Skip to content

Choosing the languages you write in

Enabling languages, setting a default and a fallback, keeping brand terms untranslated with a glossary, and why this steers generation rather than translating anything.

Updated

Localisation chooses the languages your organisation writes in, and steers every generation towards writing in them properly in the first place.

It is not a translation feature. Nothing here translates existing content, there is no translation provider and no translation memory, and no post has per-language variants. What the settings do is add instructions to the generation prompt: write in this language, use natural idiomatic phrasing, be conservative where the language is not fully quality-assured, and keep these terms exactly as written.

Quality is reported honestly per language rather than claimed to be equal everywhere. A language may be measured as high or good quality, in beta and not fully measured, or experimental with quality not measured at all.

BEFORE YOU START

  • The "administer" permission to change the language settings or the glossary. Reading needs nothing extra.
  • A decision about which languages you actually write in.
  • Your brand and product names, if you want them protected from translation.

STEPS

  1. Open Localisation, or go to /app/localisation.
  2. Tick each language you write in. Each shows its native name, a right-to-left marker where that applies, and its measured quality.
  3. Choose the default language for new content, and the fallback. Both are limited to the languages you enabled.
  4. Choose "Save languages".
  5. Add glossary terms: the term, the language it applies to, and either a preferred wording or a tick saying never translate it.
  6. Remove a term when it no longer applies.

WHAT YOU SHOULD SEE

"Saved.", and the glossary listing each term as either "Do not translate" or with its preferred wording.

WHAT THIS WILL NOT DO

  • It will not translate anything you have already written.
  • It will not switch English off. English is the anchor every other language falls back to, so its checkbox is permanently disabled and turning a language off never leaves anyone with nothing.
  • It will not accept a language it does not know. An unknown code is filtered out, and a default or fallback outside the enabled set snaps back to English.
  • It will not claim equal quality. A language marked experimental is marked experimental in the interface.

WORTH KNOWING

Glossary rules are per term and per language, and adding the same pair twice updates the existing rule rather than creating a duplicate. The rules reach both text generation and the image journey, so a protected brand name is protected in both.

IF IT DOES NOT WORK

  • "administer permission required." means your role cannot change these settings.
  • "Those settings could not be saved.", "That term could not be saved." and "That term could not be removed. It is still in your glossary." are the three fallbacks. The last one is precise: the term is still there.
  • "No approved terms yet." is the empty glossary.
  • If a generation comes back in the wrong language, check that the language is enabled: a request for a language you have not enabled falls back, and the generation records that a fallback was used.

COMMON QUESTIONS

How do I get a post in two languages?

Generate it twice, once in each. There is no per-language variant of a single post in this product.

Does a right-to-left language need anything else?

No. The generation instructions tell the model not to add direction markers or reorder punctuation, and the interface marks those languages in the list.

  • Writing posts with AI

    Generating a first draft from a brief, checking what it will cost first, and what the output is and is not.

  • Setting your brand voice

    Reusable steers for generation - tone, length, reading level, audience, words to avoid and words to include.

  • Managing brands

    How to create a brand, what your plan allows, and why the brand count on the page can differ from the number you can see.