How to Build a Multilingual Website: URL Structure, hreflang, and the Translation Workflow

A machine-parts manufacturer in Taichung launched an English site straight from machine translation. Three months later it was nearly invisible in Google's English results, overseas inquiries stayed at zero, and the rebuild cost another NT$80,000. Eighty percent of a multilingual website's success is decided before development starts — in URL structure, hreflang, and the translation workflow, not in the translation itself. This article breaks down all three.

When It Fits, When It Does Not

Good fit:

• Export-oriented manufacturers needing an English B2B product catalog to capture overseas inquiries
• Tourism and hospitality businesses serving Japanese-, English-, and Korean-speaking visitors
• Cross-border e-commerce selling into specific language markets
• SaaS products going global, with both UI and marketing pages in multiple languages

Hold off if:

• Your customers are 100% local and the foreign version is only there to "look international"
• You update content less than once a month but want five languages at once
• The budget only covers machine translation with no human review
• No one on your team can maintain foreign-language content or answer inquiries in those languages

Alternative Solution Matrix

Compare four approaches before committing:

The Full Process (Tools and Deliverables Included)

Decide the URL structure first

Per Google Search Central's official guidance: subdirectories (example.com/en/) are cheapest to maintain and keep domain authority consolidated; subdomains (en.example.com) offer deployment flexibility but dilute authority; ccTLDs (example.jp) send the strongest geo signal at the highest cost. Then map every language version to its counterparts with hreflang annotations so Google serves the right language to the right user.

The five phases

• Language and market audit (1 week): deliverable is a market-priority spreadsheet
• Information architecture and URL planning (1 week): Figma wireframes plus an hreflang mapping table
• Development and string externalization (2–4 weeks): Laravel lang files, translation strings managed in Crowdin
• Translation and localization (2–3 weeks): first pass via the DeepL API, then review by native-speaker editors
• QA and launch (1 week): hreflang validation, per-language coverage check in Search Console

Real Cost Breakdown

• A bilingual custom site typically adds 30–50% on top of the single-language quote
• Professional human translation: roughly NT$2–5 per character; about NT$20,000–50,000 per language for 10,000 characters
• Machine translation plus human review saves roughly 50–60% versus fully human translation

Hidden costs people forget:

• CJK webfont file size and font licensing fees
• RTL layout rework for languages such as Arabic
• Customer-service emails and system notification templates also need translating
• "Translation sync" labor on every site update: roughly 2–4 hours per language per release

Implementation Reality vs Client Expectations

• Expectation: translation is a one-off payment. Reality: content keeps changing, so translation is a subscription-style cost.
• Expectation: having an English version means Google will rank it. Reality: each foreign-language version needs its own keyword research and content optimization — ranking is a separate race.
• Expectation: launch machine translation now, fix it later. Reality: low-quality copy damages brand trust and SEO quality signals at the same time, and redoing it costs more.

Common Pitfalls and How to Avoid Them

• Broken reciprocal hreflang (A points to B, B never points back) → manage all pairs in one mapping table and validate with a checker before launch
• Forced IP-based redirects → Google recommends a suggestion banner that lets users switch themselves
• Strings hard-coded in templates → externalize every string into lang files from day one
• Dates, currencies, and units not localized → use locale formatting functions instead of hand-built strings
• Untranslated SEO meta (title, description, og tags) → include meta in the translation string list and the acceptance checklist

Success Metrics and the 90-Day Roadmap

• Day 30: confirm in Search Console that every language version is indexed with no hreflang errors
• Day 60: compare organic traffic and inquiries per language; invest more content in the best-performing market
• Day 90: review whether the translation sync workflow and costs match the plan, then decide on the next language

Decision Checklist: Should You Go Multilingual?

• ☐ I have a specific foreign target market, not just "it would be nice"
• ☐ Customers in that market actually search for my product in that language
• ☐ I can name who maintains the content for each language after launch
• ☐ Someone can answer inquiries or support tickets in that language
• ☐ I have chosen one structure: subdirectory, subdomain, or ccTLD
• ☐ The translation budget covers both first-pass translation and native review
• ☐ I accept that translation is an ongoing cost, not a one-off fee
• ☐ Every site update reserves translation sync hours
• ☐ SEO meta and system emails are inside the translation scope
• ☐ I will redo keyword research for each foreign market
• ☐ Someone owns the Search Console data for the first 90 days
• ☐ I will validate with one language first, not five at once

If you cannot check at least 8 items, build a strong single-language site first.

FAQ

How much does a multilingual website cost?

A custom bilingual site typically adds 30–50% on top of the single-language budget. At ScriptWalker, multilingual builds start from NT$120,000 (two languages included), with each additional language from NT$30,000 plus translation fees.

Should I choose subdirectories or subdomains?

For most Taiwanese SMEs, subdirectories (/en/) are the right call: cheaper to maintain and they consolidate domain authority. Consider subdomains or ccTLDs only when markets are run independently or regulations require it.

Can I just use machine translation for everything?

Not for direct publishing. Use DeepL for the first pass plus native-speaker review — about 50–60% cheaper than fully human translation, without the brand and search damage of raw machine output.

Will adding another language later be painful?

Not if strings are externalized and URLs plus hreflang are planned in phase one. Adding a language then mostly means translation fees and configuration hours, with no architectural changes.

Next Step

ScriptWalker's multilingual website service starts from NT$120,000 (two languages included), with each additional language from NT$30,000 plus translation fees — URL structure, hreflang, and translation workflow planned in one pass. Want to know which language your industry should launch first? Book a free 30-minute consultation:

• Email: [email protected]
• Phone: 0916-224-047
• LINE: @ufv9089p