A marketing director once told me her company had "done" international. They had translated four hundred pages into German, French and Japanese, put them on the same domain, and waited. Six months later the traffic curve in those markets was flat. Nothing was broken, exactly. The site simply did not exist as far as local searchers were concerned. That gap between a translated website and a visible one is what multilingual SEO is actually about, and it catches out far more companies than anybody admits publicly.
Translation Answers a Different Question Than Search
A translator is given a sentence and asked what it means. A searcher is never given a sentence. They type whatever phrase happens to be in their head, and that phrase is shaped by habit, by local competitors and sometimes by a word that has no clean equivalent in the source language.
The classic example is a term like "cell phone plan". Translate it literally into German and you get something grammatically correct that almost nobody searches for, because the actual query is built around Handyvertrag. Same product, same intent, different vocabulary. Multiply that across a product catalogue and you have a site that answers questions no one is asking.
This is why keyword research has to be redone in each market rather than translated. The output is not a list of equivalents. It is a list of the phrases real people use, ranked by how often they use them, which sometimes forces a change to page structure. A market where users search by problem rather than by product category needs different landing pages, not different words on the same ones.
The Technical Layer Everyone Gets Half Right
Search engines have to work out which version of a page to serve to whom, and they need help. That help comes from the hreflang annotation, a set of tags that map every language and region variant of a page to every other. Google's own guidance on telling search engines about localized versions is short and worth reading in full before anybody touches a template.
Most implementations fail for boring reasons. A page lists its alternates but forgets to reference itself. The German page points to the French one, but the French one never points back, so the set is incomplete and gets ignored. Region codes get invented, or a country code is used alone, which is not valid. The hreflang specification is strict about using recognised language and region codes, and a single typo quietly disables the whole cluster.
There is also the question of where the content lives. Subdirectories, subdomains and country domains all work, and the choice matters less than picking one and staying consistent. What does hurt is mixing approaches across a site, or leaving machine generated pages accessible in markets you never intended to enter.
Multilingual SEO Is Mostly a Content Problem
Once the plumbing is right, the remaining gap is editorial. Search engines reward pages that satisfy the query better than the alternatives, and a translated page is competing against locally written ones. If the local competitor has current pricing in local currency, examples from that market and a support number that does not sit in another time zone, a faithful translation will lose even when it reads beautifully.
This is where transcreation earns its keep. Rather than rendering the source, a writer rebuilds the argument for a different audience, keeping the intent and discarding the phrasing. It costs more per word and is worth it on the twenty pages that carry commercial weight. The other three hundred and eighty can be translated conventionally.
The same logic applies to conversion. A page can rank and still fail, which is why multilingual website SEO should be measured against local signups and enquiries rather than against impressions. Ranking is the cheap half. Persuading is the expensive half.
Ranking problems on a translated site usually trace back to text that was moved between languages without anyone owning its accuracy. The same weakness shows up far more expensively in official paperwork, where certified translation services exist precisely because an unverified rendering of a diploma or a contract has no standing at all. Search engines quietly punish careless translation, while registries reject it outright.
What a Sensible Rollout Looks Like
Start with two markets, not seven. Pick the ones where you already see unexplained traffic or where sales has live opportunities, because those give you a baseline to compare against. Run native keyword research for each, map it to your existing pages and accept that some pages will need to be merged or split.
Then get the hreflang right before publishing rather than after, since retrofitting means recrawling. Publish the commercial pages first and the supporting content second, because thin support content with no commercial page behind it produces traffic that never converts and muddies the data.
Give it two quarters before judging. International rankings move slowly, particularly on a domain with no local links. Building those links is a separate project, and often the slowest part, since a German publisher has no reason to cite an English page.
For teams formalising this properly, PoliLingua's overview of multilingual SEO best practices is a useful structural checklist to hold your own plan against, particularly on the governance questions that decide who owns what after launch.
The Part Nobody Budgets For
Maintenance. Every product update, price change and new landing page creates work in every language, and the version that gets forgotten is always the one with the smallest team behind it. Sites drift out of sync within a year unless someone owns the process.
The organisations that do well here treat languages as products with owners rather than as copies of the English site. That single change in framing tends to fix the budget question, the ownership question and the measurement question at once. Multilingual SEO is less a technical discipline than an operational one, and the hreflang tags are the easy part.
