Key takeaways
- Google AI Overviews pull from many passages across many pages, so a cluster of focused pages gives you more chances to be cited than one big article.
- Internal links tell crawlers and models how your pages relate. Link supporting pages to the pillar, the pillar to supporting pages, and related siblings to each other, using descriptive anchor text.
- Most of the cluster advice online is practitioner opinion, not tested evidence. Treat it as a sound structure to test, not a guaranteed ranking lever.
- Content format matters too. In July 2026, listicles, product pages and how-tos made up about half of Google AI Overview citations, so a cluster should mix formats.
- Measure what happens after you publish: which pages get cited, which get crawled, and which cluster gaps remain.
What is actually being claimed here
Let me be upfront about something. I could not find Google documentation or an independent, controlled study showing that internal linking directly increases your odds of appearing in AI Overviews. What exists is practitioner guidance from agencies like Seobility, Globerunner and Devenup, and it all points the same direction: cover a topic in depth, connect the pages sensibly, and make each page easy to extract an answer from.
That direction is reasonable. It matches how search systems work (crawl, understand relationships, retrieve passages), and it matches the old topic-cluster playbook that has been working for classic SEO for years. But "reasonable" is not "proven for AI Overviews", so I'll flag where something is opinion and where it is data.
The guide below treats clusters and internal links as the structure, and passage-level quality as the thing that actually gets you cited.
Why clusters fit how AI Overviews are built
AI Overviews and AI Mode use what Google calls query fan-out: a single question is split into several related sub-questions, each retrieved separately, and the results are stitched into one answer. Globerunner describes the same idea, noting that being the best page for one keyword does not guarantee you appear in a generative result.
Seobility gives a clean example. A query about AI in digital marketing might fan out into sub-topics like how AI is used in content creation, AI tools for SEO and keyword research, AI-driven ad targeting, ethical considerations, the effect on marketing jobs, and case studies. A well-built cluster mirrors that: a pillar page for the broad concept and a supporting page for each sub-question.
Here's the logic. Each sub-question is its own retrieval. If you only have one page covering it all at medium depth, you compete on every sub-question with a page that is not the best answer to any of them. If you have a focused page per sub-question, each can win its own retrieval.
A caveat on the data. Promptwatch Data's ChatGPT query fan-out report measured an average of about 2.15 searches per response in early December, easing to about 1.84 by early March, which is far below the "3 to 8+" figure often quoted. That is ChatGPT behavior, not Google AI Overviews, so don't treat it as proof of how Google fans out. What it does tell you is that fan-out sub-queries are real, short, and specific, and that the number per response varies. Its recommendation is useful either way: cover the sub-questions a fan-out splits into (comparisons, pricing, alternatives, how-tos), build a cluster of focused pages rather than one mega-page, and write headings that read like short search queries.
Step 1: Pick the topics you want to own
Start with the three to six themes your business should be known for. Seobility's skincare example is a good model: routines for different skin types, ingredients and their benefits, sustainable beauty practices.
A few rules I'd apply:
- Choose topics where you have real experience or data. AI systems cite sources that add something, and thin rewrites of the top results rarely do.
- Keep each cluster to one theme. A cluster that sprawls across five loosely related ideas is just a pile of posts.
- Check what already triggers AI Overviews in your space. Search your core queries and note which sub-questions the overview answers.
For keyword and question research, tools like AlsoAsked and AnswerThePublic show the follow-up questions people ask, which are a decent proxy for fan-out sub-queries.
Step 2: Map the cluster before you write
Build a simple map: one pillar, then supporting pages, each with a single intent.
| Page role | Job | Example (natural skincare) |
|---|---|---|
| Pillar | Broad overview, links out to every supporting page | The complete guide to natural skincare |
| Supporting: how-to | Answers one procedural question | How to build a sustainable beauty routine |
| Supporting: list | Ranks or groups options | 5 natural ingredients that hydrate dry skin |
| Supporting: comparison | Resolves an "X vs Y" question | Organic vs natural skincare products |
| Supporting: product or category | Commercial intent | Natural moisturizers for dry skin |
Notice the format mix. Promptwatch Data's July 2026 report on content types cited by Google AI Overviews shows where citations went that month:
| Content type | Share of AI Overview citations (July 2026) |
|---|---|
| Listicle | 18.0% |
| Product page | 16.3% |
| How-to | 15.1% |
| News article | 13.5% |
| Video | 5.9% |
| Social post | 5.1% |
| Landing page | 4.7% |
| Comparison | 3.6% |
Two details are worth noticing. Product pages were the most cited format every day from July 28 and ended the month at 17.9%, against 16.2% for listicles. And listicles averaged about 26% in Q1 2026, so their share has dropped. Only citations with a classified content type are counted. The practical takeaway from the report is to give product pages the same attention as editorial content and keep list-style pages without relying on them alone. A cluster that is only blog posts leaves a big slice of citation types uncovered.
Step 3: Write the pillar page
The pillar should be a real page, not a table of contents with ambitions. Aim for:
- A clear definition or direct answer to the main question near the top.
- A short section on each sub-topic that gives a useful answer and then links to the full supporting page.
- A logical heading structure, with headings phrased close to how people search.
- Original material wherever you can: your own data, examples, screenshots, or first-hand experience.
The pillar earns its place by orienting the reader and the crawler. It should not try to out-detail the supporting pages, or you will end up with two pages competing for the same query.
Step 4: Build the supporting pages around single intents
Promptwatch's guidance from the fan-out data is blunt: a page aimed at one intent beats a page that half-covers two. I agree. Each supporting page should:
- Answer its core question in the first few lines.
- Expand with detail, examples and caveats.
- Use H2s and H3s that read like short queries.
- Put the entity and category first in titles and headings ("Best CRM for small agencies 2026" rather than "Our thoughts on picking a CRM").
- Link back to the pillar and to two or three closely related siblings.
For drafting and optimization help, content tools such as Frase, Clearscope and MarketMuse can show you which subtopics competing pages cover that yours doesn't.
Step 5: Link the cluster properly
This is the part most people get half right. The consistent advice across Seobility, Globerunner and Launchmind is:
- Every supporting page links to the pillar.
- The pillar links to every supporting page.
- Related supporting pages link to each other where it helps the reader.
- Anchor text is descriptive. "Read more" tells nobody anything.
Globerunner adds a point I like: a coherent linking structure makes it easier for search engines to crawl and for AI models to stitch related passages together during fan-out. Whether Google's systems use links that way for AI Overviews specifically is not documented, so call that a plausible mechanism rather than a fact.
Anchor text examples
| Weak | Better |
|---|---|
| Click here | how to build a sustainable beauty routine |
| Read more | our guide to hydrating ingredients |
| This article | organic vs natural skincare products |
A few linking habits that help
- Link in the body, in context, not only in a footer or sidebar. A link inside a relevant sentence carries more meaning for a reader and for a parser.
- Avoid linking the same destination five times in one page.
- Don't point multiple pages at the same exact anchor for different destinations. That muddies what each page is about.
- Cross-link between clusters only when the connection is real.
Step 6: Fix the technical basics so pages can be used at all
A beautiful cluster on a site that AI crawlers or Googlebot can't read is wasted. Check:
- Orphan pages. Any cluster page with no internal links pointing at it is invisible to the structure you just built.
- Broken links and redirect chains inside the cluster.
- Pages missing from the XML sitemap.
- Rendering. If key content only appears after heavy JavaScript, crawlers may miss it.
Crawlers like Screaming Frog, Sitebulb and Lumar make orphan pages and link depth easy to spot. Google Search Console tells you which cluster pages are indexed and which queries they earn impressions for.
If you want a side-by-side of what each crawler does, the broader software directory at surferstack.com covers them with reviews and alternatives.
Step 7: Keep the cluster current
Freshness matters more for some formats than others. Promptwatch Data's content half-life report tracked every page ChatGPT cited on July 6, 2026 for 90 days: landing pages and docs stayed cited longest, while news lost half its pages within 36 days. That is ChatGPT, not Google, but the pattern is a sensible prompt to schedule refreshes. Time-sensitive supporting pages deserve a review cadence. Evergreen definitional pages need it less.
When you update a page, update the links too. New sibling pages should be woven into old ones, otherwise your newest content sits unlinked at the edge of the cluster.
Step 8: Measure whether it worked
Here is where I'd be disciplined, because the evidence for this whole approach is thin and your own data is the only real test.
Track:
- Impressions and clicks per cluster page in Search Console.
- Whether your pages appear as cited sources in the AI Overview for your target queries.
- Which sub-questions in the cluster still have no strong page.
- Whether AI crawlers are actually fetching the pages.
For the AI side, Promptwatch is one option. It tracks how your brand and pages show up across Google AI Overviews, AI Mode, ChatGPT, Perplexity and others, shows which pages get cited, and logs AI crawler visits so you can see whether a new cluster page was read at all. Its content gap analysis and Content Agent are aimed at exactly the "which sub-questions am I missing" problem this guide describes.
Other trackers to look at include Semrush, Ahrefs Brand Radar and Otterly.AI. They vary a lot in how deep they go, and many only tell you whether you were mentioned. For a broader list, the GEO software directory at bestgeosoftware.com is a decent starting point.
Otterly.AI
| Tool type | Good for | Examples |
|---|---|---|
| Crawler | Orphan pages, link depth, broken links | Screaming Frog, Sitebulb, Lumar |
| Content optimizer | Subtopic coverage per page | Frase, Clearscope, MarketMuse |
| Question research | Fan-out style sub-questions | AlsoAsked, AnswerThePublic |
| AI visibility | Citations, crawler logs, content gaps | Promptwatch, Semrush, Ahrefs Brand Radar |
| Search data | Indexing and query performance | Google Search Console |
A worked example
Say you run a project management software company and want to be cited for "project management software." Devenup's version of the semantic approach is useful here: the related concepts an AI associates with the topic are scheduling, resource allocation, team collaboration and Gantt charts.
A workable cluster:
- Pillar: Project management software: what it does and how to choose
- How-to: How to set up resource allocation in a small team
- Explainer: What a Gantt chart is and when to use one
- Comparison: Kanban vs Gantt for software teams
- Listicle: Best project management software for small agencies 2026
- Product page: your own feature page, linked from the relevant sections
- Alternatives page: alternatives to the biggest competitor
The pillar links to each. Each links back with anchor text like "how to choose project management software". The Gantt explainer links to the Kanban comparison. The listicle links to your product page. That is the whole structure, and it takes a few weeks of work, not a new platform.
Common mistakes
- Building the pillar and never finishing the supporting pages. A hub with five dead ends is worse than a single solid page.
- Making supporting pages near-duplicates, which creates cannibalization where search engines can't tell which to pick.
- Treating internal links as a magic switch. They organize good content. They don't rescue thin content.
- Ignoring non-blog formats. Product, comparison and landing pages get cited too.
- Measuring only rankings. An AI Overview can cite you without you ranking first, and rank you first without citing you.
Where this leaves you
Clusters and internal links won't force a citation. What they do is give you a coherent set of focused pages, each of which can win a retrieval for one sub-question, joined by links that make the relationships obvious. Build one cluster around a topic you genuinely know, link it carefully, and compare the cited-source data before and after. If it works, expand. If it doesn't, you will at least have cleaner site architecture.