Multilingual SEO: URL Structure, hreflang, and Why Translation Isn't Localization

The most common misconception about multilingual websites is "just translate the Chinese into English and put it up." We've seen far too many sites like this: content translated, URLs unplanned, language annotations missing, keywords translated literally — the result is an English version nobody can find in search, sometimes even dragging down the Chinese version's rankings. Multilingual SEO has three layers to get right: URL structure, language annotation, and — the most overlooked — local search habits.

Layer One: How to Structure Your URLs

Every language version must have its own URL. This is non-negotiable — if you swap content via cookies or browser language while the URL stays the same, search engines will only ever index one version. The three mainstream approaches:

For most Taiwanese SMBs, our recommendation is the subdirectory: one domain, one system, no diluted authority. For the general subdomain-versus-subdirectory trade-off, see our fuller discussion in subdomain or subdirectory.

Layer Two: hreflang Annotations

hreflang is the markup that tells Google "here are this page's language/region versions," so search engines serve the right version to the right user — a Japanese user sees the Japanese version, not the English one. The practical essentials:

• Annotations must be reciprocal: if the Chinese version points to the English one, the English version must point back. One-way annotations are ignored. Each page also needs a self-referencing annotation.
• Get the language and region codes right: Traditional Chinese is zh-Hant (or zh-TW for Taiwan specifically). Don't write just zh and assume it covers both Traditional and Simplified.
• Add x-default: it specifies which version to show when none of the user's languages match — usually the language-selector page or your primary market's version.
• hreflang is not a ranking boost: it solves "serve the right version," not "rank higher." Rankings still come from each language version's own content strength.

Layer Three: Localization, Not Translation

This is the true dividing line in multilingual SEO. The same need can be searched with completely different words in different markets: Taiwan searches "行動電源," mainland China searches "充电宝," English-speaking markets search "power bank" — if your English keywords are literal translations from Chinese, you're aiming at searches that don't exist.

The correct process: run independent keyword research for each target market, then use the local terms to rework titles, copy, and page structure. Beyond that, date formats, currencies, units of measure, and locally relevant case studies all shape local users' trust. Machine translation is fine as a draft, but content that hasn't been reviewed by someone who knows the local market should not go live — and it's not just a quality issue: large volumes of unreviewed automated translation are explicitly targeted by Google's spam policies.

Translation handles the language; localization handles the market. Nine out of ten failed multilingual sites only did the former.

Pre-Launch Checklist

1. Every language version has its own URL and is directly accessible (no reliance on auto-redirects).
2. hreflang annotations are reciprocal, include x-default, and have been validated with a tool.
3. Independent keyword research has been done for each market; titles and descriptions use local terms.
4. Structured data and the sitemap cover all language versions.
5. Don't force-redirect by IP — Googlebot mostly crawls from the US, and forced redirects mean it never sees your other versions. Use a suggestion banner and let users choose.

In the multilingual projects we've done, structure and markup are roughly 30% of the work; localization is 70%. If you're taking your business into overseas markets, talk to us — getting the architecture right up front saves you the cost of a full site migration later.