About setting up a multilingual site
Translating a site looks like a copy-and-paste job with a machine translator. It isn’t.
We’ve spent months translating graphemics.net into English, and there are things we wish we’d known
before we started. Here they are in order, with the site’s real numbers and no sales pitch.
First: how much the second language actually brings in
We counted what we have today. Our WordPress API returns 545 articles, and
91 have an English link (/en/). That’s one in six.
In Search Console, over the last 28 days (through October 20, 2026), here’s what each side
delivered:
- English: 80 pages with impressions, 838 impressions and 7 clicks.
- The rest of the site (Spanish): 69 pages, 6,560 impressions and 28 clicks.
English accounts for 11% of impressions but 20% of clicks.
With a sample this small we won’t draw big conclusions, but we will draw one: a second language is
not decoration. People are searching in English for topics we already cover in Spanish, with less
competition and with clicks that weren’t in the account before.
There’s also a caveat in those numbers. The English page with the most impressions, an
explainer on YouTube Shorts views, got 198 impressions at an average position of
9.2. Other English pages show up at position 37 and as far down as 67. Translating
doesn’t rank on its own: it puts the content in the queue, it doesn’t move it up.
Second: translating isn’t the same as localizing
An article translated word for word sounds like a translated article. Readers notice by the
first sentence and leave. What we wish we’d known is that every text needs three decisions
that a machine translator doesn’t make:
- Which examples stay. An example about “a dental clinic in Cancún” is
useful to someone reading in Spanish in Quintana Roo. For someone reading in English from another
country, you have to decide whether to keep it as a real case or explain which city we’re talking about. - What people actually call what they’re searching for. “Agencia de posicionamiento web” isn’t
searched as “web positioning agency”; it’s searched as SEO agency. If the title is translated
literally, you lose the query. - Which units and formats change. Dates, currencies, thousands separators.
In Mexican Spanish we write 1.100; in English it’s 1,100. A misplaced separator costs
you trust.
Third: the technical structure gets decided before, not after
This is the most expensive thing to fix later. Before translating a single line, you need to decide how
each language will live on the site.
URLs. We use one folder per language: Spanish at the root and
English under /en/. It’s the easiest option to maintain with a single domain. The
alternatives, a subdomain or a separate domain per language, force you to build authority
separately.
Hreflang tags. These tell Google “this Spanish page
has its English equivalent here.” Without them, Google may show the wrong version or treat
the two as duplicate content. We checked them in the page code: the English homepage
declares both the Spanish and the English version in its HTML. If you don’t check by hand, you
don’t know.
The link between translations. In WordPress with a language plugin, each
translated article has to be explicitly paired with its original. Until it
is, the site treats it as just another post, with the same link and the same category as the
original. We learned that pairing is a separate step, and that if you skip it, the language
switcher sends the reader nowhere.
Slugs. The URL of each article is translated along with the title. It’s the
right call for the reader and for the search engine, but it forces you to keep a record of which Spanish
URL corresponds to which English one. Without that record, internal links break.
Fourth: not everything gets translated
This is where we’d change the most. We translated in order, from easiest to hardest, rather than in
order of demand. The right approach is the opposite: start with the pages that already get
impressions in Spanish, since those have proven demand, and leave for later whatever
has never shown up.
Of our 545 articles, a good share have never had a single impression. Translating them into English is
paying twice for content nobody is searching for. A simple rule we’d use from day
one: if the Spanish version has no impressions, the translation can wait.
Same goes for service pages. Those do go first, even if they have no impressions today,
because they’re the ones that close deals. A customer who arrives in English and finds the services page
in Spanish leaves.
Fifth: maintenance is a cost nobody budgets for
Translating is a one-time expense. Maintaining is a recurring one. Every time a Spanish article
is updated, the English one falls out of date, and nobody flags it. Six months later
you have two versions saying different things, with different figures, and the English reader gets the
older one.
What we’d do today: log the date of the last review for each translation and review
both versions together whenever the original is updated. A spreadsheet is enough.
The expensive part is not having one.
What we’d do differently, in a list
- Choose the language based on demand measured in Search Console, not on instinct.
- Define the URL structure and hreflang before the first translation.
- Translate first the pages that already have impressions and the service pages.
- Localize titles and examples; don’t just translate them.
- Pair each translation with its original the same day it’s published.
- Keep a log of reviews per language.
- Measure at 60 and 90 days before deciding whether to add a third language.
Is it worth it?
It depends on who’s searching for you. If your customers are in your city and speak your language, a second
language is an expense that can wait. If you sell to foreigners, tourists or buyers from
abroad, English stops being optional: it’s what they type when they search for you.
In our case, with a small sample, English brings in one in five clicks from
Google with just over a tenth of the impressions. It’s a signal, not a promise, and we’ll
keep measuring it.
If you’re thinking about opening your site to another language and want to know which pages to translate first,
get in touch. With your Search Console data we can tell you, in one afternoon, where
to start and what to leave out.
Figures from the graphemics.net WordPress API and Google Search Console, read on October 22,
2026 (28-day period through October 20). Impressions and clicks change daily. The comparisons on translation
costs and timelines are indicative and depend on each project.
Versión en español: Sobre montar un sitio en varios idiomas
