Building topical authority with AI-assisted content clusters

Ahrefs' guide to this subject opens with an example that explains why small businesses should care: Bicycle Motor Works, a specialist e-bike retailer whose domain scores 15 out of 100 on Ahrefs' authority scale, outranks Amazon, which scores 96, for competitive e-bike keywords. Not because of some trick, but, as Ahrefs puts it, because "a site with a low DR can outrank a much stronger domain simply by covering a niche more completely".

That's topical authority in one sentence: depth of coverage as a substitute for raw domain strength. It's the most credible path for a small site to compete, and it happens to be the strategy that AI assistance accelerates most, because the bottleneck was always the sheer volume of competent writing required. This article covers what topical authority really is, stripped of the mysticism, the evidence it works, and a concrete plan for building one cluster.

What topical authority actually is (and isn't)

First, the honest version, because this term attracts folklore.

There is no dial inside Google labeled "topical authority" for ordinary websites. The only system Google has publicly described under that name applies to news publications, helping it decide which outlets are strong on a beat like health or local politics. SEO folklore extrapolated a site-wide topical score for everyone, and Google has never confirmed one.

What Google has confirmed adds up to something that behaves like topical authority, built from unglamorous parts:

• Each page you publish on a topic can rank for its own set of queries, so covering forty related questions simply enters you into forty contests instead of five.
• Internal links between related pages help both readers and Google. Google's own SEO starter guide says links "can provide more context on a topic, both for users and search engines, which may help demonstrate your knowledge on a topic".
• Google's quality guidance asks whether content shows "first-hand expertise and a depth of knowledge". A site that answers every question a customer has about drainage reads like it's written by drainage experts, because it demonstrably is.
• Readers who find complete answers stay, click deeper, come back, and link, and all of that feeds back into how the site performs.

So the working definition: topical authority is the compound effect of genuinely complete coverage, not a score you unlock. That reframe matters practically, because it tells you the goal is to be the best resource on a topic, not to hit an article quota.

The evidence that coverage compounds

The idea has more than logic behind it.

The original demonstration was HubSpot's. In 2015, two of their researchers ran an internal experiment restructuring blog content into clusters, and found, in HubSpot's words, that "the more internal links they added between related pages, the higher those pages climbed in SERPs". That experiment birthed the hub-and-spoke model most content teams now use.

More recently, Graphite, a growth agency, studied 332 URLs across 12 company sites, bucketing pages by a topical-authority score of their own design. Pages on sites with strong existing coverage of their topic earned their first impressions and clicks measurably sooner than pages on sites without it. It's a vendor study with a proprietary metric, so treat the shape of the finding rather than any precise number, but the shape matches what practitioners see: new pages on an established topic get traction faster than new pages on a topic the site has never touched.

And the case-study record, like the e-bike example above, keeps producing specialists outranking giants inside a niche. Ahrefs' own topic hub on SEO basics, one pillar page, draws about 2,900 organic visits a month and links from 649 domains, illustrating the other half: hubs themselves become assets.

None of this repeals the basics. Coverage doesn't rescue thin pages, and a new site still waits months for traction. It changes the slope, not the physics.

The architecture: hub and spokes

A content cluster has three parts.

The hub is one comprehensive page on the broad topic: "The complete guide to heat pumps in New Zealand". It targets the big, hard keyword, gives a genuinely useful overview, and links out to every spoke.

The spokes are focused pages, each answering one specific question: what a heat pump costs to run, what size a given house needs, heat pump vs gas heating, why a heat pump ices up, what installation involves. Each targets a narrower, more winnable query, and each is often exactly the kind of question your buyers type in the week before they spend money.

The links are the part everyone skips. Every spoke links up to the hub, the hub links to every spoke, and spokes link sideways where relevant, with anchor text that says what the destination is about ("what heat pump installation costs" rather than "click here"). This mesh is what turns thirty pages into one navigable resource, for readers and for crawlers.

The discipline the architecture enforces is completeness within a boundary. One fully-covered topic beats four half-covered ones, because the compounding happens inside the boundary.

A worked cluster map

Here's what the architecture looks like filled in, for a hypothetical electrician whose profitable work is EV charger installation.

The hub: "The complete guide to home EV charger installation in New Zealand", covering the decision end to end and linking down to everything below.

The spokes, ordered by buying intent. Closest to money: "How much does EV charger installation cost?" (the real price bands, what moves them), "Can my switchboard handle an EV charger?" (the question that decides half the quotes), and "7kW vs 22kW home chargers: what's actually worth it". Mid-intent: "Do I need council consent for an EV charger?", "Tesla Wall Connector vs third-party chargers", "How long does installation take?", and "Smart chargers and cheap overnight rates: the real savings math". Early-stage: "Can you charge an EV from a normal socket?", "How much does charging an EV add to your power bill?", and "Solar plus EV charging: what works and what's hype".

That's one hub and ten spokes, each an honest question with its own search demand, each answerable with genuine expertise, and each linking up to the hub and sideways to its neighbors ("Can my switchboard handle it?" links naturally to the cost page, which links to the 7kW-vs-22kW comparison). The intent ordering is the sequencing plan: the cost and switchboard posts ship first because they're the queries typed the week someone's buying.

Two design choices worth copying. The boundary is the profitable service, not the whole trade, an "everything electrical" cluster would dilute exactly the depth signal the strategy depends on. And every spoke passes the expertise test: an electrician has real, non-generic answers to all ten, which is what makes the finished cluster read like the specialist resource it actually is. Sketch your own version in this shape, ten to fifteen genuine questions around your most profitable service, and you have the sprint plan for the next quarter.

Where AI fits, and where it can't

Building a real cluster used to be a six-month writing project, which is why only content teams did it. AI assistance changes the arithmetic at three specific points.

Mapping. A model is excellent at exhaustively enumerating a topic: ask for every question a homeowner might have about heat pumps, grouped by buying stage, and you'll have a candidate spoke list in minutes. Filter it against real search data and your own judgment about which questions buyers ask, and the cluster map that took an agency a week takes you an afternoon.

Drafting. Spokes are the perfect AI-assisted format: focused, factual, one question each. With a proper brief and an editing pass per post, the cluster's writing cost drops by most of an order of magnitude, which is the difference between "someday" and "this quarter".

Consistency. The model keeps terminology, structure, and interlinking conventions uniform across thirty pages, which humans are bad at over months.

What AI cannot supply is the reason the cluster deserves to win: your experience, your local numbers, your opinionated answers to the questions where the standard advice is wrong. A cluster of generic spokes is just scaled thin content wearing a diagram, and Google's spam policies treat volume-without-value the same whatever shape it's arranged in. The test for every spoke is the same as for any post: does it contain something a competitor's model couldn't have generated? Your role in the workflow is editor, not bystander: the map and the drafts come fast, the expertise passes are where your hours go.

A 90-day cluster plan

Here's the executable version for a small business.

• Week 1: choose the boundary. One topic, close to the money, where you genuinely know more than the internet average. Narrow beats broad: "heat pumps" beats "home heating", and "heat pumps in coastal climates" might beat both if that's your edge.
• Week 1-2: map it. AI-generated exhaustive question list, filtered by search data and buyer relevance, down to one hub and eight to fifteen spokes. Rank spokes by buying intent so the most commercial ones ship first.
• Weeks 2-10: publish spokes, two per week. Each one briefed with your material, AI-drafted, edited, fact-checked, and interlinked as it ships. A steady cadence beats a bulk drop, and two good posts a week is sustainable alongside running a business.
• Week 6-ish: publish the hub. Writing the hub mid-way is easier because the spokes exist to summarize and link.
• Week 10-12: close the mesh. Sweep every page for missing internal links, in both directions. This hour of linking is the cheapest ranking work you'll do all quarter.
• Then: watch and extend. In Search Console, impressions and average position per spoke tell you which parts of the topic are catching. Extend the cluster where the data points, refresh spokes yearly, and only then start cluster two.

Expectations, honestly set: spokes on modest-competition queries typically show impressions within weeks and meaningful clicks over months, on the usual timeline, with the cluster's pages helping each other more visibly as coverage closes. This is a compounding asset, and compounding is slow before it's fast.

The one-line version

Topical authority isn't a score you win, it's what complete, genuinely expert coverage of one bounded topic earns you, and it's the one competitive lever where a small specialist beats a big generalist. AI collapses the cost of the mapping and drafting, you supply the expertise and the editing, and one properly built cluster will outperform a year of scattered posts.