Translating a store is the easy part. Keeping it translated while the merchandising team ships three campaigns a month is the part that fails. A Shopify localization workflow is the process that decides what gets translated, by whom, when, and what happens to the untranslated string that appears on a Friday afternoon.
This guide covers the layers that need localising, where the process breaks, how to handle the gap between publishing and translating, and what to measure.
A Shopify localization workflow covers six layers, not one
Merchants think about product content. A store has more surfaces than that, and the forgotten ones are the ones customers notice.
- Theme strings — buttons, form labels, error messages, empty states.
- Product content — titles, descriptions, options, metafields.
- Collections and navigation — menu labels, collection descriptions, filters.
- Policies and static pages — shipping, returns, terms, FAQ.
- Transactional messages — order confirmations, shipping notifications, password resets.
- SEO metadata — titles, descriptions, alt text, structured data.
Layer five is the one most commonly missed, and it is the worst place to miss it: a customer who browses and buys in French, then receives an order confirmation in English, has been told the localisation is cosmetic.
Where a Shopify localization workflow actually breaks
Not in the initial translation project — that gets budget and attention. It breaks in month four, and always in the same three ways.
New content ships untranslated. A collection launches Monday and reaches the German store in English because nothing connects publishing to translating.
Edits do not propagate. Someone corrects an English product description; four translations still carry the old claim. Nobody knows, because nothing flags divergence.
Ownership evaporates. The person who ran the launch project moved on, and translation became everybody’s job, which means nobody’s.
Any Shopify localization workflow that does not have an answer to those three is a translation project with a process-shaped label on it.
What Shopify handles
Shopify’s localisation layer stores translations against the original content rather than duplicating products, which is what makes one catalogue across markets viable. Theme strings, product content, collections, policies and metafields are all translatable, and Markets handles the routing, currency and domain structure around them.
Crucially, hreflang is generated from that structure. Getting it wrong manually is a common and expensive error, so letting the platform do it is worth the constraint — see Google’s localized versions documentation for what it is signalling.
Shopify localization workflow: machine, human, or both
| Content | Approach | Why |
|---|---|---|
| Theme strings | Human, once | Small, high visibility, rarely changes |
| Product descriptions, long tail | Machine + spot check | Volume makes human review uneconomic |
| Hero products and campaigns | Human | This is the brand voice |
| Policies and legal | Human | Precision has consequences |
| Transactional emails | Human, once | Every customer reads them |
| SEO metadata | Human for top pages | Keywords do not translate literally |
The row that costs money when done wrong is the last one. Translating a meta title literally produces a phrase nobody in that market searches for. Keyword research has to happen per language, not once in English and then translated — our Markets SEO guide covers why this is the difference between a localised store and a findable one.
The publishing gap needs a policy
There will always be a window between publishing content and having it translated. Decide deliberately what happens in that window, because the default is usually the worst option.
Three workable answers: publish in the source language and translate within a defined SLA; hold the content from localised markets until translation lands; or machine-translate immediately and upgrade to human afterwards. Each is defensible. What is not defensible is having no rule, because then the answer varies by whoever happened to be publishing.
For most brands, machine-first with a human upgrade for top sellers is the honest compromise: customers see something, the team is not blocked, and quality follows revenue.
Who owns the Shopify localization workflow
Name a person. Not a team, a person. The single strongest predictor of whether a store stays localised after year one is whether one named individual has translation coverage in their objectives.
That person needs a report — untranslated strings per market, oldest first — and the authority to block a launch that has no localisation plan. Without the second, the first is a document nobody reads.
Currency, dates and units belong in the Shopify localization workflow
Language is the visible half of any Shopify localization workflow. The other half is formatting: decimal separators, date order, address formats, units of measurement and size systems.
A German customer reading 1,299.00 where they expect 1.299,00 has to stop and think about whether that is one thousand or one point three. Sizing is worse — a store selling apparel across the US, UK and EU without size conversion in the product content generates returns, which is a localisation failure appearing on a returns report where nobody attributes it correctly.
Testing what customers actually see
Test in the market context, not in the translation editor. The editor shows strings; customers see pages.
Walk the full journey per market at least quarterly: home, collection, product, cart, checkout, order confirmation email. That last step catches the transactional gap, and it is the step people skip because it requires placing a real order.
Also check for overflow. German strings run substantially longer than English, and a button that fits in English can wrap or clip in German — a layout bug that only exists in one market and that nobody sees unless they look.
What to measure in a Shopify localization workflow
- Translation coverage per market, as a percentage of published content.
- Age of the oldest untranslated string — a better health signal than the average.
- Conversion rate per market, compared against the source market as a baseline.
- Return rate per market, which surfaces sizing and description problems.
- Support tickets per market, normalised by orders — a spike usually means something is unclear rather than broken.
A Shopify localization workflow covers imagery too
Text is the obvious layer. Several others carry meaning and are routinely missed.
Images with text baked in. Campaign banners with English copy rendered into the file cannot be translated, only replaced. Either commission them per market or design them with text as an overlay, which is a decision for the design system rather than the translation process.
Size and fit guides. A conversion table is content, and an unconverted one generates returns.
Reviews. A German customer reading a wall of English reviews is being told which market matters. Machine translation with an “originally written in English” note is an accepted middle ground.
Support content. A localised store with an English-only help centre pushes customers into email, and support volume per market is the metric that reveals it.
Search terms do not translate
This is worth stating separately because it is where most localisation budget is wasted. The literal translation of your best-performing English keyword is frequently not what anyone types in that market.
Run keyword research per language before writing product titles and metadata, and expect the categories themselves to be named differently. A Shopify localization workflow that translates SEO metadata word for word produces pages that read correctly and rank for nothing, which is the most expensive kind of correct.
When to stop localising
Not every market deserves the full treatment. A market producing a handful of orders a month does not justify human translation of the long tail, and pretending otherwise spreads the budget too thin to serve the markets that do.
Tier your markets. Full human localisation for the top two or three, machine translation with human review on key pages for the middle, and machine-only for exploratory markets. Revisit the tiers twice a year against revenue, and be willing to move a market down as well as up.
Where to start
Audit coverage before buying tooling. Most stores discover that theme strings and transactional emails — the two cheapest layers to fix — are the ones with gaps, while product content everyone worries about is largely covered.
Then write the publishing-gap rule and name the owner. A Shopify localization workflow is mostly those two decisions plus a report; the tooling is the easy part, and buying it first is how brands end up with a translation platform and the same drift six months later.
If you are also deciding between one store with Markets and separate regional stores, that architecture choice constrains everything here — our comparison of Markets versus expansion stores covers it, and it is worth settling before you design the workflow.
