Location pages have a bad reputation, and mostly they have earned it. The usual construction is one template with a city name dropped into it, 38 times. Google calls that a doorway page and demotes the whole set. We run 38 of them anyway, because the demand is real: someone in Kiel looking for a software team searches for Kiel.
So the question is not whether to build location pages, but whether you can prove they differ. Proving means measuring.
How we measure similarity
We crawl all 38 pages, extract the visible text, and replace every occurrence of the city name with a placeholder. Masking is the step that matters: without it every page looks unique, because a different city name appears in every other sentence. Then we cut the text into overlapping five-word sequences and compare each pair of pages with the Jaccard index — the share of sequences the pair has in common.
With 38 pages that is 703 pairwise comparisons. The mean is the number that matters.
The result was uncomfortable
Our pages sat at 0.55 — 45 per cent unique text after masking. Each page had a decent, purpose-written part: a paragraph on the local economy, three sectors, who typically commissions work there, districts, two location-specific questions. The rest of the page was identical across all 38.
- Six process steps, word for word the same as on the services page
- Six service descriptions, written out in full on the service pages they link to
- Four reasons to work with us — statements about the studio, not about the city
- A technology list, identical to the section on /services
- A generic pricing question whose answer never mentioned the city
Together that was more than half of every page. The reflex is to write more unique text. It is the wrong reflex: 38 times 150 words about local software markets nobody researched is exactly the doorway page in a new coat.
What we did instead
We cut. Each shared block already existed somewhere it belonged — the process on the services page, the service copy on the six service pages, the reasons on the about page, the stack in the technology section. The location pages now carry a heading and a link where the repetition used to be.
After the cut: 0.39 pairwise similarity, so 61 per cent unique text. The pages shrank from 686 to 547 words — and the share that actually talks about the city rose from half to nearly two thirds.
A shorter page that mostly talks about its subject beats a long one that spends half its length talking about something else.
What we take from it
- Mask the variable part before measuring, or you are measuring the city name.
- Too little unique content is rarely the problem. Too much repetition almost always is.
- Every block of text belongs to exactly one page. Everything else is a link.
- Re-measure after every change. A number nobody recalculates is an opinion.
The 60 per cent mark is not an official Google threshold — no published number exists. It is a working line that tells you which side you are on. Knowing the number at all matters more than the number itself.