The English page was corrected. The other six still say the old figure.
Multilingual industrial content architecture that keeps specifications, technical documents and product data aligned across languages, so a correction made once reaches every version rather than depending on someone remembering which translations exist and who maintains them.
Drift is structural rather than careless. A product page is updated in the source language, the change is not flagged to translators, and each language version ages at its own rate. Within a year the same product carries different figures in different languages, and there is no reliable way to tell which are current.
A dimension, a tolerance and a part number are the same in every language. Treating them as translatable text means the same value is retyped for each market, which multiplies both the effort and the opportunity for error. Holding them as structured data that renders in every language leaves translation to the material that requires it.
Technical documents rarely exist in every language, and revision cycles differ. The model needs to express which document exists in which language at which revision, and what a visitor sees when their language version is unavailable, which is usually the source-language version clearly labelled rather than a broken link.
Machine translation is capable for high-volume, lower-risk material and less suitable where a mistranslated safety instruction, tolerance or approval statement carries consequences. The workable arrangement assigns different treatment by content type: reviewed translation for technical and legal material, machine translation with post-editing for volume, and no translation where a market does not need it.
The mechanism that holds it together is a link between source and translation with a status attached. When the source changes, every dependent version is marked as outdated and appears in a queue. Translators see what changed rather than the whole page, and editors can see at a glance which markets are behind.
A content model separating data from translatable text, a translation workflow with status tracking, document handling per language and revision, fallback behaviour, and hreflang declared across the full cluster. See multilingual industrial websites and WPML and translation.
Code Industrial is the industrial B2B practice of Code Barcelona, an agency building corporate websites and digital platforms since 2015. The same strategy, design and engineering team works on every industrial project, from the first scoping session through to life after launch.
What drifts depends on what the site carries. Multilingual industrial content architecture starts from that content.
What industrial companies ask when their language versions have grown apart.
It changes what has to be translated and what keeps versions aligned. Specifications, part numbers and units become structured data rendered per locale rather than translatable text, and a link between source and translation flags dependent versions whenever the source changes. The result is less translation volume and a visible record of which languages are current.
Fewer than most estates end up with. The useful test is whether a language corresponds to a market with a sales route and someone accountable for the content, since a language added for completeness becomes a version that drifts and misrepresents you. See how many languages should an industrial website have.
It depends on the material. For descriptive, application and news content it is capable and lets you cover more ground. For safety instructions, tolerance statements, approval wording and regulatory claims, the consequence of a subtle error is disproportionate, and those sections warrant reviewed translation with a named approver. See machine vs human translation.
The visitor should see the available version clearly labelled with its language, rather than a missing link or a silent substitution. On technical estates this is common and acceptable, since an engineer generally prefers the English datasheet to no datasheet. What matters is that the language shown is stated, so no one assumes they are reading a localised revision.
By holding them once. When a dimension or a rating is a structured field rather than text inside a translated page, it renders in every language from the same value and cannot diverge. That single change removes the most consequential category of drift, since a wrong figure in one language is more damaging than an untranslated paragraph.
Usually someone in the local market with the technical background to check that meaning survived, rather than the translator alone. For regulated or safety-related wording that approver may sit with a technical or compliance function. Making the approver a defined role in the workflow is what stops the question being renegotiated with each publication.
Yes, provided the model was built to separate data from translatable text and the targeting scheme was designed for expansion. Adding a language then means configuring the locale, extending the hreflang set and translating the fields that require it. Estates that duplicated whole pages per language are the ones where each addition costs as much as the last.
Language and market are separate dimensions and both need expressing. One language often serves several markets with different ranges and approvals, and one market may need several languages. See multi-country architecture for how the market layer is structured around the same content source.
Multilingual work usually sits alongside these.
Specifications that differ by language, or translations no one can confirm are current. Tell us which languages you run and we will tell you how we would approach the multilingual industrial content architecture.