What a semantic cocoon is, and how it structures a site that ranks

A semantic cocoon organises a site's pages around a single topic: a pillar that frames it, zooms that each address one search intent, and internal linking that connects the parent to its children, then siblings to one another. This guide is for B2B marketing and SEO leads who want a set of pages to rank, get cited by AI and stay manageable: definition, principles, architecture, rules, benefits and limits, and the method to fix the mistakes.

Definition: what a semantic cocoon is, and what it is not

A semantic cocoon is an organisation of pages dedicated to a single topic, in which each page answers a distinct search intent and is given a role, pillar or zoom. Contextual internal links connect the pillar to its zooms, the zooms to the pillar, and the zooms to one another when the reader needs it.

This definition rests on three inseparable elements:

• structured coverage of the topic, from the general to the specific;
• a role assigned to each page according to the intent it targets: inform, help compare, move towards a decision;
• an explicit internal linking logic that makes the hierarchy and the reading path visible.

Without any one of these three elements, you get an editorial grouping, not an architecture.

What a semantic cocoon is not

The most useful boundary lies in intent: the cocoon decides which page answers which question, and which page the reader should go to next. It should therefore not be confused with:

• a category and its articles, with no linking intent and no reading progression;
• categories and tags, which classify pages whereas the cocoon orchestrates them;
• a watertight silo, where each section stays isolated, with no bridge to neighbouring topics;
• a diagram that is never translated into the site's actual links;
• a pillar that repeats everything and cannibalises its zooms instead of directing readers to them.

Origin and equivalents: cocoon, silo and topic cluster

The semantic cocoon was popularised in France by Laurent Bourrelly, who turned it into an architecture method based on semantic proximity between pages. English-speaking SEOs talk about the topic cluster or the pillar-cluster model; the older silo refers to a vertical split of the site into sections. The three notions look alike, but the emphasis shifts:

The practice used here takes its logic of meaning from the cocoon and its simplicity from the topic cluster: one topic, one pillar, zooms, and no watertight silo. You link outside the cocoon whenever it helps the reader, without trying to "lock in" authority.

The 3 tests for a page to belong to the cocoon

Three quick tests check whether a page belongs to the cocoon. They do not aim for a perfect score: they spot pages that sit outside the structure, whether orphaned, redundant or without a clear role.

• Relevance test: the page can be summed up in a single promise, without a string of "and"s. If you have to list three objectives, it is too broad or badly placed in the hierarchy.
• Non-redundancy test: the page brings at least one angle, example, approach or clarification that the other pages of the cocoon do not. Otherwise, it dilutes signals and risks cannibalisation.
• Linking test: the page receives a link from its parent and sends useful links, to the pillar, to a sibling if the reader needs it, to the next step of the journey. A page that receives no internal link does not belong to the cocoon in practice, even if its topic is relevant.

A page that fails one of the tests does not necessarily disappear. It is merged, repositioned or taken out of the cocoon, depending on the decision made in the recovery plan set out below.

Architecture: one pillar, zooms, two levels

The recommended architecture has two levels (three at most): a pillar (T0) that frames the topic, zooms (T1) that each address one intent, and sometimes second-tier zooms (T2) when a T1 branches out. Each role has a scope and, above all, an out-of-scope:

The pillar: direct readers without copying

The pillar serves as a stable entry point. Its job is not to say everything, but to give an overview and direct readers: a short definition, the principles, then one paragraph and one link per sub-topic. A pillar that repeats the content of its zooms competes with them. It is the most common mistake, and the most costly, because the search engine no longer knows which page to serve for which query.

The zooms: one intent, one direct answer

Each zoom answers a precise question, with an unambiguous promise. It does not need to be long, but it must be sharp: consistent vocabulary, a stable angle, the answer at the top of the page. A zoom is never isolated. It links back up to the pillar and across to a sibling when that immediately helps the reader, to clarify a choice or complete a point.

When to add a third level

A third level is justified when a zoom covers several sub-questions that each call for their own outline. Beyond that, click depth increases with no benefit: important pages must stay reachable in a few clicks from the pillar, and no page should become orphaned. When in doubt, a well-titled section in the T1 is better than one more T2 page.

The rules that hold a cocoon together

Five rules keep the cocoon readable for search engines and manageable for the team:

• One page per search intent. If two queries call for the same outline, the same examples and the same call to action, they are one page.
• One owner per sub-topic. A notion is developed in only one place; the other pages sum it up in one sentence and link to it.
• No magic number. A pillar has as many zooms as there are distinct intents, typically 4–12.
• Two levels, three at most. A pillar and its T1 zooms, with T2 zooms only when a T1 branches out.
• No watertight silo. You link outside the cocoon when it is useful, with no PageRank sculpting, no link obfuscation and no internal nofollow.

Applying the merge rule

The merge rule (same outline, same examples, same call to action = one page) is tested on real queries, not on wording. "Semantic cocoon definition" and "what is a semantic cocoon" call for the same outline and the same examples: they belong on a single page. "Create a semantic cocoon" calls for a step-by-step outline and a different call to action: it is a separate zoom. When two existing pages fail this test, you choose the reference page and redirect the other one to it.

The topic owner, guarantor of non-redundancy

For each sub-topic, you designate the page that develops it. The other pages mention it in one sentence and link to it with a descriptive anchor. This decision is made at the brief stage, not at proofreading: a solid brief specifies the dominant intent, the unique angle, the expected proof and what the page must not cover. Non-redundancy is won before writing, not after.

Cocoon linking in 5 lines

Linking is what makes a cocoon: without it, you have a collection of pages, not an architecture. Five lines set out how it works:

These five lines are enough for a cocoon. Where to place links on the page, how to choose anchors and how to audit orphan pages are covered in the internal linking guide, which applies to the whole site.

What a cocoon delivers, and what it does not

A cocoon does not force a ranking. It makes the topic readable: the search engine better understands the themes covered, and the reader finds the right level of answer faster. The table separates what you can expect from it from what depends on other levers:

On the SEO side: coverage and internal authority

The first benefit is coverage: the zooms capture precise, long-tail queries, then send readers on to the pillar. The second is internal authority. On most sites, external links are scarce: 94–95% of pages have no backlinks at all (Backlinko, 2026). Internal linking then becomes the main lever for moving the available authority towards the pages that matter.

Publishing is not enough, though: 91% of pages never reach Google's first page (SEO.com, 2026). A cocoon does not erase this, but it avoids adding to it with pages that have no intent of their own and compete with one another. Further benchmarks are gathered in the SEO statistics.

On the reader and team side: journey and governance

In B2B, readers rarely move in a straight line: they discover, compare, validate, and only then consider taking action. The cocoon offers them the next step at the right time, without them having to look elsewhere for a sub-question the site could have covered. For the team, the benefit is stability: you depend less on a single star piece of content, each new page adds an entry point without weakening the others, and updates are made by page group.

Cocoons and AI answers: each zoom answers its own question

Generative engines (AI Overviews, AI Mode, ChatGPT, Perplexity, Gemini) no longer return just a list of links: they rebuild an answer from passages extracted from several pages. At the time of writing, AI Overviews and AI Mode have been available in France since summer 2026. A well-divided cocoon gives them what they are looking for: one page per question, with the answer at the top.

The structure of each page matters: pages with an H1-H2-H3 hierarchy are 2.8x more likely to be cited by AI (State of AI Search, 2025). A zoom with a clear outline, lists and tables lends itself to extraction better than a pillar that mixes every intent.

Organic ranking does not explain everything: 60% of AI citations come from URLs outside the organic top 20 (State of AI Search, 2025). A precise page that answers a sub-question better than another can therefore be cited without appearing at the top of the results. That is precisely the role of a zoom.

To make each zoom citable, four habits apply:

• place a direct answer of two or three sentences at the top of the page and of each section;
• structure with explicit H2 and H3 headings, with lists and tables;
• date your proof and cite its source;
• update the zooms that carry figures or practices that change over time.

These levers go beyond the cocoon: the guide to Generative Engine Optimization (GEO) details them for the whole site.

Common mistakes and recovery plan

Mistakes rarely come from a lack of content. They come from pages added without a role, without a distinct intent or without links. The most common are:

• a pillar that copies its zooms and competes with them;
• several pages on the same intent, with titles and angles that are too close;
• competing definitions rewritten across several pages, instead of one owner per sub-topic;
• orphan zooms that receive no link from their parent;
• links dropped in as a generic block, unrelated to the passage being read.

Deciding page by page: create, optimise, merge, deindex

When a cocoon exists but runs on empty, you decide page by page, based on intents and Search Console data:

• Create when the intent is not covered, generates impressions without a dedicated page and matches a step of the journey.
• Optimise when the page exists and generates impressions but has plateaued: structure, proof, angle, links.
• Merge when several URLs address the same intent, then redirect to the reference page.
• Delete or deindex when the page answers no useful intent; redirect if an alternative exists, otherwise relink to avoid dead ends.

Measure by page group

A cocoon is measured as a whole. In Google Search Console, you group the cocoon's URLs and track the group's clicks, impressions and average position, before and after each batch of changes. On the AI side, you track citations question by question. The useful signals are the number of pages generating impressions, the appearance of new queries and the pillar's progress. These readings establish a correlation, never a causation.

Depending on your situation

The full method has its own guide: create a semantic cocoon step by step, from scoping the topic to publishing in batches, in a fixed order (scope the pillar first, write it last). To see the architecture applied, the semantic cocoon examples detail three typical cocoons, each with its structure laid out in a table.

Depending on your starting point, the next read changes:

• you are starting from a list of queries: first sort them by search intent, before drawing the site structure;
• you need to choose which topics deserve a cocoon: build them into your editorial strategy;
• your site already has dozens of pages on the topic: an SEO audit spots duplicates, orphan pages and pages without a role before you create anything;
• you need to organise the production of the zooms: editorial content production describes the workflow and the roles.

Before you set the pillar and its zooms, you need to know which pages exist and which keywords they capture: for this, the Incremys GEO & SEO platform offers an audit that maps keywords, pages and AI presence by keyword.

FAQ about the semantic cocoon

What is a semantic cocoon?

It is an organisation of pages dedicated to a single topic. A pillar frames the topic, zooms each address a distinct search intent, and contextual internal links connect the pillar to its zooms, the zooms to the pillar and the zooms to one another when the reader needs it. The aim is to make the topic readable for search engines and for readers.

How many pages does a cocoon need?

There is no magic number. A cocoon has one pillar and as many zooms as there are distinct intents on the topic, typically 4–12. If two pages call for the same outline, the same examples and the same call to action, they are one page. Five sharp zooms are better than fifteen pages that copy one another.

Semantic cocoon vs silo: what is the difference?

A silo splits the site into watertight sections, linked mainly vertically. A cocoon also connects the parent to its children, but allows links between siblings and to other topics whenever they serve the reader. A silo isolates themes; a cocoon organises a topic without closing it in on itself, which avoids navigation dead ends.

Does a cocoon help you get cited by ChatGPT?

It helps without guaranteeing it. Generative engines extract passages that answer a question directly. A cocoon provides one page per question, with the answer at the top, a clear H2/H3 structure, lists and tables. These pages lend themselves to extraction better than an article mixing several intents, but the quality of the answer remains decisive.

Does a small site need a cocoon?

Yes, as soon as a topic has several distinct intents. A small cocoon, one pillar and four zooms, is enough to clarify which page answers what and to stop two pages competing. The logic matters more than the volume: each page has a role, an intent and consistent links. You then expand according to the queries actually observed.

How long before you see results?

The timeframe depends on the sector, the site's history and the competition. The first signals are intermediate: more pages in the group generate impressions, and new queries appear. Gains in clicks and rankings come next, with updates and adjustments to the linking. You read them by page group, before and after each batch.

Semantic cocoon vs internal linking: what is the difference?

Internal linking refers to all the links between a site's pages, whether or not they follow a strategy. A cocoon is an editorial architecture that uses linking according to precise rules: parent → all its children, child → parent, siblings when useful. Every cocoon relies on internal linking, but not all internal linking forms a cocoon.

Can you build a cocoon on an existing site?

Yes, and it is the most common case. You start by sorting pages: those to keep, reposition, merge and redirect. You then designate the pillar and the owner of each sub-topic, and rebuild the links around this hierarchy. Finally, you add the missing zooms, according to the intents not yet covered.