Analysis

What 1,105 programmatic pages taught us about SEO in this sector

We built a large programmatic site, ran it for eight months and measured it honestly. Here is what it returned.

Published 2026-08-18

The experiment

A combinatorial programmatic build: platform × property type × benefit, generating 1,105 URLs. Every page technically valid, every page indexed, every page targeting a plausible keyword combination.

Over three months of measurement it produced 7,964 impressions, 20 clicks and an average position of roughly 43. Click-through rate: 0.14%. Impressions were flat month on month. Eight months in, it had plateaued.

Finding one: specificity beats volume, decisively

The platform pages — anything under /integrations/ — averaged position 24.8 and took 11 of the 20 clicks from 45% of the impressions. Every generic cluster (property types, benefit pages) sat between position 47 and 68.

Deeper was also better. Depth-four URLs averaged position 24.8 and took 10 of the 20 clicks. The more specific the page, the better it performed, consistently.

Finding two: 43% of the pages had never been seen

470 of the 1,105 URLs earned zero impressions in three months. Not low impressions. None.

That is not a content quality problem, it is an index bloat problem, and it dilutes everything around it. Fewer, stronger pages would have outperformed the whole build.

Finding three: you can rank for the wrong intent forever

Sixty per cent of impressions came from software queries — ‘airbnb pms’, ‘guesty channel manager’, ‘direct booking platform’. Nought clicks, average position 55.

The site's information architecture read as a SaaS product, so Google matched it against SaaS queries. People searching for software will never click a design studio, and if they do they bounce. Ranking is worthless if the intent is wrong.

Finding four: conversational queries were the bright spot

A handful of full-sentence queries ranked between position 6 and 10 — things like ‘does [platform] offer features for setting up a commission-free booking website?’. Everything else on the site sat at 43.

That is the shape of AI-assisted and voice-led search, and it is winnable now. Visible question-and-answer content — a clear question as a heading, a direct answer beneath it — is what gets extracted.

What we would do instead, and did

Six platforms, each with a properly written money page and four supporting guides. Roughly forty pages instead of eleven hundred, each written by a person, each answering something real.

Plus a visible answer layer on every platform page using the question formats that already rank. And a redirect map consolidating the equity from the programmatic build rather than throwing it away.

The uncomfortable conclusion: the programmatic experiment was worth running, and the answer it gave was ‘do the opposite’.