Topic clusters when the engine reads instead of ranking
The cluster model still works, with the weight inverted. The pillar page used to be the asset and the cluster pages fed it. Under retrieval the cluster pages are the assets, because a machine lifts the narrow page that settles one question, not the broad one that surveys twenty.
The pillar-and-cluster model was built for a world where authority concentrated on one URL and everything else pointed at it. That world assumed a ranking contest for a broad term. Retrieval does not hold a contest for a broad term; it looks for the passage that settles a specific one.
What inverts
| Ranking era | Retrieval era | |
|---|---|---|
| The asset | The pillar page | Each cluster page |
| Cluster pages exist to | Support the pillar | Be quoted themselves |
| Ideal cluster page | Eight hundred words feeding upward | Whatever it takes to settle one question |
| The pillar's job | Rank for the head term | Orient a reader and route a machine |
| Linking direction | Mostly upward | Both ways, and sideways between siblings |
The practical consequence is that the thin supporting page — written to exist rather than to be read — stops being harmless and becomes the problem. It was never going to be quoted, and it now dilutes a subject where you want a machine to find one clear source.
Twenty pages that each settle something beats one page that surveys twenty things.
How to build one now
Not "marketing". Something where you could plausibly be among the best sources in your language and market.
Everything a person asks on the way from not knowing about this to deciding. Sales and support will produce a better list than any tool.
What does this mean, how does it work, what does it cost, what goes wrong, how do I choose, how do I start. The chain is the internal linking plan.
Title as the question. If two questions want the same page, they were one question.
A short account of the subject with an honest map of the chain. It is a routing page and a page a reader can send to a colleague, not a twelve-thousand-word attempt to contain everything.
Each page links to the next question in the chain and to its near neighbours. That is how a reader travels and how a machine works out that these pages belong together.
When the pillar still earns its length
For subjects where people genuinely want one document — a specification, a standard, a full method — a long pillar is right, and it should then be structured so that any section can be lifted on its own. The mistake is the long pillar written for length rather than because somebody needed the whole thing in one place.
Questions this settles
No, but length no longer substitutes for specificity. A long guide whose sections each settle something is fine; a long guide that surveys is outperformed by the narrow pages it surveys.
As many as there are distinct questions, which is usually between eight and thirty. Inventing questions to hit a number produces the thin pages the model no longer tolerates.
It helps a reader and makes the relationship legible in the URL, but linking matters more than path structure.