How to Map Topic Clusters and Entities for AI Visibility

Why does a topic and entity map help AI visibility?

AI answers are assembled from passages retrieved for specific questions and sub-questions. A site mapped around entities and topics helps in three ways:

• Coverage: each buyer question has a page or section that answers it, so there is something to retrieve.
• Clarity: each page has one main subject, which makes it easier to match to questions and harder to confuse.
• Connections: internal links show how products, audiences, problems and places relate, which helps engines understand the business as a whole.

Without a map, sites tend to grow overlapping pages that half-answer the same question, and leave other important questions unanswered.

How do you build the map?

1. List your core entities: - the organisation itself; - products or services; - audiences or segments, such as small agencies, clinics or enterprise IT teams; - problems or jobs, such as payroll errors, staff scheduling or compliance; - places, if you serve specific regions.
2. Collect buyer questions for each entity from sales, support, reviews, Search Console and AI answer checks.
3. Group questions into topics: questions that share one answer belong to one topic.
4. Assign each topic one home: a dedicated page, or a section on a broader page.
5. Choose pillar pages: usually one per main product, service, audience or problem.
6. Draw the links: from each pillar to its topic pages, back to the pillar, and across related topics.
7. Mark gaps and overlaps: topics with no page, and pages competing for the same topic.

A spreadsheet with columns for entity, topic, main question, page URL, pillar and status is enough.

What does an entity and topic map look like?

An illustrative excerpt for a scheduling software company:

Each row is one question with one home. Gaps show up as empty page cells; overlaps show up as the same question in two rows.

How do you build pillar pages for AI search visibility?

A pillar page is the main page for an entity or broad topic, linking to more specific pages. For AI answers, a good pillar is not just a list of links. It should:

• answer the broad question directly at the top: what the product is, who the service is for, what the problem is and how to solve it;
• summarise each subtopic in a short, self-contained section, with the key fact and a link to the full page;
• include the facts people ask most about the entity, such as prices, eligibility or key features, stated plainly;
• link to every page in the cluster with descriptive anchor text;
• stay current, since it answers the broad questions.

A pillar that summarises well can be cited for broad questions, while its cluster pages are cited for specific ones. Whether a single strong page or a cluster works better for a given topic is discussed in whether topic clusters or single focused pages perform better for LLM visibility.

How should clusters be linked?

• Pillar to cluster pages: every topic page linked from its pillar, with anchor text that names the topic.
• Cluster pages back to the pillar: near the top or in a clear "part of" note.
• Across clusters where the relationship is real: a restaurant scheduling page linking to the tip-tracking integration page, for example.
• From your homepage and main navigation to the main pillars.

Descriptive anchor text such as "tip tracking for restaurant rotas" tells readers and engines what the linked page covers.

How do you avoid overlap between pages?

Overlap splits signals and makes engines pick between near-identical pages. To avoid it:

• give every topic a single home in the map, and redirect or merge duplicates;
• when a new page is proposed, check whether its main question already has a home;
• let broad pages summarise and link, rather than repeat, the detail on specific pages;
• review the map every quarter, since new content tends to drift into existing topics.

How do you find gaps in the map?

• Buyer questions with no page: check each question in the map against your site.
• Questions where AI answers cite competitors or third parties: run key questions in assistants and see which sources win.
• Sub-questions behind broad prompts: break common buyer prompts into their parts, such as integration, pricing or segment, and check each has an answer; see what query fan-out is.
• Entities without pillars: an audience or product you serve that has no main page.

Prioritise gaps by business value and how often the question comes up.

How do you check whether the map is working?

Pick two or three pillars and their clusters, and run the buyer questions they cover in the AI engines your customers use, before and after the work. Check three things:

• Broad questions such as "What is the best scheduling software for restaurants?": is the pillar cited, and are you named?
• Specific questions such as "Can a restaurant rota track tips?": is the matching cluster page cited?
• Accuracy: do answers state your prices, integrations and segments correctly?

If broad questions cite the pillar but specific questions cite competitors, the cluster pages need sharper answers. If specific pages are cited but the pillar never is, the pillar probably lists links without answering the broad question itself.

How do you keep the map useful over time?

Give it an owner, and make it part of content planning: every new page starts with a row in the map. Update it when products, audiences or locations change, and when AI answer checks reveal new questions. A map that is maintained stops content sprawl and keeps each question's answer in one reliable place.

What do people ask about topic clusters and AI?

Do topic clusters still matter if AI reads individual passages?

Yes. Clusters decide whether a clear page exists for each question and how pages connect. Passages are retrieved from pages, so good structure improves what there is to retrieve.

How many pages should a cluster have?

As many as there are distinct questions worth answering, often five to fifteen for a main product or audience. Do not create pages for minor variations of the same question.

Should pillar pages be very long?

Long enough to answer the broad question and summarise each subtopic well. There is no ideal length; clarity and accuracy matter more.

Can tools build topic clusters automatically?

They can help group keywords and questions, but review every cluster by hand. Automatic grouping can merge questions with different intent or split ones that belong together.