When does a concept need its own page, and when does a section on the page do the same job?

When a topic deserves more depth, the instinct is to give it its own page. We tested whether the page boundary is what actually helps AI retrieval, or whether a well-developed section on the existing page does the same job. Across 20 real pages and 160 real questions, a dedicated page and an on-page section were equally findable. The URL boundary added nothing.

A common pattern in search work is to take several thin pages on a topic and consolidate them into one authoritative page with good structure, and watch it do better. The opposite advice is also common: split a buried idea onto its own page so it can be found. Both can be right. The open question is where to draw the line, and whether the answer is different for AI retrieval than for classic search.

Our earlier Findability Study showed that building a recommended page lifted answer-findability from 4% to 84%. But it compared a page existing against not existing, so it could not separate two things: having a dedicated URL, and having the content developed at all. This study separates them.

The setup

For each of 20 real pages, ContentGrapher identified one concept it said should move to its own page. We then built three versions and asked the same questions of each:

1. AThe original source. The real page as published, where the concept is thinly covered.
2. BA dedicated page. A focused standalone page about the concept.
3. CA section on the page. B’s exact content placed as a section inside the original page.

B and C carry the same concept content. They differ only in whether it lives on its own URL (B) or as a section of the consolidated page (C). That is the comparison that isolates the page boundary.

What we found

For narrow questions about the moved concept, developing it transformed retrieval: from 15% on the original page to 70% in both other versions. For broad questions about the whole topic, the dedicated page collapsed to 13%, while the original and the section-on-the-page held near 60%.

Finding 1: depth is the lever, not the URL

The dedicated page (B) and the on-page section (C) were indistinguishable on narrow questions: both 70%, a gap of essentially zero. Because they hold identical content, the only difference between them is the page boundary, and the boundary did not move retrieval. What moved it was developing the concept to real depth, up 55 percentage points over the original either way.

So the line between “needs its own page” and “needs its own section” is, for AI retrieval, not about the URL. A clearly-developed section on a consolidated page is found just as reliably as a standalone page.

Finding 2: the dedicated page has a cost

Splitting the concept onto its own page is not free. On broad questions about the parent topic, the dedicated page dropped to 13%, because it no longer contains the rest of the topic. The consolidated page with the developed section kept broad coverage at 57% and matched the dedicated page on narrow questions. It was the only version that did well on both.

The catch: found is not the same as answered

We checked the AI judges against a human reading the same passages. They agreed 90% of the time. The disagreements all pointed one way: the AI was satisfied when the answer was technically present across the retrieved passages, while a human reading the same fragments often was not. The recurring note was that the answer was buried, split across passages, or only inferable rather than stated. A chunk that contains the answer is not the same as a reader getting the answer. This reinforces the main result: structure and development are what make content retrievable, not the page it sits on.

What this means

If a buried concept needs to be found, develop it into a clear, self-contained section. You do not need to spin up a new URL to win AI retrieval, and a new URL can cost you on broader queries. Reserve the dedicated page for concepts that are genuinely their own topic, where the broad-query trade is one you want to make.

What this study does not claim

1. 01It does not measure classic search rankings. Ranking depends on links, crawl, and engagement signals we did not test. This is about AI retrieval only.
2. 02It is a point-in-time test on freshly built content, not a months-long study of how the web indexes a new page.
3. 03It tests the MOVE case, where the concept already exists on the source page. It does not test net-new content.
4. 04It does not cover every content type. Very long guides, product pages, and transactional pages are out of scope.
5. 05The AI judge panel ran marginally more lenient than a human even after calibration. The residual is disclosed in the methodology.