The first time I killed a planned service page mid-build, the client thought I had lost my nerve. We had agreed on a 22-page sitemap. By week 3, I cut it to 14. The 8 we removed were not bad ideas. They were sections pretending to be pages.
Every one of them failed the same test. I will explain it below, because it is the single rule that decides whether a site compounds or stalls.
The rule, said plainly
Sections earn promotion to pages. Never the reverse.
Most content plans work backwards. Someone makes a list of keywords, turns each keyword into a page, then scrambles to fill each page with sections. The sections exist to justify the page. That is the wrong direction. The page should exist because a section grew so specific, so load-bearing, and so different in search intent from its parent that splitting it out is the only honest option.
If you cannot point to a section that has outgrown its home, you do not have a new page. You have a duplicate.
What promotion actually means
A section graduates when 3 things are true at once. First, the search results page for its target query looks materially different from the parent page's results. Second, the buyer asking that query is at a different stage of their decision. Third, the answer you would write for it could not be lifted and pasted onto any sibling without changing the meaning.
If all 3 are true, you have a new page. If any one is missing, you have a stronger section inside the page you already have.
I run this check on every keyword that comes out of research. It is the gate between a thin sitemap that pretends to be thorough and a sitemap that actually compounds. The pages on page 1 are there because they were tested against real searchers and won, so opening the actual ranking URLs and classifying them is the only honest way to do this. Your instinct is what you would want, not what the searcher wants.
The Tenmar example, in detail
Tenmar installs glass railings across BC. When we started, the site had 5 pages. Three months later, it had 85, and lead volume was 4x higher.
People assume the win came from blasting out 80 net new pages. It did not. The win came from refusing to write 30 of the pages on the original list, because they failed the promotion test, and instead expanding 50 sections into real pages because they passed it.
Two examples from that build. "Glass railing installation" and "frameless glass railing installation" looked like 2 separate keywords on a research export. When I opened both result pages side by side, they shared 8 of the top 10 URLs. Same buyer, same intent, same format. We kept that as 1 page with a strong subsection. By contrast, "glass railing installation" and "glass railing cost" shared almost nothing in the top 10. The cost query pulled calculators and breakdown articles. The installation query pulled service pages. Different stage, different format, different page.
Multiply that decision across 80 keywords and you get either a thin site that cannibalises itself or a site that ranks across a category. The work was almost entirely about which decisions to make before anyone wrote a word.
The before-and-after, side by side
| Decision | Page-first thinking | Section-first thinking |
|---|---|---|
| Starting point | List of keywords | List of buyer questions and the SERPs that answer them |
| What gets built | 1 page per keyword | 1 page per distinct intent, with sections covering close cousins |
| Cannibalisation risk | High. Sibling pages fight for the same query | Low. Each page owns a separate result set |
| Filler risk | High. Sections exist to pad word count | Low. Sections exist because the buyer asked for them |
| What scales | Page count on the sitemap | Topical depth and authority |
| What Google sees | A site with many thin pages | A site with fewer dense pages, each one harder to copy |
The test I run before any section ships
Could you paste this section onto a different page targeting a different keyword and have no one notice? If yes, it is filler. Rewrite or delete.
This is the cleanest internal check we have, and it kills more drafts than any other rule. A section about "how long installation takes" written generically can sit on any service page in any vertical. A section about "how long installation takes when the railing wraps a curved deck on a sloped lot in North Vancouver" cannot. The first is decoration. The second is the page doing work.
Every site we have built that compounded did this. AetherHaus grew from 40 to 4,800 monthly visitors with 15 landing pages and around 32 articles, not 200. Each page carried its own weight. Maple Terroir grew organic traffic +1,500% with the same discipline. WashTech went from a generic template site to 5x revenue partly because we cut the planned page count and made the pages that survived deeper, more specific, and unswappable with each other.
The pattern across every client is the same. Fewer pages, each one impossible to substitute, beats more pages that all sort of look alike.
Why this matters more in 2026
Helpful, original content is now an explicit ranking signal, with Google's guidance directing site owners to focus on "people-first content" and to remove or rework unoriginal material rather than keep it live (Google Search Central, 2025). A site full of sections that could swap with each other is exactly what that guidance is trained to demote.
AI search makes this sharper. When ChatGPT or Perplexity pulls an answer, it pulls the most specific paragraph it can find on the open web. A swappable section is not the most specific paragraph. It is the same paragraph that already exists on 40 other domains. The model will skip it.
This is why GEO is not a separate workstream. The same rule that makes a page rank, depth that cannot be lifted, is the same rule that makes a page get cited. Across our roster, the pages that earn AI citations are the ones that pass the paste test. The ones that read like they could live anywhere never get pulled.
How to apply this on your own site this week
Open your sitemap. For every page you have, read the H2s out loud. Ask: could this H2 sit on the page next to it without changing a word? If yes, you have a candidate for merging.
Then look at your keyword research. For every keyword you planned to turn into a new page, run 2 searches in incognito. Compare the top 10 result pages between the new keyword and your existing nearest page. If the URLs overlap by more than 50%, you are planning a sibling that will cannibalise the page you already have. Keep the section, drop the page.
The work is not glamorous. It is mostly saying no to ideas that looked good in a spreadsheet. But it is the difference between a site that compounds and a site that just gets bigger.
One month of this discipline will not transform a site. 3 months shows real movement. A year is the difference between a site Google ignores and one it rewards. If you want a second set of eyes on which of your current pages should be merged, split, or rebuilt, that is the conversation worth having on a call.
Liam Lytton is the founder of The 66th, an SEO and GEO agency based in Vancouver, BC. He works founder-direct with operators across North America on building search presence that compounds.
