SEO Content Strategy: How to Build a Cluster Structure That Ranks

My GEO cluster has ten pages in it right now. Two of them are trying to answer the same question.

/generative-engine-optimisation/ and /generative-engine-optimization-geo-the-next-layer-of-search-visibility/ sit on my own site, different URLs, different word counts, near-identical intent. I wrote them four months apart and didn’t notice the overlap until I went looking for it while writing this. There’s a second pair doing the same thing on LLM SEO: /llm-seo-a-search-practitioners-honest-guide/ and /how-to-optimize-for-llm-search-the-complete-2025-guide-to-geo-ai-visibility/. Two pillar pages, one topic, both competing for the same reader.

I’m telling you this before I tell you anything else because it’s the honest starting point. Every SEO content strategy guide you’ll find gives you the same nine steps: understand your audience, do the keyword research, map the content, optimise, measure, repeat. That framework isn’t wrong. Audience research matters. Keyword research matters. You need all of it.

What it doesn’t tell you is that you can follow every one of those steps and still end up with what I’ve got. A cluster with real depth and a real problem sitting inside it.

The checklist isn’t wrong, it’s incomplete

Go and read the top ten results for “SEO content strategy” right now. Siteimprove, HubSpot, seoClarity, half a dozen others. They all converge on the same shape: research your audience, find your keywords, build a content calendar, optimise on-page, track your metrics. Some stretch it to nine steps, some compress it to six, but it’s the same skeleton wearing different clothes.

I don’t disagree with any of it. If you skip audience research you’re writing for nobody. If you skip keyword research you’re guessing at demand. Those fundamentals aren’t going anywhere.

The problem is what the checklist assumes without saying it out loud. It assumes the reader’s journey starts on Google. It assumes one persona per keyword. It assumes a cluster is “done” once every page has a home in the sitemap. None of those assumptions hold the way they did five years ago, and a strategy built only on the checklist misses the parts of the job that actually decide whether the cluster works.

The journey doesn’t start on Google anymore

Someone researching your topic today might never touch a search results page before they’ve already formed an opinion about you. They ask ChatGPT. They ask Claude or Perplexity. The tool reads across the web, synthesises an answer, and hands them something that feels complete. By the time they do search on Google, if they search at all, they’re often searching your brand name, not the generic topic.

That’s what search everywhere optimisation actually means in practice. Not “optimise for five platforms instead of one.” It means your content has to hold up wherever it gets encountered, including inside an AI answer that fans your topic out into six sub-questions you never wrote a heading for. A cluster built purely for Google’s crawler misses that a chatbot doesn’t crawl your site the way Googlebot does. It pulls a chunk, synthesises it into an answer, and moves on. If your cluster only makes sense as a whole document read top to bottom, it doesn’t survive that process well.

I’ve written about how query fan-out actually reshapes this in more depth elsewhere on the site, but the short version for cluster planning is this: your pillar page and its supporting content need to work as standalone, extractable answers, not just as chapters in a book only Google reads start to finish.

Topics, not keywords, and not one persona either

Here’s a gap that shows up early in most cluster plans. Someone maps “project management software” to a handful of keyword variants: best project management software, project management software for small business, project management software comparison. Reasonable enough on paper.

But a senior manager evaluating that software and a junior team member trying to figure out how to use it after it’s already been bought are two completely different readers. One wants a comparison built around ROI, team size, and integration risk. The other wants a walkthrough. Same rough topic, same product, entirely different job to be done.

A cluster built keyword-first tends to flatten that difference because it’s optimising for search volume, not for the actual range of people typing something close to that phrase. Building topic-first instead means starting with “who is asking this and why” before you start with “what phrase do they type.” The keyword research still happens. It just stops being the first decision and becomes the second one.

Does the cluster actually cover the funnel?

This is where most cluster diagrams look better than the clusters actually are. The pillar page exists. A handful of spokes exist. Someone’s drawn the hub-and-spoke picture in a deck and called it a strategy.

What that picture usually doesn’t show is whether the cluster covers the reader at every stage of actually deciding something. Top of funnel: what is this, why does it matter. Middle: how do the options compare, what’s the real trade-off. Bottom: how does this specific thing solve my specific problem, and can I trust the answer.

The gap I see most often sits at the bottom. Cluster plans are generous with explainer content and comparison content, then thin out right where the reader is closest to acting. If you’ve built or bought a tool, that’s the moment to explain what it actually does and how it solves the reader’s problem, without turning it into a sales page. Not “book a demo.” What does it do, step by step, and why does that specific mechanic matter for the problem the reader showed up with. That page is harder to write than a TOFU explainer because you can’t just summarise what’s already been said everywhere else. It has to come from having actually built or used the thing.

That’s also, not coincidentally, the content that’s hardest to replicate. Generic “what is X” content is well covered territory. An AI model has seen a hundred versions of it and can produce a hundred and first without much trouble. A page built on your own hands-on experience with a specific product, backed by what you actually saw happen when you used it, is a different kind of asset. It can be cited. It can’t be duplicated.

Intent cannibalisation, not keyword cannibalisation

This is where I’ll go back to my own cluster, because it’s the clearest way to explain the distinction that most cannibalisation advice misses.

The usual warning is: don’t have two pages targeting the same keyword. Fair, as far as it goes. But two pages can share a keyword and coexist fine if they’re answering different questions. A “what is generative engine optimization” page and a “generative engine optimization services” page can both rank for variations of “generative engine optimization” without stepping on each other, because one is explaining a concept and the other is selling an engagement. Different intent, different reader need, same rough phrase.

The real problem is intent cannibalisation: two pages trying to answer the exact same underlying question, for the exact same reader, regardless of what keywords each one happens to target. That’s harder to spot because it doesn’t show up as an obvious keyword clash. It shows up as two pages quietly splitting the same traffic and the same authority that one stronger page would have carried on its own.

I let my own cluster get away from me, here’s the fix

So back to my ten pages. /generative-engine-optimisation/ and /generative-engine-optimization-geo-the-next-layer-of-search-visibility/. Run the test: is a reader landing on either page trying to answer a genuinely different question? No. Both are explaining what GEO is and why it matters. Different titles, different word counts, same job. That’s intent cannibalisation, not a coexistence case.

Same story with /llm-seo-a-search-practitioners-honest-guide/ and /how-to-optimize-for-llm-search-the-complete-2025-guide-to-geo-ai-visibility/. Both are trying to be the definitive LLM SEO explainer. Neither one is doing anything the other isn’t already doing.

What I’d actually do here, and what I’m doing after this goes live: pick the stronger page in each pair, fold whatever’s genuinely additive from the weaker one into it, and 301 redirect the loser. Not because having two pages is inherently wrong. Because these four specifically aren’t serving different readers, they’re just splitting one audience across two URLs and asking Google to guess which one matters more. I’d rather make that decision myself than leave it to chance.

If you’re auditing your own cluster, the question to ask about every pair of pages that feels a little too close is simple: if I put these in front of the same reader on the same day, are they solving different problems, or just repeating each other in a different voice. If it’s the second one, you don’t have two pages. You have one page that isn’t sure what it wants to be yet.

Content isn’t just articles anymore

Google made a change in July 2026 that’s worth paying attention to here. Search Console added platform properties covering Instagram, TikTok, X, and YouTube, letting site owners see clicks, impressions, and top-performing posts for content on those platforms the same way you’d view a normal site property. That’s Google formally telling you it’s tracking multi-format visibility as part of the same measurement layer as your written pages, not as a separate world.

Attention spans are shorter than they were even two years ago. Trust in written content took a hit as unedited, ungrounded AI output flooded feeds, and readers got sharper at spotting it. Video, in both short and long form, has become a real part of how people search and how Google surfaces answers. So has structured content: tables people can scan, infographics that compress a comparison into one image, formats that don’t ask for the same sustained attention a 2,000-word article does.

None of that means every cluster spoke needs to become a video. It means format has to be chosen the same way audience is chosen, deliberately, based on what the specific reader at that specific stage actually wants and how they’re actually looking for it. A comparison page might work better as a table than a paragraph. A “how it works” explanation might land harder as a two-minute video than a wall of text. There’s no single format that wins across every persona and every funnel stage, and treating video or tables or infographics as a mandatory checkbox misses the point as badly as ignoring them entirely.

Building the cluster: a working structure

Put the pieces together and the shape looks like this. A pillar page anchors the topic and routes into spokes built around real funnel stages, not just related keywords. Each spoke is checked against the intent test before it gets built, so you’re not writing a fourth version of the same explainer under a new title. Internal links move the reader (and the crawler, and whatever’s fanning your content out into an AI answer) from broad to specific, TOFU to BOFU, in a path that actually makes sense to follow. Format gets chosen per page based on the reader at that stage, not applied uniformly across the whole cluster.

And once it’s built, it gets checked again. Not just for keyword rankings, but for the two questions that actually matter: does this cluster cover the full range of people searching this topic, from the senior manager to the junior researcher, and is every page in it still earning its place, or has something started quietly competing with something else.

Mine didn’t pass that check when I ran it for this piece. Worth running yours too.

FAQ

How many pages should a content cluster have?

There’s no fixed number. What matters is coverage, not count. A cluster with four pages that map cleanly to different stages of the reader’s journey beats a cluster with fifteen pages where half of them are answering the same question in different words.

What’s the difference between a pillar page and a cluster page?

A pillar page covers the topic broadly and links out to more specific supporting pages. Cluster pages (sometimes called spoke pages) go deep on one angle, one funnel stage, or one sub-question the pillar only touches on. The pillar is the map, the cluster pages are the destinations.

How do I know if I have keyword cannibalisation or intent cannibalisation?

Check the keyword first. If two pages rank for the same phrase but answer different questions for different readers, that’s coexistence, not a problem. If two pages answer the same underlying question for the same reader, regardless of which exact phrases each one targets, that’s intent cannibalisation, and it’s the more common issue in mature clusters.

Does a content cluster actually help with AI search visibility?

Structure helps because AI models fan a topic out into sub-questions and synthesise answers from whichever pages best answer each one. A cluster that’s mapped to real sub-questions, with each page written to stand on its own, gives an AI system more clean, extractable answers to pull from than one long page trying to cover everything at once.

Should every cluster include video content?

Only where it fits how that specific audience searches and consumes information. Google’s July 2026 Search Console update, which added performance reporting for Instagram, TikTok, X, and YouTube content, shows video and social formats are part of the same visibility picture as written pages now. That’s a reason to consider it, not a reason to force it into every spoke.

How often should a content cluster be reviewed?

At minimum, whenever you add a new page to it. That’s the moment to check whether the new page duplicates an existing one in intent, not just in keyword. Beyond that, a full review every few months catches drift before it turns into the kind of pair I found in my own GEO cluster.