconfirmed. The full method is at the end of this article.

Don't file pages into silos, their content already spans several topics

Each of these beliefs deserves a closer look, starting with filing. The silo version of topic clusters treats each page like a book shelved in a single place. Yet a page's content decides which topics it belongs to, and it is almost always several. Take an article on pricing a bathroom renovation. It is as much about quotes as about bathrooms. Intuitively, it seems a waste to lock it into one silo, since it would then be visible on one topic when it could be visible on several.

The data points clearly the same way. Only 2% of articles belong entirely to one topic. About seven in ten are shared between one and five topics, and that sharing does not hold them back. An article that spans ten topics is no less likely to rank in the top 3 or the top 10 than one that touches two. The most shared articles even appear on more queries.

Only 2.3% of articles belong entirely to one topic, and about seven in ten are shared between one and five topics.

A topic cluster is therefore built from content rather than from filing. List every page that covers the topic, including pages already filed elsewhere. They will still need linking, because internal links are what turn them into a topic.

Yet most pages lack those links. About two pages in three receive no link from another page of their topic. They rank lower than pages that get at least one, on the same site and with the same number of pages linking to them.

Pages linked from their own topic are 14% more likely to reach the top 3 (19.4% against 17.0%). They are also 12% more likely to reach the top 10 (31% against 28%) and appear on about 30% more queries. For first place, their lead reaches 11%* (8.9% against 8.1%).

* There are too few first places to confirm this gap, but the trend matches the other measures, which are confirmed.

A page with a link from its own topic reaches Google's top 3 more often, 19.4% against 17.0% overall, and at five levels of linking pages out of six.

The chart shows that this gap is not confined to one particular case. Pages linked from their own topic keep the lead whatever the number of pages linking to them, at five levels out of six, and draw level at the sixth. The top 3 gap may look small. For a single SEO lever, I find that a lot, all the more as the fix comes down to one link per page.

That link has to come from a closely related page of the same topic, which is where the task gets harder. Spotting by hand the pages that lack one means knowing every topic each page belongs to. In Unveil SEO, the topic clusters feature does that spotting. You type a subject, and it lists the pages closest to it and suggests links between them.

My reading of the results is that the origin of a link counts mostly for the first one. Once a page has a link from its topic, its next links can come from anywhere.

This first link has its limits, though. In large topics it matters less than how much the pages link to each other, and that is where the study's main result lies.

Those exchanges between the pages of a topic are at the heart of the main result. For the same number of links, articles in the most interlinked topics are 21% more likely to reach the top 3 (17.3% against 14.3%). That figure comes from the share of each topic's links that connect two of its own pages. The third of topics where it is highest (18% or more) is compared with the third where it is lowest (10% or less).

OutcomeTopics linked most inside, against least
Top 321% more chances (17.3% against 14.3%)
First place15%* more chances (7.7% against 6.7%)
Top 1014% more chances (28.2% against 24.7%)
Number of queries49% more queries

* There are too few first places to confirm this gap, but the trend matches the other measures, which are confirmed.

Articles reach Google's top 3 more often in the most interlinked third of topics, 16.6 to 18.4% against 14 to 15% in the other two thirds.

The advantage is not spread evenly. On the chart, the top 3 rate barely moves until a topic keeps about 18% of its links inside, then it climbs. Yet most topics stay below that level. A typical topic keeps 10 to 15% of its links between its own pages.

Two thirds of articles sit in topics that keep less than 18% of their links between their own pages, where the top 3 rate stays at 14 to 15%, against up to 18.4% above.

I would not turn 18% into a target, though, since larger topics naturally keep more of their links inside. The data gives a direction, and the top 3 rate still rises across the top third. Moving in that direction takes few links, because pages have only about 3.5 links each in a typical topic. In a topic of 75 pages, one link added per page more than triples the share exchanged inside.

The gap also widens as the topic grows. In topics of about 150 pages, those twice as linked inside as usual have 35% more chances of reaching the top 3 (18.2% of their articles against 13.5%).

Does that mean closing the topic off from the rest of the site, as the first rule of topic clusters demands? Closing a topic adds nothing, and neither does building it around a pillar page.

On closing topics, Ahrefs argues that a silo structure makes no sense once it bans the links readers need, and my measurements point the same way. Once a page has a link from its own topic, a link from another topic is worth about as much. Each time its links from other topics double (from none to one, from one to three), it has 7% more chances of reaching the top 3 and of ranking first, about what its next links from its own topic are worth.

The page that sends that link loses nothing either. Each time its links to other topics double, it has 4.5% more chances of reaching the top 3 and 8.5% of ranking first. What counts for a topic, then, is the links its pages exchange, not the ones it holds back. Each time the links exchanged between its articles double, they have 18% more chances of reaching the top 3, while sending fewer links outside goes with no gain.

The same holds for site sections, meaning the categories a site shows in its URLs, such as /marketing/ or /sales/. On the 24 sites that have several, a link from another section is worth about as much as one from the same section.

Links that cross those borders are the norm anyway, since 74% of links connect two different topics. None of the 42 sites strictly silos its topics by meaning, so the comparison is between more and less open topics. When a link changes topic, BruceClay.com's guide points it to the other silo's landing page. I would rather link straight to the article that fits, since a link to any article goes with better rankings for that article.

A pillar page brings no demonstrated gain

The pillar page, the third rule of topic clusters, fares no better than closing topics. It was measured in two ways, as a topic whose links converge on one central page, and as the central page itself. Convergence goes with no demonstrated gain in the top 3 or the top 10, and first place is too imprecise to judge. Being the central page goes with no demonstrated advantage over other well-linked articles either.

Being linked by closely related articles, on the other hand, goes with a better top 3, with 11% more chances each time the number of those articles linking to the page doubles. Being well surrounded seems to count more than sitting at the centre. A pillar page can still serve navigation or conversion, which is a separate matter from rankings.

These results come from blog articles, and a category page that sits above its products on an e-commerce site remains a different case. Unveil SEO lets you build your topic clusters with or without a pillar page, with no fixed structure. I suggest starting without one, to leave more room for links between the pages.

Build your topic clusters in four steps

This topic cluster with no fixed structure takes four steps to build, and they follow what the data shows, in order.

  1. 1

    List the pages that cover the topic

    Start from a subject and list every page that covers it, including pages already filed under another topic or section.

  2. 2

    Find the pages that receive no link from another page of the topic, and give each one a link from a closely related page.

  3. 3

    Add links between pages of the topic, more so in large topics. The gain shows in the most interlinked third of topics, so aim well above the usual level.

  4. 4

    Leave in place every link that points to another topic. The pillar page is optional.

These steps sit on top of internal linking, the one SEO lever a site fully controls. Start with the topic that matters most to your business, find its pages with no link from inside, and link them first.

How the study was run

Before you start, keep in mind what kind of figures these are. This is an observational study, which shows what goes with better rankings on real sites at one point in time. Only a before-and-after test can show what causes them.

  • The 42 sites come in three groups, 20 French sites drawn at random among the most visited sites or picked for their structure, 12 US sites drawn at random the same way, and 10 US sites from my earlier studies.
  • The four outcomes are ranking in the top 3, in first place or in the top 10 for at least one query, and the number of queries an article appears on.
  • Topics come from the meaning of the articles, measured with embeddings and grouped into sets of about 25. An article has a share in each topic its closest neighbours fall into.
  • Similar articles come from the same site, section and age bracket, with comparable links received, length, backlinks and breadth of subject.
  • The comparison sees the advantage of being linked, the subject of my earlier study of contextual internal links. An article that receives at least one link has 29% more chances of ranking in the top 3 (17.3% against 13.4%) and 24% of ranking first (7.8% against 6.3%). If one of the three rules of topic clusters went with a gain of that size, it would have caught it.
  • A gap is demonstrated when its 95% range sits entirely above zero. For the main result, that range runs from 4% to 38%.
  • The main result holds when any one of the 42 sites is taken out of the calculation, and it stays demonstrated when each article counts for one instead of each site. When I regroup the same articles with a different algorithm, the links kept inside those groups go with no demonstrated gain, which ties it to the real topics.
  • "Worth about as much" is a measured claim too. For links from the same topic rather than another, the range rules out any advantage above 10% in the top 3, in first place and in the top 10.

Frequently Asked Questions

Mathias Decourt

Written by

Mathias Decourt

Co-Founder - CEO

Website performance specialist helping businesses identify the few actions that truly move the needle. Turning complex data into clear, actionable insights that drive growth.