A precision-parts exporter bolted a translate widget onto its site and called it "the English version." Six months later Search Console told the real story: not a single English page indexed, Japanese and German buyers landing on Traditional Chinese, an 81% bounce rate, and three overseas inquiries a month. The owner blamed the ad budget. The actual cause was that widget-generated text never enters the HTML source, so search engines never see it and AI retrieval tools cannot cite it.
When It Fits vs When It Doesn't
Multilingual isn't "the sooner the better." Check which side you're on.
| Do it now | Not yet |
|---|---|
| You already get occasional overseas inquiries | Overseas demand is unproven, you're just curious |
| Clear target markets, four locales or fewer | Content changes weekly; translation can't keep up |
| Product copy can be standardized for translation | You close at trade shows; the site isn't the entry point |
| You need to be found in local search results | Nobody on the team can reply in that language |
| You have native-speaker review capacity | Budget under NT$80,000 but you want three locales |
Alternatives Matrix
| Option | Pros | Cons | Cost tier |
|---|---|---|---|
| JS translate widget | Free, live in ten minutes | Text never hits the source, no indexing, no hreflang | NT$0 |
| WordPress + WPML / Polylang | Mature ecosystem, editors pick it up fast | Plugin conflicts, slower pages, clunky admin as locales grow | NT$3,000–9,000/yr licence + build from NT$60,000 |
| Shopify Markets + translation app | Multi-currency and multi-language in one move | Template-bound; weak for blogs and technical docs | ~NT$1,500–5,000/mo |
| Custom i18n architecture (this offer) | Full control of URLs, hreflang and the translation pipeline | Higher upfront, 6–10 weeks | Starting at NT$180,000 |
Full Process Breakdown
Stage 1: Market and locale audit (1 week)
Pull country-level traffic from Google Search Console and Ahrefs, then build a locale matrix in Notion (language × region × currency × support coverage). Deliverables: locale decision sheet, translation scope list.
Stage 2: URL strategy and content architecture (1–2 weeks)
Lock the URL structure first, string extraction rules second — do it in the other order and you rebuild everything. Deliverables: URL spec, i18n key naming convention, Figma locale switcher design.
| Structure | Example | Best for |
|---|---|---|
| Subdirectory (recommended) | example.com/en/ | Shared domain authority, cheapest to maintain |
| Subdomain | en.example.com | Regional teams running their own site |
| ccTLD | example.jp | Going deep in one country, at higher cost |
Compare the trade-offs in Google's multi-regional site guidance.
Stage 3: Development and TMS integration (3–4 weeks)
Use Next.js with next-intl (or Laravel lang files) to extract every hardcoded string into JSON keys, add a translations table (key / locale / value / status / updated_at) so editors can fix copy from the admin, and wire up Crowdin for version control. Deliverables: staging site, brand glossary.
Stage 4: Translation rollout and SEO validation (2 weeks)
Draft with the DeepL API, review with native speakers, and add self-referencing hreflang plus x-default to every page. Deliverables: production site, hreflang validation report.
Real Cost Breakdown
- Development: from NT$180,000 (2 locales, ~40 pages); each extra locale adds roughly NT$35,000–60,000.
- Translation platform: Crowdin Pro from about US$59/month, translation memory included.
- Machine translation API: DeepL's developer plan covers the first 500,000 characters a month, then around US$5 per million.
- Native review: NT$1.5–3 per character; 40 pages is roughly 30,000 characters, so NT$45,000–90,000 — the line item people forget.
- Image and catalog rework: banners and PDFs with baked-in text need remaking, about NT$15,000–40,000.
- Hidden support cost: someone has to answer inquiries in that language, roughly 4–10 person-hours a month.
- Maintenance: NT$36,000–60,000 a year for sync translation and hreflang checks.
Reality vs What Clients Imagine
- Imagined: traffic arrives once it's translated. Reality: a new locale takes 3–6 months to rank locally; the first 90 days run on ads and your existing list.
- Imagined: AI translation is good enough. Reality: it's fine for drafts, but one wrong word in a spec, a legal clause or a CTA costs you the deal.
- Imagined: auto-detecting browser language is the friendly option. Reality: forced redirects show crawlers only one language and trap users who want to switch back.
- Imagined: switching locale just swaps words. Reality: dates, currency, address fields, name order and string length (German often runs twice as long) all change with it.
Common Traps and How to Avoid Them
- Asymmetric hreflang: every version must list all versions including itself, plus x-default — check against Google's hreflang rules.
- Publishing raw machine translation: build a glossary that locks brand terms and model numbers, and ship only after review.
- Forcing redirects by IP: use a top banner suggestion instead, and remember the user's choice in a cookie.
- Strings hardcoded in templates: require every visible string to go through an i18n key and add a CI lint to catch stragglers.
- Text baked into images: overlay banner copy in HTML so a new locale doesn't mean new artwork.
- Translating body copy but not meta: title, description, OG tags and the JSON-LD inLanguage all have to follow the locale.
Success Metrics and 90-Day Roadmap
- Day 30: all locales live, zero hreflang errors in Search Console, per-locale sitemaps submitted.
- Day 60: indexation above 90%, new-locale LCP under 2.5s via Cloudflare edge nodes, first inquiry-form conversions.
- Day 90: priority keywords on page 3 or better in target markets, overseas organic at 15% of traffic, content updates syncing across locales within 3 days.
Decision Checklist
- ☐ I already have overseas inquiries or orders
- ☐ I can name my top two target markets
- ☐ I need four locales or fewer
- ☐ Someone can reply and take calls in that language
- ☐ My product copy can be standardized
- ☐ I have native reviewers or will pay for them
- ☐ I accept NT$1.5–3 per character for review
- ☐ I have NT$180,000+ upfront
- ☐ I accept a 6–10 week build
- ☐ I understand a new locale takes 3–6 months to rank
- ☐ I'm willing to remake banners and catalogs
- ☐ I want cheap locale additions later
FAQ
Can I translate only the homepage and product pages?
Yes, and on a tight budget that's the pragmatic call. Just don't put hreflang on untranslated pages — noindex them or leave them in the source language so half-finished pages don't drag the site down.
Do Traditional and Simplified Chinese count as two locales?
They do. zh-Hant and zh-Hans differ in vocabulary, payment rails and target market, not just characters, so maintain them as two sets of content.
Can I launch with AI translation and clean it up later?
Technically yes, but the homepage, product pages and contact page must be human-reviewed first. Informational pages can follow in batches.
How long does adding a locale take after launch?
If the architecture is right, string extraction and the TMS pipeline already exist, so a new locale is mostly translation time — usually 2–3 weeks.
Let's Talk
If your site already has an "English version" that brings in nothing, don't raise the ad budget yet. Thirty minutes checking hreflang and indexation usually finds the real leak. ScriptWalker's Multilingual Website Build starts at NT$180,000, using Next.js and Crowdin to create an architecture you can extend one locale at a time, delivered with an hreflang validation report. Book a free audit.
- Email: [email protected]
- Phone: 0916-224-047
- LINE: @ufv9089p