I had a client last year with over 50 blog posts targeting all the usual “project management software” phrases – from “best project management software” through to “free project management software for startups”. Traffic was alright and the rankings weren’t bad. But when Google started incorporating more AI-driven results, this client’s site just vanished from the search results. Not a single mention. Meanwhile, a competitor who’d written maybe 15 pages on the same topic was always showing up.

The difference wasn’t in the word count or the site’s credibility. It was the way the content was set up. The competitor had built what I’ve come to call an AI-native content architecture – where pages are connected up by ideas and concepts, rather than just being variations on a keyword theme. That was a real eye-opener, and it’s what I want to explore here.

What Entity Clusters Actually Are

In SEO, an entity is just any concept, person, place or thing that Google can identify as a distinct “thing” rather than just a string of words. Google’s been building its knowledge graph around entities for ages now, but it’s more important than ever because AI systems don’t just retrieve pages – they retrieve knowledge.

An entity cluster is a group of related entities that, together, give you a really good grasp of a topic. If you’re writing about cybersecurity, your entity cluster might include things like threat actors, attack vectors, compliance frameworks, incident response and regulatory bodies. That’s not about keywords – that’s about building a real understanding of what cybersecurity actually is.

We used to be able to get away with just picking a keyword and optimising a page to rank for it. Nowadays, you’ve got to build content that’s properly connected up and shows you’re really across the topic – and AI systems can see that.

Why Keyword-First Architecture Falls Down

Traditional keyword-mapped architecture works because Google just went through your page, saw if it matched the search query and ranked it accordingly. That’s still in play to some extent, but there’s now an AI layer on top that works in a completely different way.

AI retrieval systems don’t care about keyword density – they care about whether you’ve covered a topic properly and how concepts relate to each other. They don’t care about whether you used the exact phrase someone typed. And they care about whether you’ve covered a topic thoroughly, not just if you’ve got a page that mentions it.

The thing that kept stumping me and my clients was that a site might have 50 pages targeting “project management” keywords, but if those pages never connect the dots between Agile methodology, sprint planning, team velocity and productivity frameworks – the AI just doesn’t see that site as an authority. It sees 50 disconnected pages that happen to mention project management.

Some signs you’re still stuck in the keyword era: you’ve got pages targeting similar keywords with almost identical content. There’s no clear hierarchy between your main topic pages and supporting content. Your internal links are based on anchor text optimisation rather than how ideas actually connect up. And there are obvious gaps between related subtopics your audience would naturally want to explore.

How to Build an AI-Native Content Architecture

This is where I’ve been spending most of my time with clients lately, and I’ll be honest, it’s more work upfront than the old way of doing things. But the results compound in a way that individual keyword pages never did.

Start by doing an entity audit to see what entities you cover and what gaps there are. Map every topic your brand covers to a named entity. Google’s Knowledge Graph and Wikipedia categories are useful for this, along with tools like InLinks or Semrush’s Topic Research. Then build an entity relationship map – literally drawing out on a whiteboard how your core entities connect to each other. If you cover personal finance, that might include things like tax brackets, investment vehicles, retirement accounts, compound interest, inflation and regulatory agencies. The map shows you which connections your content needs to make explicit.

After that, take a look at your content gaps – not just by search volume, but semantically. The question isn’t “what keywords am I missing?” It’s “what questions exist within this entity cluster that I haven’t answered?” Tools like AlsoAsked and Frase can help with this. So can just typing your topic into ChatGPT and seeing what follow-up questions it generates.

Finally, restructure your internal linking around entity relationships. I used to think about internal links in terms of anchor text, but now I link pages based on conceptual proximity. A page on data encryption links to pages on compliance standards, breach prevention, and cloud security because those ideas are related, not because they share keywords.Finally, the time has come to add structured data to make these relationships between entities in machine-readable. I used to view Schema as a nice to have, but now I see it as the foundation of an AI-native architecture. Without it, you’re at the mercy of the AI to magically infer the relationships you could just tell it about yourself.

Measuring this the right way

If you put a lot of work into an AI-native content architecture but still obsess over individual keyword rankings, you’re heading for madness. Your metrics need a change of pace too.

I’ve been tracking a few metrics with clients : Topic authority scores from tools like Semrush or MarketMuse give you an idea of how thoroughly you cover a subject area. How often the AI Overview pulls your content in is a great indicator of whether you’re actually getting cited in AI generated results. And then there’s entity coverage percentage – that’s how much of your topic cluster has got dedicated, authoritative content. And if you’re a branded entity, having a knowledge panel present is definitely worth keeping an eye on.

Keyword positions still count for something, but they don’t tell you as much about your visibility as they used to. A page can rank #3 for a query and still go completely unseen in the AI Overview. Coverage and citations in AI generated answers are now the real metrics to watch if you want to know where your traffic is headed.

Where to start from

If it all feels a bit overwhelming, keep it simple. Take a core topic area, spend 30 minutes mapping the entities involved, and pick out three content gaps that can be filled this month. That’s it, nothing more.

The client I mentioned at the start ended up restructuring their project management content to revolve around entity clusters instead of keyword variants. It took roughly 2 months of work. They went from zero mentions in the AI Overview to being cited 7 times within the first quarter after the restructure. And to show the power of consolidation, their total page count actually went down because they eliminated redundant content.

Fewer pages, better organisation, and more visibility – that’s the tradeoff I keep seeing, and it’s one I’d take every time.