Translation that keeps slugs, prices and tags in sync,
not just the sentences
Machine translation gets the sentence right and the structure wrong: a broken HTML tag, a shifted price format, a brand name translated by mistake. We build the pipeline to protect structure first, then translate the actual words.
What it is and who needs it
A translation and localization product translates catalogues, interfaces or content across languages while protecting everything that is not supposed to change, HTML structure, numbers, slugs, brand terms, rather than treating the whole document as free text. It fits any multilingual site, app or catalogue where translation errors in structure (a broken tag, a mistranslated brand name) cause more damage than a slightly awkward sentence would. It is not a fit for a one-off document translation a person can review line by line in an afternoon; the value is in volume and in keeping translations correct as content keeps changing.
What is inside
A structural protection layer runs before and after the translation model call, locking HTML tags, numbers, placeholders and slugs so the model only ever touches the actual translatable text. A glossary defines terms with fixed rules, your brand name, specific product names, technical terms with one correct translation, enforced as a check against the output rather than a hope the model remembers an instruction buried in a long prompt. Format rules apply per target language, quote marks, date formats, decimal separators, matching what readers in that language actually expect rather than a literal transfer of the source format. A review flow lets a human spot-check a sample before anything publishes, catching the rare case the automated checks miss.
How we build it
We start by defining the glossary and the structural rules together with you, since these decisions, what never gets translated, what always translates one specific way, are business decisions, not technical ones. The protection layer gets tested against deliberately tricky content, nested HTML, mixed numbers and text, before trusting it on your real catalogue. We launch with a defined content set and a few languages, reviewing output closely in the first batch, before scaling to the full catalogue and language set. Ongoing sync, where new or changed content gets translated automatically, gets built once the initial quality bar is proven on the first pass.
What to watch
The real risk is a structural break slipping through, a tag that gets mistranslated or dropped, a number format that silently shifts, which is why the protection layer gets tested against deliberately tricky content before trusting it on your real catalogue. Glossary maintenance is the other ongoing responsibility, a brand name or technical term added after launch needs to be added to the glossary explicitly, since the model has no way to know a new term should never be translated unless it is told. Review samples after any major content update, since even a well-tested pipeline can be caught out by a genuinely new content pattern.
Timeline and price
| Option | Price | What it covers | Timeline |
|---|---|---|---|
| MVP | from $1,500 | One content set, two to three languages, protected glossary | 3 to 4 weeks |
| Production | from $4,000 | More languages, format rules per locale, human review flow | 5 to 7 weeks |
| Full control (handover-ready) | from $6,800 | Everything in Production, plus a full handover package: architecture docs, test suite, admin access audit, and a walkthrough so your own team or another vendor can run it without us | 7 to 8 weeks |
Running cost on top of the build is usually $10 to $40 a month in model calls, depending on content volume and how often it changes.
What you own at the end
You own the translated content, the glossary, the protection rules and the full source code, running on your own infrastructure. The handover package documents exactly what the glossary enforces, so your own team can add new terms or languages without touching code.
Related
Pairs with the AI content studio when the same brand voice needs to travel across languages, and ETL and integrations hub for syncing a translated catalogue into your store. See the SEO and content service page for the editorial side of multilingual content. Real builds: the factory ERP recovery case study, where 3,040 interface strings were translated and kept structurally correct, and the fitness app case study, live in four languages. Translating catalogue updates by hand every time something changes? Get in touch.
FAQ
How much does a translation and localization product cost?
From $1,500 for translating a defined content set (a catalogue, an interface) into two to three languages with a protected glossary. Covering more languages or ongoing catalogue sync runs $4,000 to $6,800.
How long does it take?
Three to four weeks for an initial translation pass into a few languages. Ongoing sync, where new content gets translated automatically as it is added, takes longer to set up properly, typically six to eight weeks.
What is the stack?
Claude or GPT for translation itself, with a Python layer that protects structure, HTML tags, numbers, slugs, placeholders, before and after the model call, plus a glossary of terms handled as hard rules rather than left to the model's judgment.
Who owns the translations and the pipeline?
You. The translated content, the glossary and the code are yours, running on your own infrastructure, not a translation SaaS charging per word indefinitely.
How do you keep the brand name or technical terms from being mistranslated?
A glossary defines terms that must never be translated, or must always translate one specific way, enforced as a rule checked against the model's output, not just a hope that the model remembers the instruction.