Localisation

Localisation is the work that makes a product or page feel as if it was made for the place it’s used, whatever language or culture it started in. A faithful translation can still be a poor local experience: the words may be correct while the payment method is unavailable, the joke meaningless or the instruction legally wrong. Localisation asks whether the whole thing works here.

Three related terms are easy to mix up. Translation changes the language. Localisation adapts everything else as well. Internationalisation comes first: designing content, templates and systems so they can be localised without being rebuilt, for example by allowing for longer words, right-to-left scripts and different date formats.

What gets localised goes well beyond the text: date, time, number and address formats; currencies and prices; units of measurement; images, colours and icons that carry different meanings; examples, names and cultural references; legal requirements such as consumer rights, privacy notices and age limits; and voice and tone, since formality and directness land differently from one culture to another.

In Wales, public bodies work under the Welsh Language Standards, which require them not to treat Welsh less favourably than English. That makes bilingual publishing a test of maintenance as much as translation. The Welsh version needs to be as current, complete and easy to find as the English one, rather than a translation that lags weeks behind, so teams that publish both at once tend to build translation into the content workflow from the start.

Why it matters for content

Localisation affects inclusion, trust and task completion. People shouldn’t receive a second-rate version because they use another language or live in another region. Planning for localisation early lets content models, layouts and workflows accommodate variation rather than treating it as a late export job.

Maintenance is critical. Source content changes, but translated versions are often updated later or not at all. Teams need relationships between variants, visible status and clear responsibility for review. A changed delivery restriction, for example, needs to reach the source page and every affected language version.

The working tools make consistency possible at scale. A glossary or terminology database records approved translations of key terms. Translation memory stores previously translated sentences so they can be reused, which keeps wording consistent and cuts the cost of updates. A localisation style guide covers tone, formality and formatting for each language. Translators and local reviewers also need the purpose, audience and product behaviour, not isolated strings.

How AI changes the picture

AI can speed up translation, terminology checks and first-pass adaptation. It can make support for more languages affordable, particularly when paired with approved glossaries and translation memory.

Fluency doesn’t guarantee accuracy or cultural fit. Models may alter names, legal meaning or tone, and performance varies between languages, often falling for languages with less training data, Welsh among them. High-risk content needs qualified human review and testing with people who use the local version.

Structured source content helps machines and people track variants. Clear fields, locale metadata and shared terminology reduce drift, while change notifications can alert reviewers to affected variants. Someone still needs to act on them.