Localisation is a system, not a translation job

Machine-translating your store into six languages is easy. Keeping six localised stores correct, in sync, and individually findable — that is the actual problem.

International expansion fails quietly: the translated store launches, then drifts. New products ship in the primary language only, hreflang breaks in a template update, the German store still shows last season’s campaign. Builder and Visibility treat localisation as ongoing infrastructure — per-market content written rather than word-swapped, URL and hreflang structure kept mechanically correct, and every catalogue change propagated to every locale as part of shipping it.

Localised, not just translated

Word-for-word translation produces copy that is understandable and unpersuasive. Buying objections, sizing conventions, payment expectations, and search vocabulary differ per market — and the localised store has to reflect that to convert.

  • Product copy written per locale against local objections and terminology

  • Search vocabulary researched per market — literal translation misses what people actually type

  • Sizing, units, payment methods, and delivery expectations localised

  • Campaign timing per market, because commercial calendars differ

The technical layer must be boringly correct

hreflang mistakes, inconsistent URL structures, and half-localised templates cause wrong-language results and duplicate-content confusion. This is exactly the kind of mechanical correctness agents maintain well: every page in every locale carrying a complete, symmetric hreflang cluster, verified continuously rather than assumed.

Sync as a standing guarantee

The killer failure is drift. The agent treats the primary catalogue as the source of truth and propagates changes — new products, price updates, discontinued items, template changes — to every locale as they happen, flagging anything that cannot be auto-localised instead of silently leaving it in English.

How it works

01

Establish the structure

URL scheme, hreflang clusters, and per-locale templates set up correctly once — then enforced.

02

Localise per market

Copy written against local vocabulary and objections, not word-swapped from the primary language.

03

Keep locales in sync

Catalogue changes propagate to every locale automatically; gaps are flagged rather than ignored.

What you get

  • Per-market copy that answers local objections in local vocabulary

  • hreflang and URL structure kept mechanically correct at all times

  • New products live in every locale, not just the primary one

  • Wrong-language search results eliminated rather than shrugged at

Frequently asked questions

Subdirectories, subdomains, or country domains?

For most merchants, subdirectories consolidate authority best and are easiest to maintain. There are legitimate exceptions — legal entities, radically different catalogues — but the default should be the simple one.

Is machine translation good enough to start?

As a bootstrap for low-stakes pages, perhaps. For product and collection pages — where money changes hands — localisation against local vocabulary and objections outperforms it enough to matter, and that is what the agent produces.

How does hreflang actually break?

Asymmetric clusters, pages missing their return links, templates that drop the tags on new page types. It breaks silently, which is why the agent verifies it continuously instead of at launch.

Which markets should come first?

Where your data shows demand — existing traffic and orders by geography — rather than where intuition points. Analyst reads that before Builder builds anything.

Put the team to work on your store

Connect your storefront, analytics, and email stack, set a goal, and let the agents run the work end to end. Start on the free plan with your own model key.