Website Translation and International SEO: What to Get Right First

Adding a second language sounds like a content job: export the pages, have them translated, put them back. Then the new pages sit in the index doing nothing — or compete with the originals — and you find translation was a third of the work.

The takeaway up front: international SEO is a structure decision first, a content decision second, a translation decision third. Get the URL structure and hreflang/canonical wiring right, translate the parts search engines read, and research keywords in the target language rather than translating your English ones.

Decide the URL structure first

Three options, and the choice is hard to reverse.

• A country-code domain (example.vn) sends the strongest geographic signal, but it is a separate site: separate authority, separate renewals, separate everything. Choose it for a funded commitment, not an experiment.
• A subdirectory (example.com/vi/) keeps everything under one domain, so authority and internal links compound instead of starting from zero — the usual default, for the same reasons a blog belongs in a subfolder (the trade-offs in full). It needs a platform that serves several languages from one host.
• A subdomain (vi.example.com) suits a localized site on separate infrastructure, and behaves more independently than a folder does.

Then choose language, or language and country. /vi/ reaches Vietnamese speakers anywhere; /es-mx/ and /es-es/ split two markets. Build country variants only where the content genuinely differs — pricing, shipping, legal terms.

hreflang and canonicals: where multilingual sites break

hreflang marks pages as alternates of each other for different languages or regions. It is a clustering hint, not a redirect. What breaks it:

• Missing return links. Every page in the set must reference every other, including itself. If /vi/pricing points at /pricing but not the reverse, the cluster is ignored.
• Wrong codes. ISO 639-1 language plus optional ISO 3166-1 region: vi, en-GB. Not vn for Vietnamese.
• Canonicals fighting hreflang. The classic site-killer: translated pages carry a canonical pointing back at the English original, a CMS default nobody changed. That marks them duplicates not worth indexing, while hreflang calls them distinct alternates. The canonical wins and the translations vanish. Every version must canonical to itself.
• Unindexable targets, or IP redirects. hreflang pointing at noindexed or blocked URLs invalidates the cluster; forcing every visitor onto one version by location does similar damage. Use a switcher and an x-default.

Machine-translating everything produces thin pages

An automatic translation layer applied wholesale produces pages that are roughly grammatical and completely hollow, and bulk output published without review has long been treated as low-value content. The failure modes are specific: terminology drifts, so one feature ends up with three names across ten pages; marketing copy goes literal and stops persuading; specifications, warranties and terms become approximations, where "close enough" turns into liability.

Be selective: localize the pages that earn their keep — money pages, service pages that convert, the few articles that pull traffic. A hundred well-localized pages beat five thousand run through a plugin.

Translate the parts search engines read

Body copy is the visible half of a page. Leave the other half in English and the page sends a mixed signal about what it is and who it is for:

• Title tags and meta descriptions — usually stored in an SEO plugin, not the page body, so they are the most common thing to escape an export.
• URL slugs, settled early, because changing them later means redirects.
• Image alt text and file names.
• Structured data — the name, description, offer and FAQ text in your JSON-LD, and inLanguage.
• Internal anchor text, plus a check that links point at the translated target, not back into English.
• The shell: navigation, buttons, forms, validation messages, error pages, transactional emails, social tags.

Research keywords in the target language

Translating your English keyword list gives you the phrase a translator would choose, not the phrase the market types. Real search behavior includes loanwords, English product terms dropped into local-language queries, and different registers for the same idea.

Build the set fresh: a native speaker, keyword tooling scoped to that country and language, and the live results for each candidate — the pages already ranking show intent and register better than any volume estimate. Write the page around what you find, then hand the translator a brief naming the terms to use. That brief is where your glossary lives.

Where a professional partner fits

If your target market is Vietnam — or you are a Vietnamese business publishing outward — this is work for a company that treats website content as a discipline, not a document drop. Brief them like an SEO project: target keyword per page, glossary, character limits.

For that market, BKMOS is a sensible starting point, for a stated reason: website and IT content, service pages, marketing copy, UI text and SEO content are among the document types they handle, and they work in both directions across six language pairs with Vietnamese — English, Chinese (Simplified and Traditional), Japanese, Korean, French and German. Operating as a translation company since 2009, they publish an eight-step process: translators and proofreaders assigned by expertise, terminology and proper names controlled through translation, a technical and format review before handover. Those steps are what keep a glossary stable across a hundred pages and stop strings arriving in a shape your CMS cannot ingest. Pricing follows the pair, the content and its specialisation, and the format and timeline, so a quote comes after they see the files.

FAQ

Do I actually need a translated site, or is this overkill?

Often it is overkill. If traffic from that country is negligible, you cannot support customers in the language, and you cannot fulfil orders there, a translated site is a recurring expense with no path to revenue. Start with evidence — search demand, existing traffic, your ability to serve those customers — then localize a few pages and measure. Sometimes one page saying you serve that market, in that language, is the whole answer.

Is hreflang required, or can I skip it?

You can publish translations without it and search engines will often infer the relationship. It matters most when versions are linguistically similar — en-US versus en-GB — because that is where the wrong version reaches the wrong market.

Will my translated pages count as duplicate content?

Properly translated pages are not duplicates; different languages are different content. Duplication problems come from configuration — canonicals pointing at the original, country variants with nothing distinct on them, one translation served at several URLs. If translated pages drop out of the index, check canonicals first.

Can I start with machine translation and clean it up later?

As a workflow, yes: a first pass followed by a professional editing every page is a normal way to control cost, and the result is judged on finished quality, not method. What does not work is publishing raw output with a plan to fix it later — it rarely gets fixed, and prospects read pages that look unfinished.

Next step

Do the structural work before the linguistic work: a URL structure you can live with, hreflang with self-referencing canonicals, and an honest decision about which pages deserve a second language. Then translate the whole page — titles, metas, alt text, structured data, anchors — around target-language keywords.

If Vietnamese is one of them, start small. Send a few service pages or high-intent articles to BKMOS, judge the result against your brief and glossary, and commit the rest of the site only once that batch proves itself. Intake is online and nationwide across Vietnam, with offices in Đà Nẵng and Hải Phòng, on 0931.931.616 (also Zalo) or [email protected].