Hub and spoke, and why the arrows matter more than the titles
A pillar page earns its position by receiving links from pages that are narrower than it and about the same subject. Publish 20 articles around a topic with no internal linking discipline and you get 20 orphans competing with each other. Point them at a hub and the hub becomes the page Google understands as the authoritative one on the broad term.
So each supporting article in the plan gets a link direction. "Up to pillar" is the default: the article's primary internal link points at the hub. "Lateral" means it points sideways at a sibling in the same cluster instead, which is how a cluster stops being a fan of unconnected spokes.
4 rules are enforced on every cluster that comes back:
- At least 2 articles per cluster point up, so the hub actually accumulates links.
- At least one points laterally, so the cluster interlinks internally.
- No 2-node loops. If A links laterally to B, B does not link back to A.
- No target keyword is repeated across articles, because 2 pages chasing one phrase is the cannibalization problem this structure exists to prevent.
What the 4 inputs change in the generated clusters
Pillar topic. The broad subject the hub will own. Pick something a single page could plausibly rank for. "Email deliverability" works. "Marketing" does not, and the plan you get back for it will be generic because the input was.
Audience. Shifts the angle of every title. "B2B SaaS growth marketers" produces different supporting articles than "small business owners" for the identical pillar, because the questions differ.
Business type. SaaS, ecommerce, agency, marketplace, media, local, or other B2B. This mostly moves the intent mix. Ecommerce pulls the plan toward transactional and commercial pages; media pulls it almost entirely informational.
Number of clusters. 3 to 8. Each cluster gets 4 to 6 supporting articles, so the plan sizes between 12 and 48 articles plus the pillar. Start at 5 unless you know you have the production capacity for more.
What a topic cluster generator returns
One pillar definition, then N cluster cards.
The pillar card carries the page title, the single broadest keyword it can realistically own, and 5 to 8 concrete things the pillar has to cover to deserve the hub position. Those are subjects, not instructions. "SPF, DKIM and DMARC alignment" is a mustCover item. "Provide value to the reader" is not, and the plan is built to avoid producing that kind of line.
Each cluster card gives a theme name, one line on why the theme belongs under the pillar, and its articles. Run it on "email deliverability" for B2B SaaS growth marketers at 5 clusters and one card comes back like this:
Cluster: Authentication and domain setup
Why: Nothing else in the plan works until the sending
domain authenticates cleanly.
- How to Set Up SPF, DKIM and DMARC for Cold Email
keyword: spf dkim dmarc setup
intent: informational
links up to: Email Deliverability: The Complete Guide
- DMARC Policy: p=none, quarantine or reject?
keyword: dmarc policy quarantine vs reject
intent: informational
links laterally to: How to Set Up SPF, DKIM and DMARC for Cold Email
The chip on each card shows the split, so a cluster reading "4 up, 1 lateral" tells you the shape at a glance without opening it.
It reports no search volume, and that is deliberate
There is no keyword API behind this tool. No volume, no keyword difficulty, no CPC, no traffic estimate, anywhere in the output.
That is a design decision, not a missing feature. A language model asked for a monthly search volume will produce a number, and that number will be invented. It will look plausible, it will carry 2 significant figures, and it will have no relationship to anything measured. Plenty of cluster planners print those numbers anyway. Treating fabricated data as real is worse than having no data, because you will make a prioritization call on it.
What you get instead is qualitative: intent tags per article, and keywords chosen to be narrower than the pillar and distinct from each other. When you are ready to sequence the plan, take the keyword column into a real keyword tool and pull the volumes from a source that measured them.
The run also asks you to pass a human check first, since it calls a language model.
What you still have to do after the clusters are built
The plan is a structure, not a schedule. It does not tell you what to publish first, how long each piece should be, or which articles to skip.
Sequencing is yours. The pattern that works: publish the pillar first even if it is thin, then the 2 or 3 supporting articles per cluster with the clearest commercial intent, then fill in. A pillar with no spokes yet ranks for nothing, but it gives every article you publish afterward somewhere to point.
Check the plan for overlap before you brief anything. Across 5 clusters and 25 articles, 2 titles will occasionally be closer in scope than their distinct keywords suggest. Merge them. 2 thin pages on one subject is the exact problem clustering is meant to fix.