How to build a topical map, worked on 525 real phrases

Before writing a single guide, I needed an answer to three questions: which topics deserve a page, which pages belong together, and which ones come first. A topical map answers all three. The example below starts from a file of 525 search phrases with Google Ads volumes for the US, dated 2026-10-22.

Four steps from seed to map

Seed topics come first: the subjects the site can cover with its own tools, templates and data. Expansion turns each seed into the phrases people actually search, each with a monthly volume. Clustering groups those phrases by the need behind them. The map then places every cluster under a pillar hub and gives each search intent exactly one URL.

In this file, expansion ended at 525 phrases, clustering at 12 clusters, and the map at 9 pillars. The rest of this page works from that output.

Clean the list before you count it

Phrase lists from different sources disagree on case and repeat each other. Counting before cleaning inflates every cluster. The normalizer above splits on commas, semicolons and new lines, trims, lowercases and removes repeats, in that order. The same rule fits in a few lines of JavaScript. Run on Node 26.8.2 on 2026-10-28 with the six phrases from the figure, it prints the four distinct ones:

const pasted = 'n8n pricing, N8N Pricing; n8n cloud pricing, what is apify; apify pricing, Apify pricing'

const phrases = [
  ...new Set(
    pasted
      .split(/[,;\n]/)
      .map((p) => p.trim().toLowerCase())
      .filter(Boolean)
  ),
]
console.log(phrases.length, phrases)

4 [
  'n8n pricing',
  'n8n cloud pricing',
  'what is apify',
  'apify pricing'
]

Applied to the whole keyword file, the same rule leaves 525 distinct phrases and no duplicates.

How big each cluster is

FIG. 03 · data

Phrases per cluster in the keyword file

SEO tools and data96

n8n86

Apify83

Amazon sellers48

YouTube research47

AI for developers44

AI search33

Ad intelligence30

Local leads25

Data studies14

Avatars and IndieWeb11

Finance news data8

Data table

Item	phrases
SEO tools and data	96
n8n	86
Apify	83
Amazon sellers	48
YouTube research	47
AI for developers	44
AI search	33
Ad intelligence	30
Local leads	25
Data studies	14
Avatars and IndieWeb	11
Finance news data	8

The three largest clusters, SEO tools and data, n8n and Apify, hold 96, 86 and 83 phrases. Those are also the three pillars the plan starts with.

Phrase count is not demand, though. The table adds up the monthly volumes in each cluster and shows its largest single phrase:

Cluster	Phrases	Summed volume	Largest phrase	Its volume
SEO tools and data	96	129,200	serp api	12,100
n8n	86	190,490	n8n	165,000
Apify	83	46,770	apify	40,500
Amazon sellers	48	81,270	amazon price tracker	27,100
YouTube research	47	180,070	youtube transcript	110,000
AI for developers	44	737,270	claude code	550,000
AI search	33	538,700	google ai mode	368,000
Ad intelligence	30	177,940	meta ad library	90,500
Local leads	25	572,450	google business profile	550,000
Data studies	14	1,600	youtube statistics	1,300
Avatars and IndieWeb	11	1,530	indieweb	1,000
Finance news data	8	1,560	earnings call transcript	480

One phrase carries most clusters. In Local leads, the largest phrase is 550,000 of 572,450. In AI for developers it is 550,000 of 737,270.

A summed volume like that says a head term exists, not that 44 pages would each find readers. I read the summed column as a warning, not a ranking.

The zero-volume phrases stay in the map: 57 of the 525 have a volume of 0. The plan keeps them because a page with a working tool or a download can earn its place without measured volume.

From clusters to the map

The 12 clusters became 9 pillars, and not every cluster became a pillar of its own. A small cluster folds into a broader section, and one cluster has no page at all in the first version of the map. Size alone did not decide it: a cluster earns a pillar when several of its pages can stand on a tool, a template or data of my own.

The map also sets an order. Pages with that kind of first-hand work behind them go first; clusters that only have volume wait until there is a measured signal of demand. A pillar gets its hub page only once it has 3 published pages, so the map never shows empty sections.

Do it for your own site

Start the seed list from topics where you have your own tool, data or template, not from the topics that have volume; a page you cannot back with first-hand work will not hold its place. Expand those seeds with any volume source you trust, but keep one market and one date in the file, so the numbers compare.

Group the phrases by the need behind them and settle every unclear case with one rule: a tool does the task, a template deploys it, a guide explains it. Give each intent one URL, and add a hub only when a section has 3 pages to show.

Then order the pages by what you can test and show soon, not by summed volume. A cluster with a big head term and nothing of yours behind it can wait; the map is a plan you revise, not a ranking.

What I tested and what is untested

The counts are deterministic: the same file gives the same chart. The clusters were assigned by the need behind each phrase, not by comparing search results.

I have not checked them against SERP overlap, so two clusters that share result pages would show up here as separate. The volumes are averages for the US and say nothing about other markets. Every number here comes from that one keyword file; there are no outside sources to cite.

Drafted with an agent from my keyword file and notes; the cluster counts were computed by the agent on 2026-10-23, and the code example was run by the agent on 2026-10-28.

Reviewed by David S on 2026-10-28.