What I learned deleting 93% of my own website
Eleven sports, eleven subdomains, fifty-five languages, and an entity nobody could identify.
For about a year AnalyticsZone was eleven websites. One per sport: run.analyticszone.app, swim., bike., walk., health., sleep., and so on. Each one translated into roughly fifty-five languages. Fifteen thousand URLs in total.
It got almost no readers, and the reason turned out to be embarrassing once I finally looked at the right number.
The number I had not looked at
I had spent months on on-page work — schema, internal linking, page speed, the whole checklist — without once pulling the actual list of queries the domain appeared for. When I did, it was this: “gym analytics”, “data analysis app”, “fitness data analytics”. Not one query about sports science. Not one about critical power, or swim pace, or gait. And for the brand name itself, position 61.
Google had not decided my content was bad. It had decided the site was generic analytics software, and then answered the queries that classification implies. Everything I had been optimising was downstream of a category error.
Eleven hosts is eleven websites
Search engines treat a hostname as the unit of reputation. swim. and bike. were not two sections of one site; they were two separate small sites that happened to share a name, each starting from nothing, each with a handful of links. Whatever authority the project had was divided eleven ways and then divided again across the languages.
Consolidating them into one host with paths — /en/swim/, /en/bike/ — was not a cosmetic change. It was the difference between eleven weak entities and one that can accumulate.
Most of the pages should never have existed
Fifty-five languages sounds generous. In practice it meant that a page I had written once existed fifty-five times, fifty-one of them in machine translation nobody had read. When I audited the French, I found 325 occurrences of “running” translated as software execution — charge d’entraînement en cours d’exécution, training load “currently executing”. That had been live for months.
The health section had 35 leaf pages thin enough that merging each one into its category page lost nothing: 95 pages averaging 467 words became 54 averaging 766. Elsewhere there were near-duplicate articles covering the same topic in different words — three separate pieces on running training zones, written independently, sharing almost no sentences and every single heading. And there was an entire eleventh site that was a find-and-replace clone of another one, fully indexed, competing with its own source.
The final shape is one host, about 1,100 pages, and four languages indexed — English, Spanish, French and Catalan — plus Russian, Japanese, Korean and German where I have evidence somebody actually reads them. The other translations are still online for the people who land on them; they are simply no longer asking to be judged.
What I would tell myself two years ago
Publishing more pages is the easiest thing to do and the last thing that helps. A page only earns its place if a reader would be worse off without it, and translation multiplies whatever you had — including the mistakes.
Before optimising anything, check what a search engine thinks you are. If the answer is wrong, nothing downstream of it can work.
And the audit that finally moved things was not a tool. It was reading my own site in a language I speak, on a page I had never opened.