A 1,000-Node Content Tree for $0.35: Diverge-Then-Converge Topic Architecture

Content planning is the most underrated step in SEO, and the one that punishes mistakes most. Without a topic architecture, two pages can fight over the same keyword while entire areas of coverage stay empty. We used to build these architectures by hand, and it was slow enough that we could not afford to do it for every site — a multi-thousand-node site took weeks of meetings, and any change to the catalog meant starting over. Until we handed the work to a diverge-then-converge process, this bottleneck never moved.

Diverge first, then converge

The diverge phase is deliberately wasteful: AI generates candidate topics in bulk across multiple dimensions — category, attribute, scenario, device, region and more. Quantity matters here, because the goal of divergence is to map the full possibility space, duplicates and noise included; a dimension you skip leaves a whole branch missing from the tree forever. In one real pass this produced 1,221 candidate topics. The converge phase then applies rules and clustering to deduplicate and organize them into a hierarchical tree: rules remove the obvious duplicates, clustering groups semantically related topics under the same branch. Our record converged to 1,324 nodes — parent pages, hub topics and leaf articles arranged with clear intent boundaries.

The whole tree for about $0.35

The entire 1,000+ node content tree cost about $0.35 to generate. When heavyweight planning collapses to a negligible cost, the mindset changes: you stop hoarding the architecture and start re-running it. A topic tree stops being an annual project and becomes something you refresh whenever the business, the catalog or the market moves — run the diverge-converge pass again, diff the old and new trees, and the missing pieces are obvious.

What the tree actually unlocks

A tree is not decoration. With it, three things become checkable before a single word is written: sibling differentiation — every brief knows exactly how it diverges from the pages next to it, so no two articles cannibalize one keyword; internal linking — link equity flows along the tree's relationships instead of wherever a writer happens to link; and pillar coverage — the big topics with no hub page are visible at a glance. Because every node carries an intent, “who is this page for and where does it end” stops being a meeting-room argument and becomes a fact you can read off the tree.

The 5-step diverge-then-converge checklist

• Define the dimensions — category, attribute, scenario, device, region; dimensions set the ceiling on divergence quality.
• Diverge with AI in bulk — generate thousands of candidates in one pass; tolerate duplicates and noise.
• Converge with rules plus clustering — clean with rules first, then cluster into a hierarchy.
• Organize into a tree with intents — a node is a title plus an intent, not a title alone.
• Plan briefs and internal links from the tree — differentiation and linking are the actual output.

Building and maintaining topic trees this way is what our SEO content orchestration module does before a single article is written.