What a Semantic Cocoon Is, and What It Is Not
A semantic cocoon is a page architecture organized around a primary search intent, where each page addresses a distinct aspect of the topic and links to others according to a strict hierarchical logic. It is neither a blog category, nor a navigation menu, nor an accumulation of articles on the same theme.
The most common confusion is believing that a cocoon simply means linking pages that talk about the same thing. Linking is a consequence of the architecture; it is not its definition. What defines the cocoon is the segmentation: each page answers one and only one query, and the whole covers a semantic field without two pages competing for the same intent.
The term comes from Laurent Bourrelly’s work on information architecture. The underlying principle is older and perfectly documented by Google: a search engine understands a site by its link structure as much as by its words. An isolated page that receives few internal links is treated as peripheral, regardless of its editorial quality.
A cocoon is designed before a single line is written. Rebuilding the architecture after publishing forty pages costs two to three times more than establishing it from the start, because it then requires handling redirects, existing cannibalization, and external links already pointing to incorrect URLs.
The semantic cocoon contrasts with the dominant practice of publishing content haphazardly, hoping that sheer volume will eventually produce results. Thirty structured pages outperform two hundred stacked pages, and they cost less to produce. This is the principle we apply to every natural referencing service, even when the client comes with a history of existing content to rework.
Why Architecture Trumps Content Volume
The reasoning lies in how authority is distributed within a site. Understanding this mechanism is enough to explain why two sites with comparable content achieve opposite results.
How Internal Authority Circulates
Each page has a capital of popularity, fueled by links it receives from external sources and from other pages on the site. It transmits a portion of this capital to each page it links to. A page that emits forty links dilutes this capital across forty destinations. The same page emitting six links concentrates it.
On most sites we audit, the homepage concentrates the bulk of external popularity and then disperses it across a menu of fifty entries, a footer with thirty links, and a sidebar of recent articles. The result is mechanical: no internal page receives enough authority to rank for a competitive query.
A semantic cocoon reverses this logic. The pillar page captures the capital, transmits it to a small number of truly relevant child pages, and these child pages link back to the mother page and their direct siblings. The capital circulates within the silo instead of evaporating into general navigation.
Click Depth: An Underestimated Indicator
Depth is measured by the number of clicks required to reach a page from the homepage, following the shortest path. Beyond three clicks, exploration becomes rare, and indexing becomes irregular. Beyond five clicks, a page can remain unvisited by the crawler for months on a site with average authority.
A cocoon-based tree structure mechanically guarantees shallow depth, because each level has only a limited number of children. In contrast, an article library sorted by date reaches depths of six or seven from the second year, with older content ending up on page eight of a pagination that no one browses.
Building a Cocoon from A to Z: The Six-Step Method
The sequence matters as much as the content produced. Each step conditions the next, and skipping the first renders the other five useless.
Step One: Map Search Intentions
The starting point is not a list of keywords; it’s a list of questions your customers actually ask, in the order they ask them. A business leader looking for an agency rarely starts with ‘web agency’: they start with ‘how much does a website cost,’ then ‘showcase site or e-commerce,’ then ‘how to choose a provider.’
This mapping is built using Search Console data if the site has a history, with search suggestions and related questions otherwise, and especially with what sales representatives hear on the phone. This last source is the richest and least exploited.
Step Two: Define the Pillar Page
The pillar page targets the broadest generic query for the covered field—the one with the most volume and highest competition. It doesn’t aim to say everything: it sets the framework, provides a complete but synthetic answer, and distributes to child pages for details.
Its classic mistake is exhaustiveness. A twelve-thousand-word pillar page that covers everything absorbs the intent of its own child pages and prevents them from ranking. The right measure is between two thousand and four thousand words, with a clear transition to each sub-page.
Step Three: Segment Child Pages Without Overlap
Each child page addresses a distinct intent, verifiable by a simple test: if two pages could answer the same query typed into Google, the segmentation is poor, and they should be merged. This test prevents cannibalization, which is the most common failure mode for hastily built cocoons.
The number of child pages is generally between four and eight. Below that, the cocoon lacks sufficient material. Above eight child pages, the pillar page disperses its capital, and it becomes preferable to introduce an intermediate level.
Step Four: Write the Pillar Page Before the Child Pages
The order of production is not neutral. The pillar page defines the vocabulary, level of detail, and boundaries of each child page. Writing it first avoids overlaps and provides an anchor point for the upward links that each child page will need to place.
In practice, many teams start with the easy pages and postpone the pillar page, because it is the most demanding to write. The result is a headless silo for several months, where child pages link to each other without ever concentrating capital on the generic query.
Step Five: Implement Linking During Writing
Internal links are not added as an afterthought; they are written with the text. A link placed in a sentence that naturally calls for clarification reads better, gets clicked more often, and transmits a clearer semantic signal than a link mechanically inserted during a review pass.
Concretely, each new page starts with a list of three to five destinations to cover: the mother, two siblings, possibly a service page. The writing process seeks the place where each integrates naturally. If no natural place exists for a given destination, it’s a sign that the link should not have been there.
Step Six: Measure and Adjust Quarterly
A semantic cocoon is managed using three indicators. The average position of the pillar page for its generic query indicates whether the silo captures broad intent. The number of silo pages receiving at least one impression indicates whether semantic coverage is working. Page alternations for the same query signal cannibalization issues to address.
These three reports are generated in Search Console by filtering on the silo’s directory. A quarterly review is sufficient; a monthly review produces noise without additional information, as SEO movements are best appreciated over longer periods.
Internal Linking Rules Within the Cocoon
Once the architecture is established, linking follows three simple rules, strict adherence to which makes the difference between a functional cocoon and a haphazard collection of linked pages.
Sibling Page Links: The Most Misapplied Rule
A child page links to its mother page, to its immediate siblings, and to its own children if it has any. It does not link to a page in another silo, unless there is a strong editorial reason, because each link leaving the silo drains capital out of the cocoon.
In practice, two to four links to siblings are sufficient. Placing them within the body of the text, where the subject naturally calls for clarification, produces a much stronger signal than a block of related links at the end of the page, which search engines weigh lightly and readers ignore.
Anchors: Neither Exact Everywhere Nor Generic Anywhere
An anchor describes the target page and contains its keyword, but varying the phrasing is preferable to mechanically repeating the same expression. An internal anchor profile composed entirely of exact match anchors is an artificial signal, whereas natural variation strengthens the lexical field.
Empty anchors remain forbidden. “Click here,” “learn more,” and “this link” transmit no semantic information and waste a signal that nothing else can replace.
Orphan Pages: The Invisible Leak
An orphan page receives no internal links. It exists in the sitemap, Google knows about it, but it receives no capital and behaves as if it weren’t there. On a site with more than a hundred pages, there are almost always some, usually old content whose parent page was deleted during a redesign.
Their detection involves a full crawl compared to the list of URLs in the sitemap. This is one of the twelve points we systematically check during an SEO technical audit, and it is often the one that produces the quickest gain, as it only requires adding links to reactivate already written content.
Link Density: A Matter of Common Sense Rather Than Quota
No official threshold exists for the number of internal links per page, and circulating figures are approximations. The useful criterion is readability: a reader should be able to follow the text without being interrupted every two words by a blue anchor. In a three-thousand-word content piece, about ten well-placed internal links go unnoticed; thirty turn the text into a directory.
Position on the page matters as much as the number. A link placed in the first third of the content, within a paragraph that develops the target page’s subject, carries more weight than an identical link relegated to the end of the page. This internal hierarchy explains why automatically generated blocks of related articles produce so little effect.
A Three-Level Tree Structure Example
The principle becomes concrete with a real case. Let’s take the “website creation” silo as we built it for our own site.
At the first level, the pillar page covers website creation as a whole: what a professional site is for, what distinguishes a serious project from a purchased template, how a project unfolds, and what the budget ranges are. It answers the generic query and directs to details.
At the second level are the typology pages, each addressing a distinct intent: the showcase site, the online store, custom development, the redesign. A business that already knows what it wants lands directly on the correct page from Google, without going through the pillar page. Each links back to the mother page and its two closest siblings.
At the third level are the variations by profession, grouped under a website creation by sector page, which itself acts as an intermediate pillar for thirteen professional pages. A restaurateur looking for a site for their establishment finds a page that discusses reservations, menus, and customer reviews, not a generic page.
The same reasoning applies to geographical areas, with one page per market served. The rule governing this third level is real differentiation: a professional page that merely replaces the sector name in generic text produces near-duplicate content and weakens the entire silo. Each page must address the specific challenges of its profession, its expected functionalities, and its specific queries.
Errors That Cause a Cocoon to Fail
Five failures systematically recur, and none involve advanced technical skills.
- Building the cocoon in the menu rather than in the content. A three-level dropdown menu does not make a cocoon. The links that matter are those placed in the body of the text, contextualized, with a descriptive anchor.
- Publishing the pillar page last. Many start with the easy pages and postpone the pillar page. The child pages then remain without a mother for months, and the silo does not exist.
- Mixing two intentions on the same page. A page that explains a concept and sells a service does neither well. Separate informational from transactional content, and link them.
- Neglecting the speed of deep pages. A perfect cocoon on a slow site produces nothing, because the crawl budget is exhausted before reaching the third level. Core Web Vitals determine how often Google returns to crawl.
- Never revising the architecture. A cocoon is a living entity. Each new page must find its place in the existing hierarchy; otherwise, it becomes an orphan or cannibalizes a sibling.
These five points explain the majority of cocoons that produce no results after a year. None are corrected by adding content; all are corrected by revisiting the structure.
A sixth error deserves to be added because it concerns sites that have done well: forgetting that the cocoon must serve conversion as much as SEO. A perfectly structured silo that never prompts action produces traffic but no clients. Each child page must include a natural exit to the corresponding service page, formulated as a logical continuation of the reading and not as an advertisement banner.
This exit is placed where the reader has understood the problem and is looking for a solution, generally after the section that explains the method. Placed too early, it interrupts the demonstration and degrades the content’s credibility. Placed only at the end of the page, it only reaches the minority who have read everything.
How Long and How Much It Costs
The question always arises, and the honest answer depends on the starting state. For a new site, designing the tree structure represents two to four days of work; content writing then follows at a pace the budget allows.
For an existing site, you must add an audit of the current state, detection of cannibalization, a plan for merging redundant pages, and a redirect plan. This recovery phase explains the cost difference between a cocoon implemented during a website redesign and the same work carried out three years later.
The first effects on rankings appear between the third and sixth month, once Google has recrawled everything and recalculated the internal distribution. The clearest gains almost always concern second-level pages, which move from the third page to the first when they finally receive the capital the structure intended for them. For budget estimates on creation, the article detailing how much a website costs in 2026 provides the complete breakdown, and the price simulator gives an estimate in minutes.
One last point on effort distribution. For a given budget, we dedicate approximately one quarter of the time to architecture and intent research, and three quarters to production. Projects that fail almost always reverse this ratio, dedicating half a day to the plan and the rest to writing. Yet, the plan is the only part of the work that, if done poorly, renders everything else ineffective.
The Cocoon Applied to a Local Service Site
The model described so far applies to editorial content. For a business site selling a service in a given area, the segmentation follows a different logic worth detailing, because it is the most frequent case among our clients.
Three axes intersect: the service, the client’s profession, and the geographical area. Addressing all three in the same tree structure produces an unmanageable number of combinations, and especially nearly identical pages that cannibalize each other. A roofer in Toulon and a roofer in Nice do not need two pages saying the same thing by changing the city name.
The rule we apply is to choose a primary axis and only decline that one. In a market where competition is based on profession, the axis is sectoral, and geography appears only in the content, not in the tree structure. In a market where local anchoring is the primary selection criterion, such as local SEO for local businesses, the axis is geographical, and the profession remains transversal.
The second axis is then handled through content and linking, not by multiplying pages. A professional page can perfectly mention the areas served, with a link to the corresponding geographical page, without needing to create the Cartesian product of the two. This discipline avoids the pitfall of hundreds of automatically generated pages, which inflate the number of URLs and degrade the perceived quality of the entire domain.
Key Takeaways
A semantic cocoon is not a linking technique; it is an architectural decision made before production. It involves segmenting a topic into distinct intentions, prioritizing these intentions, and circulating authority within the silo rather than dispersing it through general navigation.
Three checks are enough to know if your site has one. Is each strategic page reachable within three clicks from the homepage? Do two pages compete for the same query in Search Console? Are there pages that receive no internal links? Three unfavorable answers signal an architecture that needs to be redone, and the rework will produce more effect than six months of additional publishing.
The rest is a matter of discipline: one intent per page, descriptive anchors, links within the body of the text, and a revision of the hierarchy with each addition. The projects we deliver in our achievements all start with this tree structure, defined before the first mockup.
