Search-led information architecture: how we sized our own site by demand

Most information architectures are organised by the wrong thing. They mirror the company: a Services section, a Blog, a Resources tab, an About page. That is the org chart, not the way a person or an AI arrives at the site. People arrive on a topic, with an intent, phrased as a query. If the structure does not match the way demand actually shows up, every page fights the navigation instead of being lifted by it. A search-led IA flips the default: the topic is the spine, and content type becomes a secondary lens.

Size the structure by real demand, not opinion

The method is demand-weighted topical clustering. You gather the full keyword universe for a domain, group it into candidate topics (an L1 category with a handful of L2 sub-topics under each), and attach the real monthly search volume to every node. Now the structure has weight behind it: the order of the navigation, the internal-link priority and the depth of coverage all follow demand rather than a hunch. The single most useful test of a proposed IA is simple: what share of the total real demand does this structure actually house? An architecture that leaves half the demand homeless is not an architecture, it is a preference.

We first proved this out at scale for a B2B trade publisher. We pulled roughly thirteen thousand seven hundred keywords for the sector, clustered them, and then scored several candidate information architectures against one number each: the percentage of top-of-funnel demand each one could house. Different internal stakeholders had proposed different structures, some editorial, some commercial, and the demand test settled it objectively. The version that won housed effectively all of the priority demand. The one built purely around editorial pillars housed almost none. The output was a two-layer navigation: a demand-weighted spine as the primary structure, with the editorial pillars kept as a secondary layer, so the site served both the reader arriving on a query and the newsroom that thinks in beats.

Then we ran it on ourselves

Selling this is easy. Living by it is the harder part, so we pointed the same method at laurelinlabs.com. The spine we chose is our own Four Pillars framework, which means the site structure is now a live demonstration of the thing we consult on. There is one hero hub, AI Search Visibility, then the four pillars beneath it, then three cross-cutting tracks. Every service, framework, lab build and article maps to exactly one node. We pulled live UK search volume for the whole space and let it order the pillars.

The interesting part was the data cleaning, because raw search volume lies if you trust it blindly. Two passes mattered. First, false intent: the term that looked like our biggest topic by a mile was contaminated by a query that returned over a million searches but meant something completely unrelated, people looking for somewhere to eat, not a marketing concept. Left in, it would have made the emptiest pillar look like the largest. Out it went. Second, navigational demand: the two biggest legitimate numbers were people searching for Google's own products by name, not for a service. Real, but not our commercial battleground, so we set them aside rather than let them dominate the spine. Once the noise was cleaned, the honest ranking put our positioning bet, AI search visibility, at the top on merit rather than on hope.

Why topic-first future-proofs the content

Three things make this worth the effort. First, query fan-out means AI search decomposes one question into many parallel sub-queries, so the unit that gets cited is the cluster, not the single post. An IA built around topic hubs mirrors the way retrieval actually works. Second, the one-home rule: because every node is a topic, any future article, tool, service or project has exactly one obvious place to live, so the structure never needs a rethink as the site grows. Third, a topic-first tree makes coverage visible. Ours immediately exposed one pillar with no content at all, Trust Signals, which is precisely the gap we then filled with our own Non-Commodity self-audit. A structure that tells you where you are thin is doing its job.

This article is itself a small piece of that discipline. It sits in the Topical Content pillar, it earns its place by carrying method and numbers rather than restating a definition, and it links out to the frameworks it depends on. That is the standard, and the way to prove a standard is to hold yourself to it in public.