Skip to main content

How are glossary terms applied during AI or translation jobs?

Updated over a month ago

Glossary terms in Elovate are automatically integrated into your translation workflow to ensure consistent terminology, whether you’re using AI (ChatGPT) or a traditional translation provider like DeepL.


1. Automatic application during jobs

  • When you create a translation job in Elovate and select a glossary, the system sends your glossary rules along with the translation request.

  • Both DeepL and ChatGPT will follow these rules to keep brand names, technical terms, or other protected words consistent.

Example:
If your glossary contains:

  • Source: Elovate Boost

  • Target: Elovate Boost (do not translate)
    …then the provider will always keep Elovate Boost unchanged in the translated text.


2. How it works with different providers

  • DeepL: Glossary terms are applied directly through DeepL’s glossary API, ensuring they are strictly enforced.

  • ChatGPT: Glossary terms are added as part of the translation prompt, instructing the AI to respect those terms. While this offers flexibility, there’s a small chance the AI may deviate if the prompt is unclear — so precise wording in the glossary is important.


3. Supported job types

Glossary terms can be applied to translations for:

  • Products

  • Categories

  • CMS pages

  • CMS blocks

  • Attributes

  • Magefan blog posts


4. Limitations to be aware of

  • Glossary scope – Terms only apply if the glossary is selected during job creation.

  • Case sensitivity – Some providers treat terms differently if the case doesn’t match; keep entries consistent.

  • Multiple languages – You’ll need to add translations for each target language individually.


5. Best practices for maximum effect

  • Keep glossary entries short and precise.

  • Avoid adding full sentences — use key terms and phrases only.

  • Test your glossary in a small job before running large translations.

  • Update glossaries regularly as your product catalog or branding changes.

Did this answer your question?