Topic clusters in 2026: still working, but built for fan-out now

Topic clusters still work in 2026, and AI search has made them more useful, not less. A cluster is a pillar page that covers a topic broadly plus a set of linked pages that each answer one specific question in depth. Google says its AI Overviews and AI Mode answer by "issuing multiple related searches across subtopics", and an Ahrefs study of 4 million AI Overview citations found that most cited pages did not rank in the top 10 for the original query. A well-built cluster gives that fan-out a strong page to cite for each sub-question, instead of one pillar competing for all of them.

What changed and what did not

The core idea has not changed: cover a topic thoroughly, organize the coverage so people and crawlers can follow it, and link the pieces together. What changed is what each piece is for. In the classic model, cluster pages existed largely to pass relevance and links up to a pillar that targeted the head term. In AI search, each cluster page is a candidate source in its own right. It needs to be the best available answer to its question, readable and quotable on its own, because the system citing it may never look at the pillar.

Planning a cluster for fan-out

1. Choose a topic you can genuinely cover better than anyone ranking now, because you have the experience, the data, or the customers. Coverage you cannot make excellent is coverage you should skip.
2. Collect the real questions. Sales calls, support tickets, reviews, the People Also Ask results, and the follow-up suggestions in AI Mode all reveal sub-questions; see customer question mining.
3. Group the questions by the answer they need, not by keyword wording. Two phrasings with the same answer belong on one page; one phrasing that needs two different answers belongs on two.
4. Write the pillar's opening passage as a clear definition of the topic, naming the entity and stating the core answer in the first sentence.
5. Give each cluster page one job. Open with a direct answer under a heading that states the question, then add the depth, sources, and examples.
6. Publish the pillar and the first cluster pages together, so the structure exists from day one, and add pages as new questions appear.

A worked example: an accounting firm

Suppose a CPA firm that advises small businesses wants to own the topic of S corporation elections. A keyword tool would suggest dozens of variations of "S corp" and "S corp tax." Planning for fan-out starts from the questions an owner actually asks before deciding, and gives each distinct answer its own page:

• Pillar: what an S corporation election is, who it suits, and the main trade-offs, with links to every page below.
• When does an S corp election start saving money, with a worked comparison at a few profit levels.
• What counts as reasonable compensation for an S corp owner, and how the IRS looks at it.
• How to file the election, the deadline, and what late-election relief requires.
• How state taxes change the math, with the firm's own state covered in detail.
• When an S corp is the wrong choice, written honestly, including the cases where the firm would advise against it.

Each page answers one question a person would type or ask an assistant, and each is worth reading on its own. When an AI Overview fans "should my LLC be an S corp" out into sub-questions about savings, salary, and deadlines, the firm has a strong candidate for every one. Our guide for accounting firms covers the vertical specifics, including the structured data and directory listings that matter for CPAs.

Linking it together

Google's link guidance is the right standard here. "Every page you care about should have a link from at least one other page on your site," and "good anchor text is descriptive, reasonably concise, and relevant to the page that it's on and to the page it links to." For a cluster that means three kinds of links:

• Pillar to every cluster page, with anchor text that names the specific question each page answers.
• Every cluster page back to the pillar, and to the two or three siblings a reader would logically need next.
• Cluster pages to the relevant product, service, or pricing page, where the topic connects to what you sell.

Avoid generic anchors such as "learn more" and "click here." They waste the one place you get to tell a crawler, and a reader, what the next page is about. More on this in the glossary entry on internal linking.

The scaled-content trap

The fastest way to ruin a cluster is to build it out of near-identical pages. In March 2024 Google introduced a spam policy against "scaled content abuse": producing many pages primarily to manipulate rankings, whether by people, automation, or both. A cluster of 40 pages that each swap one keyword into the same template fits that description. Google's helpful content guidance asks a simpler question that catches the same problem: is the content made "primarily to help people" or "to attract search engine visits"? If a cluster page would not be worth reading on its own, it should be merged into another page or not written.

Measuring whether the cluster works

Measure the topic, not the pillar. Kevin Indig's "Topic Share" approach, published in May 2025, measures how much of a topic's organic traffic a site captures across all of its keywords relative to competitors, which reflects rankings, search volume, and SERP features together. For AI search, add a second measure: across the questions in the cluster, how often each engine cites one of your pages. A cluster that wins a growing share of both is building topical authority in the sense that matters.

Expect the cluster to take months, not weeks. Pages need to be crawled, indexed, linked to, and tested against competing answers before either measure settles. Judge the cluster on the trend across the whole topic, and fix individual pages that stay uncited after the rest have moved.

How SEO for AI Agents measures this

Our audit evaluates the pages a cluster is made of. The passage extractability check shows, page by page, whether the opening passage under each heading answers cleanly on its own, with the word count and heading behind every paragraph. The title quality check flags titles that do not match the question a page answers, which is the most common sign of a page built for a keyword rather than a question.

For the AI outcome, we put the cluster's questions to each engine several times and record which of your pages are cited, how often, and which competing sources were cited instead, with the verbatim answers as receipts through the source of citation check. For how the fan-out itself works, see how Google AI Overviews pick citations.