Translation workflows that don't break SEO

Translation looks like a content task and behaves like an infrastructure project: it touches every URL, every indexed page, and every market's search visibility at once. Done well, each language multiplies your discoverable surface. Done carelessly, it manufactures duplicate content, breaks URLs, and leaves markets invisible. Here's the machinery and the failure modes.

How Shopify's translation layer actually works

Translations in Shopify are overlays: the base resource (product, collection, page, theme string, metafield) exists once, and translated values attach per language, rendered when a market's locale is active on its subfolder (/de-de/ serving German overlays of the same underlying pages - the Markets architecture from Phase 1 completing itself). What this buys structurally: no duplicated resources (last lesson's anti-pattern designed away), automatic URL mapping between language versions, and translation coverage that's auditable - you can query what's translated and what's falling back to the base language. The native Translate & Adapt app covers manual and basic machine translation; translation apps and TMS integrations layer on workflow, memory, and vendor management for scale. The unglamorous but critical detail: translated handles. Shopify supports translating URL handles per language (/de-de/products/leinenhemd), and localized URLs are a real relevance signal - but changing handles later breaks URLs per language, so the Phase 1 handle-convention discipline now applies per market, decided before launch, not after indexing.

The workflow tiers, honestly priced

Raw machine translation (one-click app output, unreviewed): fast, cheap, and readable - and it shows, in exactly the places that cost you: idiom in marketing copy, size/fit nuance, category terms that miss how locals actually search ("Strickjacke" vs "Cardigan" is a keyword decision, not just a translation). Google also treats mass unreviewed MT skeptically - at worst as the thin duplicate content it often is. MT plus native review - the working standard for most catalogs: machine drafts, native speaker reviews with a terminology list (your Phase 2 closed attribute lists become the glossary - consistency doctrine, now multilingual), priority given to high-traffic pages. Native/transcreation for the surfaces where voice is the product: homepage, brand pages, campaign content, your best-performing guides (Phase 4's content, which deserves better than word-for-word - informational queries differ per language, sometimes per culture). The allocation rule: tier by page value, not by uniform policy - bestseller product pages and money collections get review; long-tail products can launch on MT and upgrade by traffic evidence.

The ways translation breaks SEO, specifically

The audit list, drawn from real wreckage: partial translation (half-translated markets read as low-quality; GSC per-market coverage from Phase 5's sitemap lesson shows it as indexing gaps - translate complete page-sets per market, prioritized, rather than scattered percentages of everything); untranslated metadata (title tags and meta descriptions forgotten while body content translates - the most visible fields, most often missed; audit them explicitly); keyword-blind category terms (collection titles translated literally instead of matched to local search language - per-market keyword checks on your taxonomy's translations are an afternoon that determines whether whole collections rank); broken hreflang from mixed setups (next lesson entirely); and translation overwriting live optimizations (a re-sync from a translation vendor flattening manual improvements - governance: one system of record per language, change flow in one direction). Every item is cheap before launch and expensive after indexing - which is the summary of this entire lesson.